For the last two posts we’ve been talking about Polymarket, the world’s largest prediction market, and the AI strategies that quietly extract millions of dollars from it every year.

If you missed them: Post 1 explained the 10/90 pricing quirk that makes it beatable. Post 2 walked through ten AI architectures the pros are using.

But there’s been an elephant in the room I want to address directly: Polymarket is blocked for US users. You can’t legally trade there. The “work-arounds” (VPNs, offshore accounts) are real but risky, and not something I’d recommend for any business owner with real assets to protect.

So why have we spent two posts on it?

Because the smartest play isn’t trading Polymarket at all. The smartest play is using Polymarket as your crystal ball, and executing on the regulated, deep-liquidity, fully-legal markets you already have access to.

Welcome to Day 3.

The Core Idea

Prediction markets are now sharp enough that their prices move BEFORE related stocks do. The crowd in a prediction market, putting real money on real questions about real events, often connects the dots faster than Wall Street analysts.

When that happens, there’s a window. Sometimes seconds. Sometimes minutes. Sometimes hours. In that window, you can buy or sell the related stock before the rest of the market catches up.

You don’t have to bet on Polymarket. You just have to read it.

You’re reading the same data the AI traders are reading. You’re just executing in a deeper, safer, fully-legal market.

The Four Plays

Play #1: Regulatory Decision Front-Running

The setup: Polymarket has markets on FDA approvals, SEC rulings, FTC decisions, agency actions, and court cases. These markets often move on filings and signals that take 12-48 hours to make mainstream financial press.

The trade: Polymarket odds on “Will FDA approve [Drug X] by Q3?” jump from 40% to 70% in four hours. The pharma stock holding that drug hasn’t moved yet. You buy the stock (or call options) before Wall Street notices.

Real example pattern: A small-cap biotech’s Polymarket approval odds shift from 35% to 65%. The stock is still trading sideways. Within 48 hours, an analyst note drops and the stock jumps 18-25%. You captured the move.

Data sources needed:

• Polymarket API (free) — for the live odds

• A database mapping each regulatory market to the affected ticker(s)

• News API to confirm the underlying signal

• Your broker’s API for execution

AI system:

• Continuous polling of regulatory-themed Polymarket markets

• Threshold detection (e.g., “alert when probability shifts more than 15 points in less than 6 hours”)

• Stock price comparison (has the underlying stock moved correspondingly yet?)

• Alert layer that flags trade opportunities to you for review

Build cost: $8,000-15,000 + ~$300/month

Best for: Biotech, fintech, energy, and other heavily regulated sectors.

Play #2: Geopolitical Defense and Energy Trades

The setup: Polymarket has rich markets on wars, conflicts, sanctions, and geopolitical outcomes. Defense contractors (Lockheed Martin, RTX, General Dynamics, Northrop Grumman), oil majors, and shipping stocks all move on these outcomes - but often slowly, as institutional analysts wait for mainstream confirmation.

The trade: Polymarket odds on “Will [conflict] escalate by [date]?” jump 20 points in a day. Defense ETF (ITA) and individual contractor stocks haven’t moved yet. You buy ITA or specific contractors before the analyst notes drop.

Inverse trade: Polymarket “ceasefire by X date” odds jump. Oil futures haven’t fallen yet. You short oil or buy airlines (which benefit from cheaper fuel).

Data sources needed:

• Polymarket API (geopolitical markets)

• OSINT feeds (we covered these in Post 2 - ISW, GDELT, Twitter analyst lists)

• Defense/energy ETF and stock pricing

• News confirmation layer

AI system:

• Combined Polymarket + OSINT monitoring

• Sector mapping (which conflict outcome affects which stocks)

• Confidence scoring based on multi-source signal alignment

• Trade alert system with suggested instruments and sizing

Build cost: $15,000-30,000 + ~$700/month

Best for: Anyone with a brokerage that allows sector ETF and option trading.

Play #3: Crypto Spot Trades on Regulatory and ETF Markets

The setup: Polymarket has markets on crypto ETF approvals, SEC enforcement actions, and regulatory decisions. These markets are unusually sharp because crypto-native traders have deep knowledge of upcoming events.

The trade: Polymarket “Will SEC approve [token] ETF by Y?” odds spike from 25% to 60%. The underlying token hasn’t rallied yet. You buy the spot crypto on Coinbase or Kraken before the ETF approval rally.

Critical: This requires you to be comfortable holding crypto. If that’s not your world, skip this play entirely.

Data sources needed:

• Polymarket API (crypto-themed markets)

• Coinbase/Binance/Kraken APIs for crypto execution

• SEC filing tracker

• Whale wallet monitoring (optional but valuable)

AI system:

• Polymarket crypto-market monitoring

• Cross-reference with on-chain whale activity (the strategy we covered in Post 2)

• Spread detection between predicted probability and current spot price

• Execution layer for crypto exchanges

Build cost: $12,000-25,000 + ~$500/month

Best for: Investors already active in crypto who want a systematic edge.

Play #4: Earnings and Macro Decision Plays

The setup: Polymarket has markets on earnings beats, Fed decisions, jobs reports, and economic data releases. The crowd consensus is often sharper than Wall Street analyst estimates because money’s on the line, not reputation.

The trade: Polymarket says 30% chance the Fed cuts rates next meeting. Bond futures are priced as if it’s 50% (rate cuts are priced in to bond yields). You short the rate-cut trade in bond futures, or buy financial sector stocks that benefit from rates staying higher.

Another version: Polymarket “Will [Company] beat Q3 earnings?” sits at 65%. Analyst consensus says 80%. You either short the stock before earnings (if you trust the prediction market) or stay away from it entirely.

Data sources needed:

• Polymarket API (earnings and macro markets)

• Bond futures pricing (CME, Tradier)

• Stock options pricing

• Analyst consensus data (Refinitiv, Bloomberg, or free sources like Finviz)

AI system:

• Multi-source consensus comparison (Polymarket vs. analyst vs. options-implied)

• Divergence detection (when Polymarket disagrees materially with consensus)

• Risk-adjusted trade sizing

• Time-boxed exit logic around the event date

Build cost: $20,000-40,000 + ~$800/month

Best for: Sophisticated traders comfortable with options and futures.

The Brokers You Need

For execution, you’ll need a broker with API access. The three best options:

Alpaca (alpaca.markets) — free stock trading API, designed specifically for algorithms. Best for stocks and ETFs. Easy onboarding. Built for exactly this kind of use case.

Interactive Brokers (interactivebrokers.com) — pro-grade platform with the widest range of instruments (stocks, options, futures, bonds, international). More complex setup, but if you want full control, this is the gold standard.

Tradier (tradier.com) — solid middle ground. Good for options trading via API at low cost.

For crypto execution: Coinbase Advanced Trade, Kraken Pro, or Binance.US all have APIs you can integrate with.

The Full System Architecture

If you wanted to build the complete signal-to-trade pipeline, here’s what it looks like:

1. Data layer — Polymarket API + news APIs + OSINT feeds + your broker’s market data

2. Signal layer — AI (Claude or GPT) interprets the Polymarket movements and decides if they’re material

3. Mapping layer — database connecting prediction markets to specific stocks/instruments

4. Decision layer — AI scores confidence, suggests trade direction and sizing

5. Human review layer — YOU approve trades before execution (essential, especially while building trust in the system)

6. Execution layer — broker API places the trade automatically once approved

7. Tracking layer — logs every signal, every trade, every outcome for ongoing model improvement

Total build cost: $30,000-80,000 depending on which plays you want to run

Total ongoing cost: $1,500-3,000/month

Realistic time to build: 8-16 weeks

The Critical Warning

Before you build this or hire someone to build it, three things you need to internalize:

1. Prediction markets are not always right. They’re sharper than the average analyst, but they’re not psychic. Maybe 60-65% of strong signals lead to actual stock moves. The other 35-40% are false positives. Your system has to handle losses without panic.

2. The window is closing. Institutional investors are getting access to prediction market data through ICE (the parent company of the New York Stock Exchange). The edge that exists today will compress in 18-24 months. Build this now or don’t bother.

3. This is NOT a hands-off money printer. The pros running these strategies have human oversight on every trade. AI flags the opportunity. Humans approve the execution. Anyone selling you “fully automated prediction-market-to-stock trading” is selling you a way to lose money quickly.

The Real Takeaway for Business Owners

Even if you never want to trade a single stock based on prediction market data, the architecture above is a master class in modern AI system design:

• Pull from multiple data sources (most businesses pull from one)

• Use AI to interpret, not just monitor (most businesses just monitor)

• Build a human-in-the-loop decision layer (most businesses automate either too little or too much)

• Map signals to specific actions (most businesses see signals but don’t act on them)

• Track outcomes to improve the model (most businesses fly blind)

Replace “Polymarket” with “your industry’s leading indicators.” Replace “stocks” with “your business decisions.” The template is identical.

The companies that win the next decade won’t be the ones with the biggest budgets. They’ll be the ones reading the world a few hours faster than their competitors and executing while everyone else is still gathering information.

The Polymarket-to-stock signal play is just one example. The pattern is universal.

Stay smart,

The SmartOwner Team

LEGAL DISCLAIMER

This article is published by SmartOwner for general informational and educational purposes only. Nothing contained herein constitutes investment, financial, legal, tax, or trading advice, nor a solicitation, recommendation, or endorsement of any specific security, trading strategy, prediction market, platform, or product. SmartOwner is not a registered investment advisor, broker-dealer, or fiduciary. All trading and investment activities involve substantial risk of loss, including total loss of capital. Past performance is not indicative of future results. You are solely responsible for verifying the legality of any activity in your jurisdiction. Before making any financial decision, consult a qualified, licensed professional. SmartOwner and its affiliates accept no liability for any loss arising from reliance on the information in this article.

In yesterday’s post we broke down the central pricing quirk inside Polymarket: contracts priced at 10 cents come true about 14% of the time, and contracts priced at 90 cents come true only about 85% of the time. Human emotion creates a measurable gap between what the crowd thinks and what actually happens.

If you missed it, read it here: HOW TO BEAT POLYMARKET — it explains the foundation of everything below.

Today we’re going deeper into the how. Each strategy below includes the data sources, the AI system architecture, and a rough sense of what it costs to build. Even if you have no interest in trading, the architecture maps directly onto problems in your business.

The Framework First

Every strategy follows the same four-step pattern:

1. Pull data from public sources nobody is reading systematically

2. Use AI to process it faster than humans

3. Compare AI’s interpretation to a target market price

4. Act before the slow-moving crowd catches up

Keep this in mind as you read. The strategies aren’t really about Polymarket. They’re about applying this template to any market where humans set prices emotionally.

The 10 Strategies

1. The Calibration Fade

What it is: AI scans markets priced above 82% and takes the opposite side. Retail traders systematically overpay for certainty, so betting against “sure things” wins on average.

Data sources needed:

• Polymarket API (free) — gives you live prices on all 2,000+ active markets

• Historical resolution data from Polymarket (free, scrapeable)

• That’s it. This strategy doesn’t need outside data — just the platform’s own.

The AI system:

• A nightly script that pulls all active markets via the Polymarket API

• A filter that flags any market priced between 82% and 95%

• A scoring engine (Claude, GPT-5, or similar) that estimates the AI’s own probability of each event happening

• An automated dashboard that ranks markets by how much the price exceeds the AI estimate

• Optional: auto-execution layer using Polymarket’s order API

Stack: Python script + Polymarket API + LLM API (Claude/GPT) + Supabase for storage + n8n or Make.com for orchestration

Build cost: $2,000-5,000 in dev time + ~$200/month in API and infrastructure costs

Business lesson: Anywhere consumers are extremely confident about an outcome, there’s usually a hidden gap. The “obvious” winner in your category is often overpriced. Build a simple AI scoring system that estimates true probabilities vs. market prices for anything you buy or sell.

2. Longshot Reversal

What it is: AI scans 5 to 10 cent contracts and identifies the rare ones that are actually underpriced because the crowd is missing specific niche information.

Data sources needed:

• Polymarket API (live longshot markets)

• Wikipedia base rates (free) — for historical analog comparison

• Domain-specific data depending on the market (sports stats APIs, weather feeds, etc.)

• A historical database of past Polymarket longshot outcomes

The AI system:

• A query layer that pulls every market priced under 15 cents

• For each market, an AI lookup that finds 5-10 historical analogs (e.g., “rookie wins MVP in their debut year”)

• A base-rate calculator that determines true historical probability

• A flagging system for markets where true probability exceeds market price by 5%+

• Human review layer (AI flags candidates, you make final call)

Stack: Same as #1, plus a vector database (Pinecone, Weaviate) for analog matching + custom scrapers for domain-specific stats

Build cost: $5,000-15,000 + ~$400/month

Business lesson: Most “hidden gem” opportunities really are duds. The few that aren’t share a pattern — specific data the crowd doesn’t have access to. AI that does proper base-rate analysis on opportunities is more valuable than gut instinct.

3. The 72-Hour News Overreaction

What it is: When bad news drops, political market prices crash too hard and revert within three days. AI catches the overcorrection and bets on the bounce-back.

Data sources needed:

• Polymarket order book data (real-time prices and volume)

• NewsAPI ($449/mo) or SerpAPI Google News ($75/mo)

• Twitter/X scraping for sentiment (Apify $49/mo)

• Optional: GDELT Project (free) for global news event tracking

The AI system:

• Continuous monitoring of news feeds and social sentiment around politicians/events with active markets

• Spike detection: when a sentiment score drops sharply AND a market price moves >15% in <2 hours, the system flags it

• AI estimates the “fair value” reversion target based on historical patterns

• Time-boxed trade execution (48-72 hour exit window built in)

• Auto-exit logic if reversion doesn’t materialize within the window

Stack: Python + news APIs + LLM for sentiment analysis + n8n for orchestration + Polymarket order API for execution

Build cost: $8,000-15,000 + ~$700/month

Business lesson: When news breaks in your industry, the first two hours are usually the worst time to react. The market overcorrects. Build a 72-hour rule into your decision-making — especially for hiring, firing, pricing changes, and PR responses.

4. The Politics Underconfidence Trade

What it is: Political markets are systematically too cautious about favorites. A candidate at 65% is usually really at 75-80%. AI bets into the favorite.

Data sources needed:

• Polymarket political markets API

• RealClearPolitics polling averages (scrape weekly)

• 538 archive data (free, downloadable)

• Predictit history (free, for cross-platform calibration)

• State-by-state voter data where applicable

The AI system:

• Polling aggregator that builds a true probability estimate for each candidate

• Comparison engine that flags markets where the Polymarket price is at least 10% below the aggregated polling estimate

• Confidence scoring based on poll quality, sample size, and recency

• Position sizing based on the gap and confidence

Stack: Python scrapers + LLM for poll analysis + statistical model layer + Polymarket execution layer

Build cost: $10,000-20,000 + ~$300/month (polling data is mostly free)

Business lesson: Don’t assume your industry’s pricing bias is the same as everyone else’s. Some markets overprice certainty (most). Some underprice it (some, like politics). The first job is measuring which one applies to your category.

5. Cross-Platform Arbitrage

What it is: Polymarket and Kalshi often price the same event differently. AI watches both simultaneously and trades the spread.

Data sources needed:

• Polymarket API (free)

• Kalshi API (free with account)

• A market-matching layer (which Polymarket question = which Kalshi question)

The AI system:

• Parallel API monitoring of both platforms

• Semantic matching engine (uses LLM to identify when two differently-worded questions refer to the same event)

• Spread calculator that flags any gap > 5%

• Risk-management layer (since you’re holding positions on two platforms simultaneously)

• Auto-execution layer that places opposing bets to lock in the spread

Stack: Python + both platform APIs + LLM for semantic matching + dual-platform execution layer

Build cost: $15,000-25,000 + ~$500/month (legal review costs extra)

Business lesson: Never look at just one source of pricing. Build AI that monitors multiple marketplaces, competitors, regions, or channels for the same product or service. Most business owners are checking one source and calling it research.

6. Real-World Data Markets

What it is: Polymarket has markets on shipping, weather, satellites, physical events. These markets are sparse but have direct real-world data feeds.

Data sources needed:

• MarineTraffic API ($200/mo) for ship positions

• FlightAware API ($300/mo) for cargo flights

• USDA crop reports (free, RSS feeds)

• NOAA weather APIs (free)

• USGS earthquake feeds (free)

• Satellite imagery providers (Planet Labs, Sentinel Hub) for advanced setups

The AI system:

• Real-time data ingestion from multiple physical-world feeds

• Mapping layer connecting each market to its relevant data feed

• AI model that converts raw data (e.g., “200 ships in queue at Suez”) into probability estimates

• Comparison to Polymarket price

• Alert and execution layer

Stack: Python + multiple physical data APIs + LLM for synthesis + custom modeling layer + Supabase

Build cost: $15,000-40,000 + ~$800-1,500/month (API costs scale with usage)

Business lesson: Physical-world data is wildly undervalued. If your business touches logistics, real estate, agriculture, construction, or manufacturing, real-time sensor and satellite feeds are your biggest untapped advantage. Most competitors aren’t using it because they don’t know how to start.

7. The Whale-Lag Copy Strategy

What it is: Polymarket runs on a public blockchain. Every trade is visible. AI tracks the wallets of proven winners and copies their bets within seconds.

Data sources needed:

• Polygon blockchain data (free via RPC nodes — fast ones cost $200-500/mo)

• Polymarket wallet leaderboard data (publicly scrapeable)

• Historical wallet performance data (build your own from blockchain)

The AI system:

• Continuous blockchain monitoring (every block, every transaction tagged as Polymarket-related)

• Wallet ranking system based on historical PnL (profit and loss)

• “Smart money” whitelist (top 50-100 wallets by track record)

• Real-time alert when whitelisted wallet places a bet > $10K

• Auto-execution layer that copies the bet at reduced size (e.g., 5-10% of whale’s position)

• Filter for unusual patterns (avoiding manipulation)

Stack: Python + Polygon RPC node + blockchain indexer (Alchemy, QuickNode) + custom analytics + execution layer

Build cost: $20,000-40,000 + ~$700-1,200/month (RPC node costs)

Business lesson: In your industry, who are the “whales” whose moves predict where things are going? Are you watching their public signals — job postings, patent filings, real estate purchases, SEC filings, executive moves? The signals are free. Most businesses are too busy to read them. AI can read them for you.

8. Court Docket and Regulatory Front-Running

What it is: Polymarket has dozens of legal/regulatory markets. Court filings are public but almost nobody reads them daily. AI does, and beats the news cycle by hours or days.

Data sources needed:

• CourtListener API (free, you just need an account and token)

• PACER (federal courts, paid per document but cheap)

• SEC EDGAR API (free)

• FCC ECFS (free)

• Federal Register API (free)

• State court systems (varies by state, some have APIs)

• SCOTUSblog and similar specialized sites (scrape)

The AI system:

• Continuous polling of court and regulatory APIs (every 15-60 minutes)

• LLM-based summarization of new filings

• Matching engine that connects filings to active Polymarket markets

• Materiality scoring: does this filing actually change the probability?

• Alert system with confidence ranking

• Optional auto-execution for high-confidence matches

Stack: Python + multiple government APIs + LLM (Claude is best for legal docs) + Supabase + n8n + alert layer

Build cost: $10,000-25,000 + ~$400/month (most data is free; LLM costs are the main expense)

Business lesson: Every regulated industry has a slow-moving paper trail almost nobody reads. Whoever reads it first has an information edge measured in days. AI makes reading it effectively free. This is one of the highest-ROI builds for legal, healthcare, finance, construction, pharma, energy, and real estate businesses.

9. Geopolitical Open-Source Intelligence

What it is: Markets on wars, conflicts, and ceasefires move on ground-truth events 24-48 hours before mainstream coverage. AI watches the channels everyone else ignores.

Data sources needed:

• Twitter/X scraping via Apify ($49-200/mo)

• ISW (Institute for the Study of War) daily reports (free, RSS)

• GDELT Project (free) for global event tracking

• Telegram channel monitoring (specialized tools or Apify)

• Foreign-language press (translation via LLM)

• Curated OSINT analyst lists (compile manually)

The AI system:

• Multi-source ingestion of OSINT feeds

• LLM-based translation and summarization (multiple languages)

• Daily “ground truth” briefing generation

• Comparison layer that scores each active geopolitical market against the briefing

• Confidence-weighted position recommendations

• Manual review step (geopolitics has high stakes, automation alone is risky)

Stack: Python + Apify/scraping infrastructure + Claude (excellent at translation and synthesis) + custom analytical layer + alert system

Build cost: $15,000-30,000 + ~$600/month

Business lesson: The news everyone is reading is, by definition, already priced in. The edge is in the channels everyone is ignoring — industry Discord servers, niche forums, regional newsletters, foreign-language press. AI can monitor 10,000 sources for the cost of one human.

10. Resolution Source Hunting (The Cleanest Edge)

What it is: Every Polymarket question has a specific data source that determines the outcome. Sometimes the source reveals the answer before the market notices. AI polls every source every few seconds.

Data sources needed:

• Polymarket’s resolution criteria for each market (parse from market descriptions)

• Direct access to every government, financial, and data source Polymarket uses for resolution

• Examples: USDA reports, Bureau of Labor Statistics, Federal Reserve releases, election commission feeds, weather feeds, sports APIs, Wikipedia, specific government websites

The AI system:

• Database of every active Polymarket market and its resolution source

• A polling layer that hits each resolution source every 15-30 seconds

• Change detection (any update to the source)

• LLM interpretation: does this change resolve a market, and which way?

• Automated execution within seconds of detection

• Critical: extremely fast infrastructure (this is a speed game)

Stack: Python + dedicated VPS for low-latency polling + multiple API/scraper integrations + Claude for interpretation + Polymarket order API with optimized execution + monitoring dashboard

Build cost: $25,000-50,000 + ~$1,000-2,000/month (infrastructure for speed costs more)

Business lesson: The data that determines major outcomes in your industry is usually public, on a schedule nobody is watching. Automating monitoring of those exact feeds is one of the highest-ROI things AI can do for any business. The question isn’t whether the data exists. It’s whether you’ve automated reading it.

The Reality Check on Returns

Before you assume any of these are easy money, here’s the truth:

• Pros running these strategies make 12-20% annually, not 40%

• There aren’t enough mispriced contracts at any one time to scale infinitely

• Liquidity dries up at the extremes — your own buying pressure moves prices against you

• Fees, slippage, and capital lockup eat margin

• The best operators have seven-figure capital bases and dedicated infrastructure

This isn’t a side hustle. It’s a real business.

The Real Takeaway for Business Owners

You’re not trying to beat Polymarket. You’re trying to beat your competitors.

Every architecture above maps to something in your industry:

• #1 Calibration Fade = build AI scoring for “obvious” market outcomes vs. true probability

• #2 Longshot Reversal = base-rate analysis on opportunities that look like duds

• #3 News Overreaction = sentiment monitoring with 72-hour decision buffers

• #4 Underconfidence = measure YOUR industry’s pricing bias direction

• #5 Arbitrage = monitor multiple platforms/competitors/regions in parallel

• #6 Real-World Data = connect physical sensor and satellite feeds to business decisions

• #7 Whale-Lag = automated monitoring of dominant players’ public signals

• #8 Court/Regulatory = paper trail monitoring for regulated industries (massive ROI)

• #9 OSINT = multi-source niche channel monitoring

• #10 Resolution Source = automated polling of data sources that determine outcomes you care about

Total estimated build cost to implement ANY ONE of these in your business: $5,000-50,000. Total ongoing cost: $200-2,000/month. Total competitive advantage: years.

The companies that win the next decade won’t be the ones with the biggest budgets. They’ll be the ones who built AI systems to read the world more quickly and accurately than their competitors.

The Polymarket traders are doing it visibly, in a public market, with money on the line. Your competitors are doing it quietly, in your industry, with your customers.

The question is whether you’re watching.

Stay smart,

The SmartOwner Team

LEGAL DISCLAIMER

This article is published by SmartOwner for general informational and educational purposes only. Nothing contained herein constitutes investment, financial, legal, tax, or trading advice, nor a solicitation, recommendation, or endorsement of any specific security, trading strategy, prediction market, platform, or product. SmartOwner is not a registered investment advisor, broker-dealer, or fiduciary. All trading and investment activities involve substantial risk of loss, including total loss of capital. Past performance is not indicative of future results. Prediction market access varies by jurisdiction and may be restricted or prohibited where you live. You are solely responsible for verifying the legality of any activity in your jurisdiction. Before making any financial decision, consult a qualified, licensed professional. SmartOwner and its affiliates accept no liability for any loss arising from reliance on the information in this article.

If you’ve never used Polymarket, the whole thing can sound like Vegas with extra steps. It’s not. It’s actually one of the most accurate forecasting tools on the planet — and there’s a single pricing quirk inside it that explains why a small group of AI-powered traders are quietly winning a lot of money.

More importantly, that same quirk shows up in almost every market you make decisions in: real estate, stocks, hiring, B2B services, used cars. Once you see it, you can’t unsee it.

Let’s break it down.

How Polymarket Pricing Works

On Polymarket, every “yes/no” question has a price between 1 cent and 99 cents. That price IS the probability the crowd thinks the event will happen.

• A contract at 10 cents = the crowd thinks there’s a 10% chance it happens

• A contract at 50 cents = the crowd thinks it’s a coin flip

• A contract at 90 cents = the crowd thinks there’s a 90% chance it happens

If you buy a “yes” contract for 10 cents and it happens, you get paid $1. That’s a 10x return. If it doesn’t happen, you lose your 10 cents.

If you buy a “yes” contract for 90 cents and it happens, you get paid $1. That’s only about an 11% return. If it doesn’t happen, you lose 90 cents.

So the price tells you two things at once: how likely the crowd thinks it is, AND how much you stand to win or lose.

What “Resolve in Favor of the Buyer” Means

When the question gets answered in real life, the contract “resolves.” Either “yes” wins or “no” wins. The winners get paid $1 per contract.

So when researchers say “contracts priced at 10 cents resolve in favor of the buyer more than 10% of the time,” it means: when they looked at every contract priced around 10 cents on Polymarket and tracked what actually happened, those contracts came true MORE OFTEN than 10% of the time.

If the crowd was perfectly accurate, contracts at 10 cents should come true exactly 10% of the time. That’s what “10 cents” literally means — a 10% probability.

But they don’t. They come true around 14% of the time.

Why This Is A Big Deal

If you bought every single 10-cent contract on Polymarket, here’s what would happen on average:

• You’d be wrong 86% of the time (losing 10 cents each time)

• You’d be right 14% of the time (winning 90 cents each time, since you get $1 back on a 10-cent bet)

The math: for every 100 contracts you buy:

• 86 losses × 10 cents = -$8.60

• 14 wins × 90 cents profit = +$12.60

• Net result on a hypothetical $10 invested: roughly +$4 in this simplified example (a 40% return)

The crowd is underpricing longshots. They’re selling 10-cent contracts that should really cost about 14 cents.

(Note: this is a simplified illustration that ignores trading fees, market liquidity, and the fact that you can’t actually buy “every” 10-cent contract. Real-world returns vary significantly. The point isn’t the specific number — it’s that a measurable gap exists between crowd pricing and reality.)

The Same Thing Happens on the High End

“Contracts priced at 90 cents resolve less than 90% of the time” means the opposite problem.

When the crowd thinks something is 90% likely, it actually only comes true around 85% of the time. So if you buy “yes” at 90 cents, you’re overpaying.

The smarter move is to bet AGAINST it. Buy “no” at 10 cents. That “no” contract should be priced at 15 cents (because the “yes” should be 85, not 90). So you’re getting a discount betting against the favorite.

Why This Happens (The Human Emotion Part)

Two psychological patterns drive this:

1. People love lottery tickets (the longshot bias).

Imagine a market: “Will an unknown rookie win MVP this year?” Real probability: 3%. The crowd prices it at 5-6 cents because winning $1 from 5 cents feels exciting. People want the dream. They overpay for the lottery ticket.

This is the same reason people buy actual lottery tickets despite the terrible odds. Hope is worth something emotionally. So lottery-style bets get overpriced.

2. People want certainty (the favorite bias).

Imagine a market: “Will the incumbent President finish their term?” Real probability: 92%. The crowd prices it at 95-96 cents because being on the winning side feels safe. People will pay a premium to “win” even if the return is tiny.

This is the same reason people buy bonds with terrible yields — safety has emotional value. So safe bets get overpriced too.

The result: prices are systematically wrong at both extremes. Too high on favorites, too high on longshots.

Why AI Beats Humans At This

A human looks at a 10-cent contract and thinks “ugh, I’ll probably lose, why bother.”

A human looks at a 90-cent contract and thinks “sure thing, easy money.”

Both reactions are emotional. Both are wrong.

AI doesn’t care about feeling smart or feeling safe. It just does the math:

• Runs the historical data

• Finds prices that are mispriced relative to true outcomes

• Places a portfolio of bets that exploits the gap

• Doesn’t get bored. Doesn’t get scared. Doesn’t chase losses.

That’s the entire edge. Not prediction — just removing the emotional discount that humans build into prices.

The Business Translation

This is the part that actually matters for your business. The same thing happens in every market with human buyers:

• Real estate: “hot” houses sell for more than they’re worth, “ugly” houses sell for less

• Stocks: popular stocks trade at premiums, unpopular ones at discounts

• Used cars: “luxury brand” 10-year-old cars cost too much, “boring brand” 5-year-old cars cost too little

• Hiring: candidates with prestigious resumes are overpaid, candidates with non-traditional backgrounds are underpaid

• B2B services: big-name agencies charge 3x more for the same work as smaller ones

Wherever humans set prices based on emotion (excitement, fear, prestige), there’s a calibration gap. Whoever measures the gap and acts on it — wins.

The AI advantage isn’t magic. It’s just the patience and discipline to do the math while everyone else is reacting to feelings.

That’s the lesson hiding inside the prediction market story. And it’s playing out in your industry right now, whether you can see it or not.

Stay smart,

The SmartOwner Team

LEGAL DISCLAIMER

This article is published by SmartOwner for general informational and educational purposes only. Nothing contained herein constitutes investment, financial, legal, tax, or trading advice, nor a solicitation, recommendation, or endorsement of any specific security, trading strategy, prediction market, platform, or product. SmartOwner is not a registered investment advisor, broker-dealer, or fiduciary. All trading and investment activities involve substantial risk of loss, including total loss of capital. Past performance is not indicative of future results. Prediction market access varies by jurisdiction and may be restricted or prohibited where you live. You are solely responsible for verifying the legality of any activity in your jurisdiction. Before making any financial decision, consult a qualified, licensed professional. SmartOwner and its affiliates accept no liability for any loss arising from reliance on the information in this article.

That is not a typo. Same business. Same services. Same town. We changed the website, and Google went from quietly ignoring them to shouting their name from the rooftops.

I have been doing digital marketing for over 15 years. I have never seen a swing like this from a rebuild alone. So I want to walk you through exactly what we did, because the playbook behind it is now public and free.

Words of the Day: SEO vs AIO vs GEO

For 25 years there was one word for getting found online: SEO. There are now three, and they are three different jobs.

SEO (Search Engine Optimization) is getting found in classic Google search. AIO (AI Optimization) is making your own website easy for AI engines to read and trust. GEO (Generative Engine Optimization) is getting the rest of the web to talk about you so AI recommends you by name.

Using a restaurant as an example. SEO is the restaurant being in the best location in town. AIO is writing a menu so clear that a stranger could describe your food perfectly to a friend. GEO is having food critics, regulars, and the local paper all talking about you, so when someone asks “where should I eat,” your name comes up.

For your business, this matters because customers now find restaurants three different ways. They drive by. They read the menu online. They ask a friend. If you only show up in one of those, you are invisible to the other two crowds. And the other two are where almost nobody else is working yet.

The 12% problem

Here is the number that forced us to split the work.

Only about 12% of the sources ChatGPT cites also rank on the first page of Google.

Read that again. If only 12% of what ChatGPT cites is also on Google’s first page, then 88% of what AI engines pull from is coming from places your regular SEO has never touched. Ranking on Google used to be enough. It is not anymore.

So we stopped running it as one job and started running it as three:

You do not abandon Google. Classic search still drives most of the click traffic for most businesses. You just stop pretending it’s the only game.

What we actually did to the client’s site

We did not do all 26 pages of the playbook on this one. We did the foundation.

We rebuilt it in clean code. The old site was a heavy, plugin-stuffed setup. We rebuilt it in a modern framework that ships almost no junk code and loads in well under a second. Speed is a powerful ranking factor. Faster pages get cited more often.

We let the AI crawlers in. Most websites are accidentally blocking ChatGPT and Claude from reading them at all. This is binary. Blocked means zero AI visibility, no matter how good your content is. We explicitly opened the door.

We added a cheat sheet for AI. We placed plain-text files at the root of the site that tell AI engines exactly what the business does, where, and for whom. Together these act as a fast, authoritative source AI can quote from.

We structured the content so answers are easy to grab. Direct answers up top, real FAQ pairs, comparison tables. Pages with comparison tables and FAQ blocks see roughly 44% more AI citations.

That is the short version. Foundation work. Nothing exotic.

A few things worth saying out loud.

Clicks more than doubled. Nearly 2,900 total clicks since launch, with a best single day of 94.

The site moved from page 2 to the top of page 1. This is the one that matters most. Page 2 of Google is famously where you hide a dead body, because nobody clicks it. The site went from a 20.8 average position to as high as 2.4 on its best recent day.

Over 820,000 total search appearances since launch. In under 8 weeks. One day alone hit more than 113,000 impressions.

Individual pages are climbing too. Their Order of Protection page is sitting at position 5.9. Their Consumer Protection page is at 10. Those are page-one spots for pages that did not meaningfully exist before.

The honest part

I am not going to tell you this happens every time, because it does not.

This client had room to climb. The old site was genuinely holding them back, which means there was a lot of trapped value to release. A business already ranking well will not see an 8x impression jump from a rebuild. The gap between “we fixed your foundation” and “you will 8x overnight” is a canyon, and I am not going to pretend otherwise.

I will also be honest that two months is early. Rankings can wobble. We watch these numbers weekly, not as a one-time win.

What I will say with confidence is this. The old site was invisible to a degree the owner did not know. Most owners do not know. That is the actual problem.

How to check your own site, free, in 60 seconds

You do not need to hire anyone to find out where you stand. We built a free scanner so you can check this yourself.

Go to scan.digitaltreehouse.com

Type in your website. No signup. It takes about 60 seconds. Here is what it does for you:

1. SEO check. Your meta tags, schema, page speed, and the Google fundamentals.

2. AIO scan. Whether your llms.txt file exists, whether AI crawlers can reach you, and whether your content is readable by AI engines.

3. Live AI search. This is my favorite part. It actually asks ChatGPT and Gemini about your business and shows you what they say back.

That last one is worth the 60 seconds by itself. Most owners have never once asked an AI engine to describe their own business. The answer is often wrong, outdated, or blank. Better you see it than your customer.

When the scan finishes, you also get the full 26-page playbook as a free download. The same one we use internally on every client engagement.

The Bottom Line

The rules for getting found changed, and most businesses have not noticed yet.

Google still matters. It still drives most of the clicks. But AI engines are now a second front door, and the work that opens that door is different work. Clean code. Crawler access. Content an AI can actually quote. Most of it is foundation, not magic.

The client result I shared is real, and it is unusually large because the old site was unusually broken. Yours might not be that broken. Or it might be more broken than you think. There is exactly one way to find out.

Run the scan at scan.digitaltreehouse.com. You will get your own SEO, AIO, and GEO grade in about a minute, and the full playbook to fix whatever it finds. Hand both to whoever runs your website.

Page 2 is where websites go to be forgotten. You do not have to live there.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

Word of the Day: Connector

A connector is a small piece of software that lets an AI tool reach inside another app you already use, read what is in there, and take action on your behalf.

Think of a connector like a key on your key ring. The AI is you. The key opens a specific door. One key for QuickBooks. One for your email. One for Canva. The AI does not need to learn each app from scratch every time. It just picks up the right key and walks in.

For your business, this matters because the moment AI can reach into your accounting software, your payment processor, and your inbox at the same time, the boring back-office work that eats your Saturdays starts to disappear. Without connectors, AI is just a chat window. With connectors, it is an employee.

What Claude for Small Business actually does

Here is the plain English version.

You already pay $20 or more a month for Claude (or you should). On Wednesday, Anthropic added a toggle inside Claude that turns it into something different. Flip the toggle, and Claude can now plug directly into the tools you already use to run your business.

The system ships with 15 prebuilt workflows. Planning payroll. Closing the books for the month. Onboarding a new employee. Chasing overdue invoices. Running a sales campaign when your numbers dip.

You approve everything before it sends, posts, or pays. Claude does the work. You hit the green button.

This is free if you already pay for Claude. There is no upsell. No “Small Business Edition.” It is a toggle inside the version you are already paying for.

One thing that will trip people up: where this actually lives

This is the part the headlines skip, and it matters, because if you get it wrong you will spend twenty minutes hunting for a button that is not there.

The toggle is not in the Claude website. It is in the Claude desktop app, in a mode called Cowork. Cowork is the version of Claude that can actually do multi-step work across your files instead of just talking about it.

The plain English distinction: regular Claude in your web browser can chat, and it can connect to a few tools on its own. But the packaged small-business workflows, the payroll planner, the invoice chaser, the month-end close, only exist inside the desktop app. If you only ever use Claude in a browser tab, you have not seen this yet.

(For the technical readers running their own AIOS setups: yes, the same plugin can be pulled into Claude Code through Anthropic’s plugin marketplace, but that is not the official path and the two do not sync. For everyone else, ignore that sentence. Desktop app, Cowork mode. That is the route.)

Why this matters more than the press release admits

For years, the gap between big company AI and small company AI has been an integration problem, not a brains problem. ChatGPT and Claude have always been smart enough to do the work. The trouble was that you, the owner, had to manually copy data from QuickBooks, paste it into the chat, ask a question, get an answer, then return to HubSpot or QuickBooks or wherever and make the updates.

Every step was a place to give up.

Connectors close that gap. The data flows directly. The AI does the work. The owner reviews and approves. The Saturday afternoon you used to spend reconciling your books becomes ten minutes on Tuesday. (Want to skip the approval step? You’ll need to set up an AIOS via Claude Code, that guide coming soon.)

There is also a quieter story here. Anthropic is betting that the next big AI battle is not over who can serve Walmart. It is over who can serve the 50-person HVAC company and the 25-person landscaping crew. The first company to make AI feel like an employee for those owners wins the next decade.

That is a market worth showing up for. And until Wednesday, almost nobody was.

Action steps if you want to try it this week

Here is the order I’d do it in. Remember: this lives in the desktop app, not the website.

1. Make sure you have a paid Claude plan. The package works on Pro ($17/month if you pay yearly, $20/month month-to-month), Max ($100 or $200/month), or Team ($20 per seat/month). Honest heads-up: Cowork uses up your usage limits faster than regular chat, so start small and don’t panic if you hit a limit early.

2. Download the Claude desktop app at claude.com/download. Install it, sign in.

3. Open the desktop app and sign in if prompted. You’ll see modes including Chat, Code, and Cowork. Click into Cowork.

   

4. In Cowork’s left sidebar, click Customize. Under Plugins, click the + button. Find Small Business and click Install.

   

5. Once installed, just type “get me started” and Claude will walk you through setup. Then connect ONE tool to begin. I’d start with QuickBooks if you have it, or Gmail/Google Workspace if you don’t.

6. Ask a small, low-risk question first. “What did I spend on advertising last month in QuickBooks?” or “Summarize the last five emails from my best customer.” Reading only. No money moving.

7. If that works, add a second tool. Then a third. Build trust through small wins before you let it near anything that sends or pays.

8. Fair warning: Cowork is still labeled a research preview. It is powerful but not bulletproof. Watch what it does early on, and approve every action until you trust it.

There is also a free training course Anthropic built, AI Fluency for Small Business. On-demand, fourteen short lectures, about an hour total. Reach it from Anthropic’s Claude for Small Business page. If you live near Chicago, Tulsa, Dallas, Hamilton Township NJ, Baton Rouge, Birmingham, Salt Lake City, Baltimore, San Jose, or Indianapolis, Anthropic is running free half-day in-person workshops in those cities.

In other AI news this week

Amazon retired Rufus. If you have shopped on Amazon in the last year, you have seen Rufus, the little AI chat bubble in the corner. Amazon killed it. Search now triggers an assistant called “Alexa for Shopping” that sits front and center in the search bar across Amazon.com, the app, and Echo Show devices. Type “what’s a good skincare routine for men” and you get an AI answer, not a list of products. This is the same shift Google made with AI Overviews, and it carries the same lesson: if your business sells anything online, the search bar is becoming a conversation. Your product listings now need to answer questions, not just match keywords.

Anthropic and the Gates Foundation are putting $200 million into AI tools for global health and education. Worth noting as a signal of where serious public-benefit money is going.

The Bottom Line

For the first time, a major AI company built a product where you are the customer they were thinking about. Not the Fortune 500 CFO. Not the developer. You. The owner running a real business with a real to-do list that gets longer every week.

Connectors are the unlock. They are what turn AI from “interesting chat tool I sometimes open” into “thing that runs my back office so I don’t have to.”

A few catches worth repeating. It lives in the desktop app, it is still a research preview, and it will make mistakes. Go in with your eyes open and your finger on the approve button.

But the direction is unmistakable. This is the moment AI stops being something happening to other people’s businesses and starts being something running inside yours.

Need help getting started? That’s what I’m here for, reply to this email.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

For thirty years, software companies built behind closed doors for two or three years, launched, and then dragged their feet on updates. AI just broke that model. A new kind of company is emerging that ships rough versions to real customers on day one and shapes the product around them in real time.

Today’s newsletter is about why that shift is happening, what it means for any business that buys (or builds) software, and how I’m running this exact playbook right now on a product called OmniLegislation. I’ll also tell you why OpenAI just quietly launched a four-billion-dollar service company, because it’s the same story from the other direction.

Let’s get into it.

Word of the Day: Adaptive Software

Adaptive software is a product that gets shaped around its customers in real time, instead of being designed in a lab for two years and then launched at them.

Think of it like the difference between a tailor and a department store. A department store stocks one hundred different sizes of the same shirt and hopes one of them fits you. A tailor measures your shoulders, asks where you wear it, and adjusts the fabric in front of you. Same shirt category. Completely different result. The tailored one fits perfectly.

For your business, this matters because the software you buy is about to get a whole lot more tailored. And the software you might build for yourself is about to get a whole lot more possible.

The old way is breaking

For decades, software worked like this. A company spent two or three years building a product behind closed doors. They guessed at what customers wanted. They launched. Then they sat in quarterly planning meetings deciding which features to add over the next year.

If you wanted them to change something for your specific business, you got one of two answers. “That’s on our roadmap” (meaning “no, but politely”) or “we can do that as a custom build” (meaning “open your wallet”).

AI just changed the math on all of this.

What’s actually new

Recently I read an interview with Boris Cherny, the engineer who built Claude Code at Anthropic. He said something that stuck with me.

“The best person to write accounting software won’t be an engineer. It will be an accountant.”

Think about that for a second. The person who knows what accounting software should do is an accountant, not the person who knows how to code. Until now, those two people had to find each other, work together for two years, and pray they understood each other. AI just made the accountant capable of building the software directly.

That changes who builds what. It also changes how fast it gets built. A small team, or even one person who knows the work, can now:

• Build a rough version in a weekend

• Put it in front of paying clients on Monday

• Customize it quickly as requested

• Test new ideas with real users instead of focus groups

• Run several different versions of the same product at the same time

• Ship daily instead of quarterly

That used to be a dream. Now it’s a Tuesday.

The proof: a story I’m living right now

I’m not going to ask you to take any of this on faith. Let me show you what it looks like in practice.

Right now, my team at DigitalTreehouse is building a product called OmniLegislation. It’s a tool that watches every state and federal bill being introduced, debated, and voted on, and alerts the right people the moment something changes. Attorneys care about this. Lobbyists live and die by it. Today, the existing tools in this space are expensive, slow, and built around twenty-year-old assumptions.

Here’s the part I want you to pay attention to. We are not building this product in a basement and then unveiling it. We are doing the exact opposite.

We are letting attorneys and lobbyists use our legislative monitor right now, for free, in exchange for one thing. Honest feedback. What works. What doesn’t. What’s missing. What slows them down. What they wish it could do that no tool currently does.

In return, the software is getting shaped around their actual workflow, week by week. The features that get built next are not coming from a strategy deck. They are coming from real users telling us, “if Omni could just do this one thing, you would replace three tools I currently pay for.”

That is adaptive software in motion. The customer is the product manager. AI is the engineering team. We are the connective tissue.

If you happen to be an attorney or a lobbyist reading this, reply to this email. We will get you in.

See our Legislative Scanning Tool live.

The line is disappearing

Here’s the strange part. The old distinction between “software companies” and “service companies” is starting to blur.

A consulting firm can now build software for a client that becomes the spine of a larger platform. A software company can now offer hands-on customization without going bankrupt. Both ends meet in the middle.

OpenAI is doing both ends of this at once, in public. Yesterday they launched a brand-new arm of the company called the OpenAI Deployment Company, with more than four billion dollars behind it. The job of the new unit is to help businesses actually use AI inside their real workflows, not just give them a chatbot and a login. The quote from OpenAI is almost an admission of guilt. The bottleneck, they say, is no longer access to the model. It’s understanding how to use it. It’s connecting the model to your messy data, your permissions, your approvals, your existing tools, and your team.

Last night they followed that with a developer update that goes even further. One of their own engineers, talking about a project called Symphony, said this:

“Six months ago we made a pretty wild decision: no human-written code in the repo.”

They forced themselves to use AI to build their own product. The result? A 500% increase in shipped work. Inside the company that builds the AI, the humans are no longer the ones typing. The humans are the ones deciding what to build next.

If OpenAI is moving in this direction, every other software company will follow. They have to.

The trap

I want to give you the good news and the warning together, because too many people only hear the first half.

The good news: you can now get software that fits your business instead of forcing your business to fit the software.

The warning: this only works if there is discipline behind it.

Customization without discipline is a slow-motion disaster. If you say yes to every weird request from your team or clients, you end up with Frankenstein code that no one can fix. AI does not solve this. AI just makes it cheaper and faster to create the mess.

The companies that will win in this new era are the ones that move quickly and stay simple at the same time. They customize where it counts. They refuse the edge cases. They watch for patterns in software usage and client requests and turn the repeating ones into core features. They keep one foot in the boutique business and one foot in the platform business.

That’s harder than it sounds. Strategy still matters.

What you can actually do this week

If there’s a tool your business needs and either the software doesn’t exist yet, or the one you’re currently paying for doesn’t quite fit how you actually work, you can build a rough version of it this month. Not a polished one. A rough one. The goal is to get something working that solves the most painful part of the problem. You can make it pretty later.

Here are the tools to start with, depending on what you’re trying to build.

Lovable (lovable.dev). The friendliest starting point if you have never built anything before. You describe what you want in plain English and Lovable builds you a working website or app you can share with a link. Great for internal tools, client portals, landing pages, and simple workflow apps. Around $25 a month.

Bolt (bolt.new). Similar to Lovable but a little more developer-flavored. Good for slightly more complex apps where you want more control over how things look and work under the hood. Free to start, around $20 a month for serious use.

Claude Code (claude.com/claude-code). The right tool when your project gets bigger and you need real code running on real servers. This is what we use at DigitalTreehouse for production builds. It comes with a Claude Pro subscription at $20 a month, or you can get heavier usage through Claude Max plans.

Replit (replit.com). Best for full apps you want to host online without thinking about servers. It writes the code, runs it, and publishes it for you, all in one browser tab. Around $25 a month for the AI features.

Here is the simplest possible way to start. Pick one tool from the list. Open it. Type one sentence describing the problem you want to solve. See what it builds. Then talk to it like a coworker. Tell it what is wrong. Tell it what to change. Tell it what to add. Keep going until it works.

Once you have something rough, put it in front of two or three real users you trust. Watch what they actually do with it, not what they say they want. Fix the things they trip over. Ignore the things they only suggested. Repeat that loop for two weeks and you will have something better than what most software companies ship in a year.

If you want help, reply to this email. This is exactly what we do at DigitalTreehouse, and it’s the same playbook we’re running on OmniLegislation right now.

The Bottom Line

For thirty years, software companies built first and listened second. The good ones listened a little faster. None of them listened in real time.

That era is ending. Not because the technology is finally smart enough to figure out what we want. Because the cost of making something, putting it in front of a real user, and changing it on the fly is finally close to zero.

OmniLegislation is one example. Yours could be next. The companies that win the next ten years will not be the ones with the most features on day one. They will be the ones who learn the fastest from real customers and have the discipline not to lose themselves in the noise.

Build rough. Ship early. Listen hard. Cut ruthlessly. Repeat.

That’s it. That’s the whole thing.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

A small business in Austin, Texas, makes vegan cheese. They ship it across the country in dry ice. Tens of thousands of orders a year, mostly during the holidays. They had a great Q4 last year. Best ever.

Then the founder, a woman named Kirsten Maitland, looked at the bank account and felt sick. The numbers were wrong. The profit she thought she’d made wasn’t there.

She started digging. A year of invoices. Hundreds of pages every single week. Fees stacked on top of fees on top of fees. She was working through it by hand and getting nowhere.

So she opened Claude and uploaded the whole pile.

What happened next is the most useful AI story I’ve read this year. And every single business owner reading this needs to know about it.

Word of the Day: Discrepancy Detection

Discrepancy Detection is using AI to compare two sets of data, like an invoice and a contract, and flag every place they don’t match.

Think of it like having a forensic accountant who works for free, never sleeps, and doesn’t get tired on page 400 of a 600-page bill.

For your business, this matters because every vendor you pay, your shipper, your software subscriptions, your merchant processor, your insurance company, is sending you bills that almost nobody actually checks line by line. The errors are not always honest. And until now, catching them required a person, a spreadsheet, and a free week.

You don’t need any of that anymore.

What Rebel Cheese actually did

Here’s the play, step by step. I’ll keep it simple because the lesson is simple, even if the work involved some real thinking.

Step one. She handed Claude a year of shipping invoices and her actual carrier contract. She didn’t ask Claude to “find errors.” She asked it to look for patterns. Within minutes, Claude surfaced things a human auditor would have spent weeks finding. Bulging packages charged at premium rates. Weight overages priced under a clause buried on page nineteen of the contract. A new weight limit the carrier had quietly added in early 2025 and never told her about.

Step two. She used the analysis to negotiate. When she sat down with the carrier, she didn’t show up frustrated. She showed up with data. She knew exactly which contract clauses were doing the damage and exactly how much they had cost her. The carrier’s response when she asked why nobody told her about the new weight rule? “Well, you should have caught it.”

That’s the line that should make every business owner reading this sit up straight.

Step three. She built a tool so it never happens again. She used Manus, an AI orchestration platform, to build a small web app that compares every weekly invoice against her contract automatically. Anything overcharged by more than ten cents gets flagged and turned into a credit request, which goes to the carrier the next day. The carrier has approved every single claim she’s submitted.

She now saves between $1,500 and $4,000 every single week.

Her total monthly spend on the AI tools that do this work? About $200.

The part that matters most for you

I want you to read this next sentence twice.

Kirsten Maitland is not a developer. She does not have an engineering team. She does not have a data analytics department. She told Fast Company that a few years ago, this kind of project would have meant hiring a consultant for tens of thousands of dollars and waiting six months for results.

She built it herself in an afternoon.

That is the whole story. That is what AI actually changed. Not the chatbots. Not the cute image generators. The fact that one motivated person with a problem and a Claude subscription can now do work that used to require a small team and a budget approval.

This is Vibe Coding. Building software by describing what you want in plain English, letting AI handle the technical part, and shipping something that solves a real business problem. It’s not theory. It’s a vegan cheese shop saving $200,000 a year.

What you can audit this week

Pull out your last six months of bills from any of the following and try this yourself:

• Your shipping carrier (UPS, FedEx, USPS, DHL)

• Your merchant processor (Stripe, Square, PayPal, your bank)

• Your software subscriptions (Microsoft, Adobe, Google Workspace, Salesforce)

• Your phone and internet provider

• Your business insurance

• Your payroll provider

Then do exactly what Kirsten did. Open Claude. Upload the invoices and your contract or your original quote. Ask it to find every line that doesn’t match what you agreed to pay.

You will find money. I promise you. Every business owner I know who has done a real audit of their vendor bills has found errors. It’s not always malice. Sometimes it’s a billing system that quietly raised a rate. Sometimes it’s a fee that was waived in your contract but billed anyway. Sometimes it’s a “fuel surcharge” that doesn’t match the formula in your agreement.

The reason most business owners never find these errors is not that the errors are hidden well. It’s that finding them used to take a forensic accountant a week and cost more than the errors were worth.

That math just changed.

The honest part

I have to be honest with you about something. This is harder than I’m making it sound.

Kirsten didn’t just paste invoices into a chat window and watch money fall out of the sky. She had to figure out how to structure her data so Claude could compare it against the contract cleanly. She built two CSV files first, one with her negotiated rates, one with the actual invoice charges. She wrote a real “requirements document” before she asked Manus to build anything. She tested. She iterated.

The first attempt didn’t work. Probably the second didn’t either.

That is what real Vibe Coding looks like. The work works on you. The gap between “Claude found something interesting” and “I have an automated system saving me $3,000 a week” is a canyon. Crossable, but a canyon.

Most people give up at the first failed prompt and tell themselves AI doesn’t work. They’re wrong. They just stopped too early.

The Bottom Line

Every business has money leaking somewhere. Vendors overcharge. Contracts go unchecked. Errors compound. For decades, the only way to find that money was to pay a human a lot of money to look for it, which usually cost more than the money you’d recover.

That equation is broken now. The audit has gone from a project to an afternoon. From $20,000 to $200 a month.

Rebel Cheese found a quarter of a million dollars hiding in their shipping bills. What’s hiding in yours?

Quick action steps for you

If you want to try the Rebel Cheese play this week, here’s the order I’d do it in.

1. Pick one vendor where you spend the most money each month. Just one. Don’t try to audit your whole company at once.

2. Download or save as PDF the last six months of invoices from that vendor.

3. Find your original contract or pricing agreement with them. (If you can’t find it, email your account rep and ask for a copy.)

4. Go to claude.ai and start a new chat. (The free plan works for this. Claude Pro at $20/month works better.)

5. Upload your contract first. Tell Claude: “This is my contract with [Vendor]. Please read it and tell me back the rates and fees I should be paying.”

6. Then upload your invoices. Tell Claude: “Here are six months of invoices from this same vendor. Please compare them to the contract you just read and list every line item where the invoice charged me more than the contract specifies.”

7. Read what comes back carefully. Spot-check two or three of the items yourself to make sure Claude got it right.

8. If you find real errors, write your account rep a calm, factual email listing the discrepancies and asking for credit.

That’s the whole play. If you get stuck, reply to this email. We do this kind of work for clients at DigitalTreehouse every week.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

90 to 95 percent of people who visit your website leave without doing anything. They land. They glance. They bounce. Gone.

You spent money on ads to get them there. You hired someone to design that page. Your team obsesses over button colors. And almost every single visitor is leaving like they walked into the wrong room at a party.

A startup called Interact AI made the case yesterday that the website itself is the problem. Not your copy. Not your design. The whole 25-year-old idea of a static page.

And they may be right.

Word of the Day: Adaptive Interface

An adaptive interface is a website (or app screen) that rebuilds itself in real time based on who’s looking at it and what they’re asking.

Think of it like the difference between a brochure and a salesperson. A brochure says the same thing to everybody who picks it up. A good salesperson asks what you need, then walks you to the exact thing that solves your problem, in the exact words you’ll understand.

For your business, this matters because your current website is a brochure. It tells the same story to a 60-year-old contractor and a 30-year-old marketing director. An adaptive interface treats them like two different visitors, because they are.

The website hasn’t changed since 2001

Stop and let that sink in for a second.

The iPhone didn’t exist in 2001. Google was a baby. SpaceX was a fantasy. But the way every company on Earth meets every customer on Earth, that has barely moved. A page of text. Some images. A “Book a Demo” button that nobody clicks.

Marketing teams have spent 25 years A/B testing button colors and calling it progress. Meanwhile, 9 out of 10 visitors keep leaving. Every sales call still starts with “let me explain what we do.” Every service business still loses leads to the form-fill black hole.

Interact AI ran a test with a software company called Sprinto. Two months in:

• Conversion doubled

• Sales cycles got cut in half

• Interact-generated leads converted 90 percent better than normal ones

• $500K in new pipeline

One company. One category. Two months. That’s not a tweak. That’s a different game.

What it actually does

When someone lands on your site, instead of seeing a homepage, they get a real conversation. Voice or text, their choice. The page composes itself around what they’re asking.

A contractor asks about pricing? They see pricing, in their language, with examples from similar contractors.

A nurse asks about HIPAA? She gets the compliance view.

A skeptical buyer says “but how does this actually work?” The page builds them a live demo on the fly.

It’s not a chatbot. A chatbot is a little box in the corner you ignore. This is the whole page, rebuilding itself for one person at a time, then booking the meeting when they’re ready.

What this means for you (today, not someday)

Look, Interact AI is brand new. They’re focused on B2B software companies right now. You probably can’t sign up for it tomorrow.

But the bigger point applies to every business owner reading this. Your website is the front door of your business, and the front door has been broken for a long time. You just got used to it.

Here’s what I’d do this week. Three things.

One. Open your own website on your phone. Pretend you’re a customer who has never heard of you. Read the first screen. Could you tell what you sell? Could you tell who it’s for? Could you find the price? If any of those answers is “not really,” your site is part of the 90 percent problem.

Two. Add a real chatbot, not a fake one. Tools like Intercom Fin (starts at about $39 per resolution) or HubSpot’s free Breeze chatbot will let your site actually answer questions instead of dumping people to a contact form. This is not adaptive interface. It’s a half-step. But it beats what you have now.

Three. Watch this space. When Interact AI or one of its copycats opens up to small businesses, you want to be early. The companies who put this on their site in 2026 are going to look like the companies who got SEO in 2005. Quietly, they’re going to eat the lunch of the ones who didn’t.

While we’re here, your AI tools just got faster

Quick second story for the vibe coders in the room.

Yesterday, Anthropic and SpaceX announced a deal that nobody saw coming. Months ago, Elon Musk publicly called Anthropic “Misanthropic” and said they hate Western Civilization. Yesterday, he rented them his entire 220,000-GPU supercomputer in Memphis.

What this means for you in plain English: Claude Code’s usage limits just doubled across all paid plans. No more peak-hour throttling. If you’ve been hitting the wall while building with Claude (or with OpenClaw, which runs on Claude), you have twice the headroom starting now.

If you’ve been waiting for “the right time” to try vibe coding a small tool for your business, the right time is now. The capacity is there. The price didn’t go up. You just got more of what you already paid for.

The Bottom Line

The website is a 25-year-old technology that bounces 90 percent of its visitors. Everybody knows it. Almost nobody talks about it. And we just spent a generation pretending button colors were the problem.

The fix isn’t a better page. It’s a different idea of what a page even is.

Is Interact AI the company that wins this? Maybe. Maybe not. The thing they’re pointing at is real either way. Your website is going to start talking back. The only question is whether that happens to your business in 2026 or 2028.

I’d rather be the one who saw it coming.

That’s it. That’s the whole thing.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

Most of it never shipped.

A year later, Siri is still the same frustrating Siri. ChatGPT and Claude run circles around it. And in April, Tim Cook announced he’s stepping down, with John Ternus taking over as CEO on September 1.

Yesterday, Apple quietly admitted what everyone already knew.

What Apple actually did

Bloomberg broke the news. Apple is going to let users choose from a range of outside artificial intelligence services to power features across its software, slated for iOS 27, iPadOS 27 and macOS 27 this fall. Bloomberg

The internal name is Extensions. It will let iPhone users choose from multiple third-party AI models from companies like Google and Anthropic, including the ability to set custom voices in Siri depending on which external model is responding. 9to5Mac

In plain English: starting this fall, you’ll open Settings on your iPhone and pick Claude or Gemini (and probably ChatGPT) to run Siri, Writing Tools, and Image Playground.

Apple is spinning this as a generous gift to users. More choice. More flexibility. Power to the people.

It is not a gift.

It’s the most valuable company in human history admitting they can’t compete in AI, so they’re letting the people who can do it for them, on their phone.

The real story is not Apple. It’s you.

Here’s the part I want you to sit with for a minute.

Apple has more cash than most countries. They have the smartest hardware engineers on Earth. They have a four trillion dollar market cap and the most loyal customer base in the history of products.

And they still couldn’t ship AI fast enough.

If Apple is not safe from being late, nobody is safe from being late. Not your accounting firm. Not your law practice. Not your HVAC company. Not your dental office.

This is the lesson that’s hiding inside the iOS 27 announcement, and almost nobody is talking about it.

The threat isn’t AI replacing you. It’s your competitors using AI replacing you.

A lot of business owners I talk to comfort themselves with the same line.

“AI can’t do what I do.”

Maybe. Maybe not. But that’s the wrong question.

The right question is this: Can AI help your competitor do what you do, but ten times faster, ten times cheaper, or ten times better?

If the answer is yes, and the answer is almost always yes, you have a problem. Because customers don’t care that you’re irreplaceable. They care that the other guy got back to them in four minutes instead of four hours. They care that the other quote came in cheaper. They care that the other proposal looked like it was written for them, not for everybody.

Apple’s customers didn’t care that Apple is Apple. They cared that Siri couldn’t do what ChatGPT could. So they used ChatGPT.

Your customers will do the same to you.

What you should actually do

You don’t need iOS 27. You need to start now. Three steps.

Step 1: Pick one task in your business this week and try doing it with AI

Not your whole business. One task. The one that takes you the most time and brings you the least joy.

Writing proposals. Drafting follow-up emails. Summarizing meeting notes. Pulling numbers from invoices. Whatever yours is.

Step 2: Time yourself before and after

This is the part most people skip. Write down how long the task took you the old way. Then time yourself doing it with AI.

If it saved you 20 minutes, you just found 20 minutes a day. That’s almost two hours a week. Almost 100 hours a year.

That’s the gap that decides who wins your market over the next three years.

Step 3: When iOS 27 lands this fall, set your default on purpose

Open Settings. Find the new Extensions section under Apple Intelligence. Pick the AI you’ve been using all summer. Done.

That’s it. That’s the whole thing.

The Bottom Line

Apple is the cautionary tale. Not the villain.

A company with unlimited money, the best engineers on Earth, and the most loyal customers in the history of products still missed it. They announced AI. They couldn’t ship it. Eighteen months later, they’re handing the keys to Google and Anthropic.

That should land hard. Because if Apple isn’t safe, the question is not whether AI will reach your industry. It’s whether you’ll be the business that moved early or the one that’s quietly handing the keys to a faster competitor.

You don’t need to bet the company. You need to pick one task this week and try it. That’s the entire game right now.

Apple is going to be fine. They have iPhones to sell. You have a business. Start moving.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

I’ll wait.

Word of the Day: Digital Confession

A digital confession is anything you type into an AI tool that you didn’t intend for anyone else to read, but that could end up in front of a judge, a regulator, or an opposing attorney anyway.

Picture every conversation you’ve ever had with ChatGPT or Claude printed out and placed in a single binder. Now imagine that binder sitting on opposing counsel’s desk in your next deposition, in a regulator’s enforcement file, or in a prosecutor’s discovery production.

That binder exists. It’s just not on your desk.

What just happened in court

In February, Judge Jed Rakoff of the Southern District of New York ruled in United States v. Heppner that conversations with consumer AI tools are not protected by attorney-client privilege.

Bradley Heppner, a former CEO facing federal fraud charges, had used Claude to prepare 31 documents related to his legal defense. Defense strategy. Anticipated charges. What he might argue. He fed information his lawyers at Quinn Emanuel had told him directly into the chatbot. He later shared those documents with his counsel.

Judge Rakoff held that none of it was privileged. None of it was work product. The reasoning:

• Claude is not an attorney. There’s no professional relationship, no fiduciary duty, no bar discipline.

• Anthropic’s privacy policy reserves the right to use inputs for training and to disclose data to regulators. That alone defeats any reasonable expectation of confidentiality.

• Heppner used Claude on his own initiative, not at the direction of counsel. So no work product protection either.

• Sharing the AI documents with his lawyer afterward did not retroactively make them privileged.

Now the kicker that should keep every defense attorney up at night. Several of the major firms analyzing the ruling have flagged that by feeding what his lawyers told him into the AI, Heppner may have waived privilege over the underlying attorney-client communications themselves. He didn’t just lose protection on the AI conversation. He may have torched protection on the original legal advice that started everything.

That’s not a hypothetical for criminal defendants. That’s a hypothetical for any client who ever pastes “here’s what my lawyer said” into ChatGPT.

Your clients are doing this right now

Let me say what every attorney reading this already suspects.

Your clients are talking to AI about their cases. They’re typing injury details, accident facts, what they told their doctor, what they’re worried about. They’re asking ChatGPT to help them respond to demand letters. They’re pasting your engagement letter into Claude and asking what it means. They’re doing it before they call you. They’re doing it the night before depositions. They’re doing it because the interface feels calm and private and they don’t know any better.

They should know better. And as their lawyer, you’re now the one who has to tell them.

Texas attorney Virginia Hammerle put it bluntly to CNN: “In my firm, we’re treating it as: Anything that somebody’s typing into ChatGPT is something that could be discoverable.”

Or as Nils Gilman, a senior adviser at the Berggruen Institute, said in the same article: “ChatGPT is not your friend, is not your lawyer, is not your doctor, is not your spouse. Stop talking to them as if they are.”

The wrinkle nobody is talking about

Heppner is the headline. But there’s a second case that may matter even more for the average user.

In May 2025, a federal magistrate judge in The New York Times v. OpenAI ordered OpenAI to preserve every single ChatGPT conversation indefinitely, including ones users had already deleted. In November, the same judge ordered OpenAI to produce 20 million de-identified ChatGPT logs to the news plaintiffs.

This applies to ChatGPT Free, Plus, Pro, and Business (formerly Team). Even chats you thought were gone. Even Temporary Chats. Only ChatGPT Enterprise, Edu, and qualifying API customers with Zero Data Retention agreements were exempt.

So even if you opted out of training on your Plus account, OpenAI is currently sitting on a comprehensive archive of your conversations because a federal court told them they have to. That archive can be subpoenaed. That archive can be ordered produced.

This is not theoretical. It is happening right now.

Right now, today: the 60-second fix for your personal AI account

If you’re on a personal AI subscription, here’s exactly what to do in the next minute. This doesn’t fix everything (the NYT preservation order still applies to ChatGPT consumer plans), but it stops your future conversations from being used to train the next model. These instructions apply to ChatGPT Free, Go, Plus, and Pro, plus Claude Free, Pro, and Max. The toggle is in the same place on every consumer plan.

ChatGPT Free, Go, Plus, or Pro:

1. Go to

https://chatgpt.com

1. and sign in.

2. Click your profile icon (top right corner).

3. Click Settings.

4. Click Data Controls.

5. Find Improve the model for everyone. Toggle it OFF.

That’s it. Future chats will no longer be used to train OpenAI’s models. (Direct link to the help article: https://help.openai.com/en/articles/8983130)

Claude Free, Pro, or Max:

1. Go to https://claude.ai/settings/data-privacy-controls

2. Find Help improve Claude. Toggle it OFF.

Important context for Claude users: in October 2025, Anthropic flipped the default. If you clicked through the “Accept” pop-up quickly, you opted IN to a 5-year retention window with your data being used for training. If you opted out, you stayed at the standard 30-day retention. Many people clicked “Accept” without realizing what they agreed to. Go check your setting today. (Reference: https://www.anthropic.com/news/updates-to-our-consumer-terms)

Google Gemini:

1. Go to https://myactivity.google.com/product/gemini

2. Click Turn off next to Gemini Apps Activity.

3. Confirm.

(Reference: https://support.google.com/gemini/answer/13594961)

For all three: deleting old chats does not pull them out of any model that’s already been trained. It only stops your future conversations from being used. So flip the switch now and don’t wait.

The practical answer for attorneys: what plan should you actually use?

This is where it matters whether you’re paying $20/month or running an enterprise contract. The differences are massive and most attorneys have no idea.

Consumer plans (ChatGPT Free, Go, Plus, Pro / Claude Free, Pro, Max). This is the Heppner danger zone. By default, your conversations train the model. Conversations sit on company servers. ChatGPT Plus and Pro are also currently subject to the NYT preservation order. If you are using these for any client matter, you have a problem.

Team plans. ChatGPT Business (the plan formerly known as Team) and Claude Team. Significantly better. No training on data by default. Proper data handling agreements. Still recommended for most small and mid-sized firms as a baseline.

Enterprise plans (ChatGPT Enterprise, ChatGPT Edu, Claude Enterprise). Strong protections. No training. Admin controls. Audit logs. Available BAA for healthcare. Not subject to the NYT preservation order. This is what AmLaw 200 firms are buying.

Enterprise API with Zero Data Retention. The gold standard for cloud AI. The conversation is processed and immediately discarded. There is nothing stored to subpoena.

Self-hosted open-source LLMs (Llama, Mixtral, etc. running on your own servers). This is the most private option that exists. The conversation never leaves your network. There is no third-party vendor in the picture. Privacy by design. The catch: it’s expensive, technically demanding, and most firms under 50 attorneys can’t realistically pull it off.

But even on the safer plans, two things commonly trip up firms that think ZDR is the finish line.

Two things ZDR does not do

ZDR is not privilege

Zero Data Retention is a vendor-side commitment that no data is stored at rest on the AI provider’s servers. That is a real and meaningful protection. It is not the same thing as attorney-client privilege.

Privilege has specific elements: a communication between a client and an attorney, intended to be confidential, for the purpose of obtaining or providing legal advice. ZDR does not satisfy any of those elements on its own. An AI is not an attorney. A prospective website visitor is not yet a client. A casual chatbot conversation is rarely framed as obtaining legal advice.

For AI work to qualify as privileged or as attorney work product, the attorney has to be in the picture. The attorney has to direct the use, supervise the output, and incorporate it into the representation. ZDR can support that posture. It cannot create it.

Anything that still exists can be subpoenaed

ZDR works because nothing exists at the vendor to hand over. But your own internal logs, audit trails, chat transcripts, call recordings, CRM entries, and backups can still be discoverable, even when stored on a private firm server. Privilege is not automatic. It has specific elements that have to be met, and AI intake conversations between a chatbot and a prospective client often fail them.

That’s why it’s extremely important to have a written policy along the lines of: AI intake is for initial screening only; substantive discussions occur only with attorneys; intake records are retained for 14 days for the purpose of conflicts and follow-up; here is our purge schedule. That policy itself becomes the firm’s defensive posture in any future discovery dispute. It establishes what was confidential, what was intended for legal advice, and what was just lead-capture metadata.

The policy alone is not enough. IT has to actually execute it. Automated purges on schedule, no rogue backups sitting in a forgotten S3 bucket, no associate’s local copy on a personal laptop. A written policy that the firm doesn’t follow is worse than no policy at all in front of a judge.

The firms that get this wrong will spend years in privilege fights they didn’t know they were going to have. The firms that get it right will have answers ready.

The law firm intake problem nobody is talking about

If you run a firm, here’s the bigger exposure most managing partners haven’t thought through.

You almost certainly have an AI somewhere on your website. A chatbot in the corner. An AI receptionist answering after-hours calls. An intake bot collecting case details before a paralegal ever sees the lead.

If that system was built on top of consumer ChatGPT, or any AI vendor whose terms of service let them use the data for training or share it with third parties, then every prospective client conversation is potentially discoverable evidence sitting on someone else’s server.

The caller describes their situation. They share sensitive details. The AI asks follow-ups. All of that gets stored by a vendor whose terms of service let them share it with regulators. Or, in the case of consumer ChatGPT right now, by a vendor under a court order to preserve every conversation indefinitely.

That’s not a chatbot. That’s a liability sitting on your homepage.

The fix isn’t complicated, but it has to be intentional. AI systems handling intake conversations need to be built specifically with Zero Data Retention from the start. The conversation ends when the session ends. Nothing stored. Nothing to subpoena.

This is exactly why we’re building AttorneyInbound.ai at DigitalTreehouse for law firms specifically. It runs on AWS Bedrock with Zero Data Retention. Anthropic never sees the conversations. AWS doesn’t use them for training. Nothing gets stored in service logs. Caller information stays inside the firm’s controlled environment, end to end.

We didn’t bolt this on at the end. It’s the foundation. Because in 2026, an AI receptionist that doesn’t have ZDR baked in is a malpractice risk waiting to happen.

Even Sam Altman is worried about this

OpenAI’s CEO said on the Theo Von podcast in July 2025 that he was “very afraid” the government would use AI chat logs to surveil people.

His exact words: “I think we really have to defend rights to privacy. I don’t think those are absolute. I’m totally willing to compromise some privacy for collective safety, but history is that the government takes that way too far, and I’m really nervous about that.”

That’s the CEO of ChatGPT saying he’s nervous about how ChatGPT conversations could be used. Take that however you want.

Some serious thinkers are pushing for new laws. Nils Gilman argued in a New York Times op-ed for legal privilege protections for AI conversations, similar to what exists for therapists and lawyers. The logic: if millions of people use AI the way they once used a therapist, the law should catch up.

The courts aren’t waiting for the law to catch up, and neither should you.

What you should actually do, in order

For yourself, today:

1. Flip the training opt-out toggle on every personal AI account you have. Steps above. Takes 60 seconds total.

2. Stop using consumer AI tools for active legal matters. Period.

3. If you’re going to keep using AI for legal work, upgrade your firm to an Enterprise or Team plan with proper data handling agreements.

For your firm:

1. Audit every AI tool deployed on your firm’s website, phones, or intake systems. Ask the vendor one question in writing: “Do you have Zero Data Retention?” If they don’t know what that means, replace the system.

2. Update your engagement letters with a clause warning clients about AI use and the privilege risk.

3. Tell every staff member: no client information in any consumer AI account, ever. Not personal ChatGPT, not personal Claude, not personal Gemini. Not even for “just a quick summary.”

4. If you’re ready to deploy AI safely (intake, scheduling, document review, client communication), do it on infrastructure designed for legal work from day one.

For your clients:

If they ask whether they should use ChatGPT or Claude to think through their case, the answer is no.

The bottom line

The rules just changed. Quietly. Without a press release.

Your AI conversations are now legally equivalent to a credit card swipe or a phone call record. They exist. They can be demanded. They can be used against you and against your clients.

The technology moved faster than the law. The courts aren’t waiting for the law to catch up, and neither should you.

If you took 60 seconds to flip those toggles, your time reading this newsletter just paid for itself for the rest of the year.

If you forwarded this to one other attorney, you might have just saved their license.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. If you’re a law firm or professional services business and want to talk about building AI receptionists, intake bots, or workflows the right way with Zero Data Retention from day one, reply to this email and ask about AttorneyInbound.ai.

This week, Perplexity rolled out something called Workflows. More than 50 of them, with new ones being added regularly. Each one is a pre-built task tuned by an expert. Market research. Document review. Website builder. SEO keyword research.

You click one. You answer a couple of questions. It does the work.

Translation: Perplexity just turned itself into a marketing analyst, a paralegal, a web developer, and an SEO specialist. All in the same browser tab.

The catch? Workflows live inside a product called Perplexity Computer, and Perplexity Computer sits behind their top-tier Max plan at $200 a month. Not pocket change. But there’s also a free version of this idea that gets you most of the way there, and I’ll show you both.

Word of the Day: Workflow

A Workflow is a pre-built, repeatable AI task that an expert has already designed for you, so all you have to do is click it and answer a couple of questions.

Think of it like a Keurig coffee pod. Someone else figured out the right beans, the right grind, the right water temperature. You don’t need to know any of that. You just pop in the pod and press the button.

For your business, this matters because the hardest part of using AI has never been the AI. It’s been figuring out what to ask it and how to ask it. Workflows skip that step. The expert is baked in.

What Perplexity actually shipped

Perplexity has a product called Perplexity Computer. It is exactly what it sounds like, an AI that can do research and complete tasks like a coworker with a laptop.

This week, they added Workflows on top of it. Here are a few of the ones they highlighted:

• Market Research. Analyzes trends, competitors, and opportunities in any market.

• Final Pass. Reviews any document to a high standard before it goes out the door.

• Website Builder. Builds a complete website from a description, ready to launch.

• SEO Keyword Research. Finds high-impact keywords and content gaps for your site.

That’s four. There are more than 50.

So when you see Workflows in the marketing, know that the full experience sits at $200 a month. Pro users can pay-as-you-go for Computer credits if they want to dip their toes in. Free users cannot.

That’s a real number, and you should weigh it against what one of those workflows would actually save you. If a Final Pass review catches a typo on a $50,000 proposal, the math is easy. If you’re using it to plan a birthday party, the math is not.

The free version that gets you most of the way

Here’s the part I love. The Rundown newsletter ran a tutorial yesterday for a free workaround that’s almost as good for one specific job: stress testing a business idea.

It uses Perplexity’s Deep Research mode, which works on the free plan with 3 to 5 runs per day depending on the source.

The recipe is dead simple:

1. Go to perplexity.ai and sign in (free account is fine).

2. Switch to Deep Research mode in the chat.

3. Paste a prompt that says: “Stress test this business idea: [your idea]. Research the market, the competition, the realistic costs, and the biggest risks. Give me a slide deck of findings.”

4. Hit run and walk away for 5 to 6 minutes.

5. Read the deck.

I tried this on a side project I’ve been kicking around. The deck wasn’t perfect. But it surfaced two competitors I hadn’t heard of and one risk I hadn’t considered. That’s worth a Saturday morning of your life.

Save the prompt somewhere you’ll actually use it again. Run one idea every weekend. You’ll burn through a year of half-evaluated ideas in a month.

Why this matters more than it seems

Step back for a second. Here’s what’s actually happening.

Six months ago, AI was a chat window. You typed, it answered, you copied the answer somewhere useful. You were doing 80% of the work.

Now AI is a coworker. With workflows. With its own computer. With pre-built specialists you can call on.

I caught a related move buried in the FutureTools newsletter yesterday: Amazon launched Amazon Quick, a $20-a-month desktop AI assistant. Microsoft and OpenAI reset their alliance. Google added Gemini to cars. Mistral debuted remote cloud agents called Vibe.

Everybody is building the same thing. AI that works on your behalf, not just AI that talks to you.

The honest part

I want to be straight with you. The marketing for these tools is always ahead of the reality.

Workflows are real and useful. They are not magic. The Market Research workflow will give you a competitor list that’s about 80% right and 20% hallucinated. The Website Builder will build something fast and ugly. The Final Pass document review will catch real issues, and also flag things that aren’t actually issues.

So the rule still applies. AI gives you a draft. You give it judgment. The draft just used to take you four hours and now takes four minutes.

That’s the whole shift. You’re not being replaced by AI. You’re being upgraded from the person who does the work to the person who reviews and approves it.

The Bottom Line

Here’s what this means for you.

The job description for a small business owner is changing in real time. You used to need to know how to do the work, or you needed to hire someone who did. Now you need to know how to direct the work and how to spot when it’s wrong.

Don’t pay $200 a month until you know you’ll use it. Test the free version on something real. If you find yourself running it five times a week, then upgrade. If you don’t, you just got a year of business research for zero dollars.

Is the work going to be perfect? No. Will it save you hours? Almost certainly. Will it change what you can build alone, with no team, no budget, no permission? Yes. That’s the part nobody is shouting from the rooftops, and that’s the part you don’t want to miss.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

The headlines are calling it a major upgrade. And it is.

But here’s what most of the headlines aren’t telling you. Claude has been doing this for months. And it does it better.

Word of the Day: Native Export

Native export is when an AI tool creates a real, ready-to-send file (like a .docx, .xlsx, or .pdf) directly inside the chat, instead of just giving you text on the screen.

Think of it like the difference between a chef reading you a recipe and a chef handing you the finished dish. Both technically deliver the meal. But one of them saves you an hour in the kitchen.

For your business, this matters because most of your work ends as a file. A proposal. A spreadsheet. A report. A contract. Native export means the AI does the whole job, not just the writing part.

Claude has been the gold standard for this

If you have a paid Claude account, you can already do all of this and more. Here’s what works today.

Claude builds real Word documents with proper headings, tables, and page breaks. It builds Excel files with working formulas, not pictures of formulas. It builds PDFs that look professional out of the box. It builds PowerPoint decks with multiple slides. One note, sometimes the layouts are slightly off on Claude PDFs so I always default to Word docs first, make my final manual edits, then convert to PDFs (File→Print to PDF).

But the part that actually matters for your business is branding. You can upload your logo, your brand colors, your fonts, and your style guide once, and Claude will use those settings on every file it creates after that. Your proposals come out looking like your proposals. Not a generic template.

I’ve been having Claude build client proposals, internal SOPs, project trackers, and pitch decks this way for months. The files come out branded, formatted, and ready to send.

How to set up Claude for branded files

Here’s how to get this running in about ten minutes.

1. Go to claude.ai and sign in. You’ll need a paid plan (Pro is $20/month) to create files reliably.

2. Click your name in the bottom left, then Settings, then look for Projects in the sidebar.

3. Create a new Project. Name it something like “My Company Files.”

4. Inside the Project, look for Project knowledge. Upload your logo (PNG with transparent background works best), your brand color codes (hex codes like #1A2B3C), your font names, and any style guide you have.

5. In the Project’s custom instructions, write something simple like: “When creating any file, use our brand colors and logo. Headings should be in our brand font. Footer should include our company name and website.”

6. From now on, when you start a chat inside that Project and ask Claude to make a file, it will use your brand by default.

That’s it. Your AI just learned your brand.

Gemini is the right starting point if you’ve never tried this before and don’t want to pay anything. It’s a fine way to test whether AI-generated files will actually save you time.

Claude is the right tool if you send branded documents to clients and care about looking professional.

The honest part

Neither tool is perfect.

Gemini’s outputs are functional but plain. The Word docs open cleanly, but you’re still going to need to add your branding manually each time.

Claude is excellent at branded files, but it’s not free. You’ll need at least a Pro plan to get reliable file generation. Worth it if you make documents for a living, probably overkill if you don’t. (Side note: the BEST branded PDF generator I’ve seen to date is Manus.ai.)

Both tools occasionally get things wrong. I asked Claude to build a 12-month cash flow projection last week and one of the formulas referenced the wrong cell. Always check the numbers before you send anything to a client.

Why this matters more than it looks

This isn’t just a Google feature drop. It’s part of a much bigger shift that happened this week.

Alphabet, Amazon, Meta, and Microsoft all reported earnings on the same day. Every single one of them posted huge AI-driven growth. Microsoft 365 Copilot now has over 20 million paid seats, up from 15 million just three months ago. AWS grew 28%, its fastest pace in 15 quarters. Google Cloud grew 63%.

Translation: AI isn’t an experiment anymore. The biggest companies in the world are pouring real money into it because real customers are paying for it. That money is what’s funding features like Gemini’s native export.

Cloudflare also announced this week that AI agents can now open accounts, buy domains, and deploy websites without a human clicking through dashboards. Stripe announced an agent wallet so AI can pay for things on your behalf. The whole infrastructure is being rebuilt so AI can do the actual work.

Native export is one small piece. But it’s the piece you can use today.

The Bottom Line

When a feature shows up in Gemini for free, that means it’s about to be table stakes. Every AI tool will have it within 60 days.

But the real lesson is this. If you’ve been waiting for AI to feel ready, it’s been ready in Claude for a while now. Branded proposals, formatted spreadsheets, real PowerPoint decks. All of it has been sitting there.

Gemini just made the floor higher. Claude is still the ceiling.

If you’re new to all this, go try Gemini today and see what AI files even look like. If you’re serious about using AI for client-facing work, set up a Claude Project this weekend and upload your brand.

Either way, stop pasting AI text into Word.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

I want to start with a confession.

For years I rolled my eyes at the “robots are coming” headlines. I’d been hearing them since I was in the Navy. The robots were always five years away. Then ten. Then “soon.”

Then recently I opened my inbox and saw this: a humanoid robot built by a Chinese company called Honor finished the Beijing half-marathon in just over 50 minutes. The human world record is 57 minutes.

The robot won. By seven minutes. A year ago, the winning robot took two and a half hours to finish that same race.

Fair. I get it now.

Word of the Day: Embodied AI

Embodied AI is artificial intelligence that lives inside a physical body and learns by interacting with the real world.

Think of it like the difference between reading a cookbook and actually cooking dinner. ChatGPT read every cookbook ever written. Embodied AI is the robot in your kitchen that has burned the rice, dropped the egg, and finally figured out how to flip the pancake.

For your business, this matters because the AI you’ve been using on a screen for the last two years is about to step out of the screen and start picking things up, walking around, and doing physical work. That changes the math on labor in a way email automation never could.

What’s Actually Happening Right Now

The half-marathon was not a one-off. Look at what’s been landing recently:

Sony’s table tennis robot beat elite human players. Project Ace won 3 of 5 matches against pros and outscored them 16 to 8 on direct serves. Ping pong is fast, unpredictable, and requires reading another human’s body language. The robot did it anyway.

Tesla expanded its robotaxi service to Houston and Dallas. Small geofenced zones for now, but Phoenix and Miami are coming this year. Waymo is already running fully driverless in both cities.

Agility Robotics’ Digit deadlifted 65 pounds. Not a stunt. A test of full-body coordination they trained through thousands of simulated tries before doing it for real.

A startup called Reframe Systems is building modular homes inside robot-run microfactories. Finished houses are already occupied in Massachusetts.

OPTIMUS: Where Tesla Actually Stands

Tesla’s robot is called Optimus. Tesla calls it “Tesla Bot” sometimes too, but Optimus is the official name. It stands 5 feet 8 inches, weighs 125 pounds, and runs on the same neural network technology as Tesla’s self-driving cars.

Here’s where it gets interesting. On the Q1 2026 earnings call earlier this month, Elon Musk had a rare moment of honesty. He admitted that of the 10,000 Optimus robots he promised would be built in 2025, basically none did “useful work.” Zero. Not “fewer than expected.” Zero.

The new plan is genuinely ambitious. Tesla is shutting down Model S and Model X production (after 14 years) and converting that entire Fremont factory line to build Optimus instead. Production starts in late July or August. A second Optimus factory is going up at Giga Texas with a long-term capacity target of 10 million robots per year.

But Musk himself said the production rate this year is, in his words, “literally impossible to predict.” Optimus has 10,000 unique parts. The current Gen 3 prototype has 50 actuators in its hands alone. The robot is real and walking around. Whether Tesla can actually mass-produce it on the timeline they’re claiming is a different question.

For context, Tesla’s own Cybertruck took over a year to ramp from first production to meaningful volume, and that was a vehicle, leveraging Tesla’s existing automotive expertise.

The target price when scale finally arrives: $20,000 to $30,000 per robot. Right now, current manufacturing cost is estimated at $50,000 to $100,000 per unit, with initial commercial pricing likely in the $100,000 to $150,000 range.

Translation: Tesla’s robot is real. It’s behind schedule. And it still might end up the most important product Tesla ever makes. All three things are true at once.

The Plot Twist Nobody Saw Coming: A Chinese Robot Ban

Here’s the wrinkle. While American companies are racing to build humanoid robots, Congress is moving to ban the Chinese ones.

Senators Tom Cotton (R-AR) and Chuck Schumer (D-NY) introduced the American Security Robotics Act in March. The bill would ban federal agencies from buying or operating humanoid robots and other ground robots made by Chinese, Russian, Iranian, or North Korean companies. A companion bill is moving in the House.

The reasoning: humanoid robots have cameras, microphones, sensors, and network connections. A robot mopping the floor at a federal facility could, in theory, be a surveillance device.

The catch is that American robot companies still depend heavily on Chinese parts. Brushless motors, rare-earth magnets, precision actuators, advanced sensors. China installed 300,000 industrial robots in 2024 alone, almost ten times what the U.S. installed. Morgan Stanley estimated that building Tesla’s Optimus Gen 2 supply chain without China would cost three times as much.

Why this matters for you: if you’re thinking about robot-powered automation in your business in the next 2-3 years, the country-of-origin question is about to get politically loaded. Federal procurement bans came first, broader market restrictions followed. Plan accordingly.

The Job Displacement Question (The Honest Version)

This is the part nobody wants to write about…

McKinsey’s projection is that automation, including humanoid robots and AI together, could displace 400 to 800 million jobs worldwide by 2030. They estimate up to 375 million workers will need to switch occupations entirely. That’s roughly 14 percent of the global workforce changing what they do for a living in less than five years.

Here is the rough timeline as I see it from the data:

2026 to 2027: Warehousing and factories. Repetitive picking, packing, palletizing, box-moving. Amazon, BMW, Tesla, BYD are already doing this. If your business depends on warehouse labor, your competitors are getting bids on robots right now. Goldman Sachs estimates humanoid robots will fill 4 percent of the U.S. manufacturing labor shortage by 2030.

2027 to 2029: Logistics and last-mile. Robotaxis expand from geofenced zones to wider service areas. Cargo aircraft fly with no human pilot (Reliable Robotics just raised $160M to get FAA approval). Sidewalk delivery robots multiply.

2029 to 2032: Service work. Fast food, hotel cleaning, basic retail stocking. Goldman projects warehouse automation will spread into hospitality and elder care, where labor shortages are already brutal.

2032 to 2035: Skilled trades and home services. This is the slower one. Plumbing, electrical, complex repairs. Robots still struggle with unpredictable physical environments. Your house is messier than a factory floor, and that buys human tradespeople a lot of time.

Unreplacable: Anything requiring trust, judgment, or human relationship. Sales. Senior care that involves emotional connection. Coaching. Therapy. Negotiation. Custom craftsmanship. Anything where your customer is paying for you, not just the output.

I want to be honest about the limitations too. The Honor robot ran a flat course on a controlled route. Sony’s table tennis robot has never had to handle a player who showed up grumpy and changed strategy halfway through. Battery life is still a real problem. Regulators have not figured out liability when a robot drops a 200-pound box on someone’s foot. And Tesla just admitted its previous production targets were fantasy.

This will not be smooth. But the direction is clear.

Meanwhile, In Non-Robot AI News

A few stories landed today that are too big to skip, even on a robot-focused issue.

Musk vs Altman, the $134B trial just kicked off. Day 1 of Elon Musk’s lawsuit against OpenAI began this week, with Musk himself on the witness stand. Musk is asking for $134 billion in damages, the removal of Sam Altman and Greg Brockman from OpenAI’s board, and a forced unwind of OpenAI’s for-profit conversion. His argument: that Altman “stole a charity.” OpenAI’s argument: that Musk is throwing a tantrum because he didn’t get his way after leaving. Witness list includes Microsoft CEO Satya Nadella, Brockman, and former OpenAI executives. Four weeks of testimony ahead. Hundreds of pages of private emails are about to spill into the public record. If you’ve been wondering whether OpenAI’s IPO is still on track, this trial is the answer.

Google signed a classified Pentagon AI deal. Google inked a contract giving the Pentagon access to its AI models for “any lawful government purpose.” 600+ Google employees signed an open letter to CEO Sundar Pichai asking him not to do exactly this. Google’s old “no weapons” pledge was quietly removed from its AI principles in 2025. OpenAI and xAI signed similar Pentagon deals last month. Anthropic is in court fighting after being blacklisted for refusing to drop its safety guardrails. Translation: every major AI lab is now an active military contractor, except the one that won’t drop its rules.

OpenAI is reportedly building a phone. OpenAI partnered with MediaTek, Qualcomm, and Luxshare to build a ChatGPT-powered cell phone, likely with its own operating system, where AI agents could potentially replace the apps you use today. This joins a rumored ChatGPT smart speaker, smart glasses, and smart lamp. If any of this ships, the iPhone gets its first real challenger in a decade.

Microsoft Outlook just got “Agent Mode.” Copilot in Outlook can now run your inbox and calendar more efficiently. Tell it what you need done, and it gets started, keeping you in the loop. Available in Microsoft’s Frontier program now. If you live in Outlook, this is worth testing.

Anthropic added Claude connectors for creative apps. Claude can now connect directly to Blender, Adobe Creative Cloud, Autodesk Fusion, SketchUp, and more. If you do creative work, this matters. The AI is no longer stuck in a chat window.

The Bottom Line

Here’s what this means for you.

The robots are not coming. They are here. A robot just ran a half-marathon and won. Another beat elite ping pong players. Tesla is about to retool a factory that built Model S and Model X for 14 years, just to make robots. Congress is already trying to ban the foreign ones.

The good news is “embodied AI” still has a long climb before it can do what you do. The bad news is that climb is happening faster than any of us predicted. A year ago, the marathon robot took 2.5 hours. Now, 50 minutes.

Don’t time the market by trying to guess exactly when robots will replace any specific job. Time the market by making sure the work you do can’t be replaced by a robot in the first place. Lean into the parts of your business that require trust, judgment, taste, and a real human showing up.

The robots are getting better at moving fast. You should be getting better at the things they can’t fake.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

Yesterday Claude broke. Not in a dramatic way. In a slow, sneaky way that I couldn’t quite put my finger on.

Responses felt shorter. Tasks that used to work in one shot needed two or three. I thought I was losing my mind. Turns out I wasn’t. Anthropic published a post-mortem yesterday admitting three different bugs had quietly made Claude dumber for weeks.

And while that was happening, OpenAI dropped a bomb on the whole industry.

Word of the Day: MODEL

An AI model is the “brain” behind a chatbot. It’s the trained software that does the actual thinking, writing, and reasoning.

Think of it like the engine in your car. ChatGPT is the car. GPT-5.5 is the new engine they just dropped under the hood. Claude is a different car. Its engine is called Opus 4.7. Gemini is another car with its own engine.

For your business, this matters because the chatbot name stays the same, but the engine changes. When ChatGPT “gets smarter” next month, it’s because OpenAI swapped in a better model. When Claude has a bad week, it’s usually the model or the plumbing around the model that’s at fault, not the app. Knowing which engine you’re driving helps you pick the right tool for the right job.

The big news: GPT-5.5 “Spud” is here

OpenAI launched GPT-5.5 yesterday. Sam Altman called it “a new class of intelligence.” The team internally nicknamed it Spud. (Don’t ask. OpenAI has always been weird with codenames.)

Here’s the part that matters for you.

For the last several months, Claude has been the quiet favorite among people who actually build things with AI. Developers loved it. Writers loved it. I loved it. Anthropic’s reputation was on a rocket ship.

Then this week happened.

GPT-5.5 scored 82.7% on a major coding test, beating Claude’s top model. It’s about 20% faster than the previous version. And OpenAI priced it at half the cost of the top competitor.

If you use ChatGPT Plus, Pro, or Business, you already have access to it. You don’t have to do anything. The engine got swapped while you were sleeping.

Meanwhile, at Anthropic, a rough week

Three separate things went sideways for Claude this week, and your business owner friends are going to ask you about at least one of them. Here’s the short version so you sound smart at the coffee shop.

1. Claude got quietly worse for a few weeks

Anthropic admitted that Claude Code (their developer tool) had three bugs running at the same time. One reduced how hard it thought. One cleared its memory mid-task. One told it to give shorter answers.

If you felt Claude slipping recently, you weren’t imagining it. Anthropic is resetting usage limits for paid users as an apology. If you’re a paid Claude user, check your account.

2. Their most dangerous model got leaked

Anthropic has a model called Mythos that’s so powerful at cybersecurity they decided not to release it publicly. It was only supposed to go to a handful of partners like Google, Nvidia, and Apple.

Someone in a private Discord got access through a third-party contractor. Anthropic says the leak looks contained. But the headline writes itself: the AI lab that built its whole brand on safety just had its most dangerous model walk out the door.

3. They launched Claude Managed Agents (this one is actually great)

Buried under the bad news was a real feature worth your attention. Claude Managed Agents now have memory that persists across sessions. Meaning an agent you set up today can remember what it did for you last week.

Quick note before you get excited. This is not the same thing as Claude Projects. Projects is where you chat with Claude and it remembers your business. Managed Agents are different. They run on their own, on a schedule, doing work while you sleep.

Until now, those agents started from zero every time they woke up. You had to re-explain the task, the tone, the customer, every single run. No more. The memory is stored as simple text files you can edit yourself.

For a small business owner, this is a big deal. Projects gave you an AI that knows your business when you chat with it. Managed Agents now give you an AI employee that actually remembers what it did yesterday.

6 things you could actually have a Claude Managed Agent do

To make this real, here’s what a Claude Managed Agent could be doing for your business while you’re at your kid’s soccer game.

1. Scan your inbox every morning at 6am and send you a one-page summary of what needs a reply, what can wait, and what to delete.

2. Summarize every Zoom call from the day, pull out the action items, and assign them to the right person in your project management tool.

3. Log client intake form submissions into your CRM, draft a personalized welcome email, and schedule a follow-up reminder seven days out.

4. Monitor your business email for invoices, extract the amount and due date, and log them in QuickBooks with the right category.

5. Read your weekly P&L and send you a plain-English recap of where the money went and what changed from last week.

6. Draft your weekly LinkedIn post based on what you actually did that week, pulled from your calendar, Slack messages, and completed projects.

The catch. Setting one of these up takes work. You can’t just tell Claude “watch my inbox” and walk away. Someone has to build the agent, connect the tools, and test it. That’s where an agency like ours earns its keep. But the point is, these are no longer hypothetical. This is what Managed Agents with memory can actually do, starting this week.

Claude also now connects to 200+ apps

One more Claude update worth noting. Claude can now connect directly to 200+ apps and chain actions together in one conversation. Spotify just joined the list for music and podcast recommendations.

Here’s what a real example looks like:

“Claude, check my Gmail for any invoices from this week, log them in my QuickBooks, and Slack my bookkeeper a summary.”

That’s three apps, one sentence, one Claude. No Zapier in the middle.

This is the direction all the big AI companies are moving. The chatbot is becoming the operating system. Your job is going to be describing what you want, not clicking through five different tabs.

What this all means (the honest version)

Let me tell you the part nobody in the AI newsletter world will say out loud.

The “leader” flips every three to six months. Last summer it was OpenAI. Last fall it was Anthropic. Right now it’s OpenAI again. By August it might be Google.

You should not pick a tool based on who’s winning this week. You should pick based on what fits your workflow, what your team will actually use, and what’s in your budget. A chatbot that’s the number two model but you use every day is worth infinitely more than the number one model sitting in a browser tab you never open.

Here’s what I’d actually do right now, in order:

1. If you pay for ChatGPT, try Spud on a real task you did yesterday. Compare the output to what you got before. See if you notice.

2. If you pay for Claude, check if your usage limits got reset. Then try the new connectors on something boring like summarizing your inbox.

3. If you don’t pay for either, keep using the free tiers. The gap between free and paid has never mattered less than it does this week. Both free tiers are excellent.

4. If you’re a builder (even a Vibe Coder like me), experiment with Claude Managed Agents and memory. This is the quiet story of the week.

The Bottom Line

This week was a lesson in how fast the ground moves.

I had a moment yesterday where I was frustrated with Claude and seriously considered switching everything over to ChatGPT. A day later, I’m glad I didn’t. Because next week might flip again.

The trap is thinking you have to pick a winner. You don’t. Pick a workflow, then pick the model that makes that workflow actually work. When a better model comes along, swap it in. The skill isn’t picking the best AI. The skill is knowing how to use whatever AI is in front of you.

Spud won this round. Claude will probably win the next one. Gemini is lurking. Someone in China will release something next Tuesday. That’s the game now.

Is it confusing? Yes. Is it exciting? Also yes. The smart move isn’t to freeze. The smart move is to keep your hand on the wheel and your foot on the gas.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

A client hands me a brand guide. Forty-two pages. Color codes, font stacks, logo clearance rules, approved photography styles, tone of voice. Someone spent real money on it.

Six months later, that same client’s marketing team is sending out proposals in the wrong shade of blue, using a stock photo the brand guide explicitly says not to use, and a headline font that’s a close-but-wrong cousin of the official typeface.

Nobody broke the rules on purpose. They were just working fast, using whatever tools they had, trying to get stuff done.

This is not a discipline problem. It’s a systems problem. And AI is about to solve it in a way that nothing has before.

Word of the Day: Design Tokens

Design tokens are the exact, locked-in values that define your brand’s visual identity: the specific hex code for your primary blue, the exact pixel size of your heading font, the precise spacing between your logo and any other element on a page.

Think of them like a recipe’s ingredient list with measurements. “A pinch of salt” is vague. “3.5 grams of kosher salt” is a token. One leaves room for interpretation. The other doesn’t.

Why this matters right now: design tokens are the technical backbone behind every AI-powered brand enforcement system we’re about to discuss. When AI tools “know” your brand, tokens are how they know it.

How It Used to Work

Before AI, maintaining brand standards across all your creative output came down to four things, and none of them were great.

The brand guide PDF. Beautiful to look at. Nobody reads it after day one. It lives in a shared drive somewhere and gets consulted when there’s a dispute, which means after the mistake has already been made.

Templates. You build a set of approved layouts in PowerPoint, InDesign, or Canva. This works fine until someone needs something that doesn’t fit a template. Then they freestyle it, and the brand police cry.

A gatekeeper. Usually one overwhelmed designer who has to review everything before it goes out. That designer is the bottleneck for the entire organization. They spend 40% of their time checking that the red is exact instead of doing actual creative work.

Tribal knowledge. “Just ask Sarah. She knows the brand.” What happens when Sarah leaves? You find out on a Friday afternoon when a proposal goes to a major client with the old logo.

Here’s the brutal honest truth about all four of these systems: they’re all human-dependent, and humans are inconsistent under pressure. The more pressure, the more brand drift.

What AI Changed

Two things happened that changed the game.

First, AI got good enough to generate creative assets from text prompts. Now instead of hunting for a stock photo, you can describe what you want and generate something original. That’s powerful but dangerous for brand consistency, because now you have unlimited content creation ability with zero built-in guardrails.

Second, the tools got smart enough to hold the guardrails themselves. This is the part most people haven’t caught up with yet.

There are now multiple ways to bake your brand standards directly into AI tools so that they become the floor, not the ceiling, of every piece of creative your team produces. Let me walk through all of them.

Option 1: Figma Plus AI Agents (Best for Product and Digital Work)

Figma has been the professional designer’s tool of choice for years for UI and digital work. Starting this month, AI agents can design directly on the Figma canvas. Instead of generating design suggestions from outside the workflow, they can now interact with real design files, components, and design systems. DEV Community

Here’s why this matters for brand standards specifically.

When AI agents work inside Figma, they access existing components, variables like color palettes and spacing rules, and the design system itself. Designs automatically follow the standards defined in the design system. The AI isn’t guessing what your brand looks like. It’s reading your locked-in tokens and building from them.

The most important new concept here is called Skills. Skills are instruction sets written in Markdown that define how AI agents should behave while working in Figma. They provide guidance on which steps to follow when generating designs, how to apply design system rules, and what spacing conventions to use. This gives agents access to the intent behind a design system, not just the assets themselves.

Think of it this way. Your brand guide tells an AI not just “use Pantone 485 red” but “here’s when to use it, how much of it, and what it should never be placed next to.” The AI now understands your brand the way a senior designer does, not just a new hire following a checklist.

There’s even a self-healing capability. When an AI agent generates a design, it can take a screenshot of the result, compare it with the expected output, and automatically adjust the design if something doesn’t match. Because the AI is working with real components and variables rather than static images, these adjustments interact with the underlying design structure.

Who this is for: Designers, product teams, agencies doing website and app work. If you’re building anything digital and your clients have established design systems, Figma’s agent integration is the most technically rigorous brand enforcement available today.

Cost: Figma Professional is $15/seat/month. The agent features are in active rollout.

What it doesn’t do well: Print, ads, video, or social content. Figma is a UI and prototype tool first. It’s not where your magazine layouts or ad creative lives.

Option 2: Adobe Firefly Custom Models (Best for Images and Campaigns)

This is the one that should genuinely excite every creative professional reading this.

Here’s the problem Adobe is solving: every marketing team has the same production problem. They can generate images. They cannot reliably generate images that look like their brand without significant rework, and they cannot produce that volume at campaign pace without accumulating style drift across dozens of contributors and tools. Shashi

Adobe’s answer is Firefly Custom Models. Leading enterprises use Firefly Custom Models to generate images aligned with campaign styles, global brand guidelines, and character themes. You train a model on your own brand assets. From that point, every generation starts from your baseline, not from a generic corpus. Adobe

This is different from a brand guide. You’re not giving the AI instructions about your brand. You’re training the AI on your actual approved creative work. It learns what your brand looks and feels like from the inside.

Teams can then deliver banner ads, social media posts, paid media ads, webpage hero images, packaging, avatars, and more with ease. Distributed teams can create images for social media and advertisements that reflect local cultures while staying true to the overall brand.

On the privacy question that every brand manager will ask: Adobe states that customer training assets are not used to improve the shared Firefly models. Content created with Custom Models remains private by default. That’s the commitment in writing. Enterprise buyers should verify what’s contractually enforceable on their specific plan.

Also important: Adobe Firefly is still the only model that is fully indemnified by Adobe for commercial use because it was trained 100% on Adobe Stock. If you’re doing a global campaign for a client that is strict about copyright, the native Firefly model is your safest choice. Feisworld

Who this is for: Creative agencies, in-house marketing teams, any designer producing high volumes of campaign assets across multiple channels. If brand drift across a large team is your problem, this is the enterprise-grade solution.

Cost: Firefly Custom Models require an enterprise license. Standard Firefly features are bundled with Creative Cloud All Apps ($54.99/month). Firefly Services API is billed separately, roughly $0.02 per image at pay-as-you-go, with lower rates on committed use plans.

Option 3: Adobe Firefly AI Assistant (Coming Soon, and It’s the Big One)

This is the one that changes everything, and it’s not fully released yet.

Adobe just launched Firefly AI Assistant, a conversational agent that orchestrates tasks across Photoshop, Premiere Pro, Lightroom, and Illustrator using natural language commands. Instead of switching between applications and navigating menus, you can tell the assistant to resize images for social media, color-grade footage to match brand guidelines, or generate logo variations. It coordinates the work across whatever Adobe tools the task requires. shellypalmer

Here’s the detail that separates this from every other AI assistant you’ve seen: it maintains context across sessions, remembering project parameters, brand guidelines, and previous decisions rather than starting from zero each time.

So when you say “create a social ad in our brand style,” the assistant already knows your brand style. It doesn’t ask. It just does it correctly.

Adobe confirmed the assistant will work with third-party AI models, including Anthropic’s Claude, alongside Adobe’s own Firefly models and partners like Google, OpenAI, Runway, Luma AI, and ElevenLabs.

This is POWERFUL. You’re getting an AI creative director that never sleeps, never forgets the brand guide, and can execute across every Adobe tool simultaneously. A human creative director costs $150,000 a year. This will cost what you’re already paying for Creative Cloud.

When does it arrive: Public beta coming in the next few weeks as of this writing.

Option 4: Canva Brand Kit (Best for Non-Designers on Your Team)

Let me be honest about what Canva is and isn’t here.

Canva’s Brand Kit 2.0 acts as a centralized hub for logos, colors, fonts, and brand guidelines. It supports multiple brand kits and includes AI-powered consistency checks across all designs. VersusTools

That’s solid. For small businesses, marketing coordinators, and teams where the content is social posts and presentation decks, Canva’s Brand Kit is genuinely useful. Upload your brand colors, fonts, and logos once. Every template in your account automatically reflects them. Share the account, and every person on your team is working from the same approved palette.

The limitation is real though. Canva’s component system provides far less granular brand enforcement than Figma at the professional tier. Canva’s Brand Kit stores colors and fonts but cannot enforce layout rules or component structures. It keeps the guardrails on the easy stuff. It doesn’t prevent someone from making a layout decision that violates your spacing rules or using a photo that doesn’t fit the brand’s visual style.

Solution: Canva’s on-brand design generation inside Claude directly addresses this key pain point by allowing users to apply their Brand Kit at the moment of creation, ensuring that presentations, pitch decks, and campaign materials reflect approved colors, fonts, and tone from the outset. Futurum Group

That last piece is actually significant. If your team is using Claude for content generation and Canva for layouts, those two tools now talk to each other with your brand baked in.

Cost: Canva Pro at $15/month per user includes Brand Kit. Teams plan starts at $10/seat/month billed annually with a three-seat minimum.

Best for: Non-designer teammates who need to produce on-brand social content, presentations, and light marketing materials without calling the design department every time.

Is Adobe Obsolete? Let’s Be Honest.

Figma’s component system provides far more granular brand enforcement. For teams that need design control, Figma occupies a different tier entirely. For product designers building digital interfaces, Figma has already won. Adobe knows this.

But for print, video, campaign photography, packaging, magazine layouts, and the full production pipeline that a graphic designer lives in? Adobe isn’t close to obsolete.

Packaging, large-format printing, publication design, pre-press workflows, this is Adobe territory. Layer masks, adjustment layers, blend modes, Photoshop still has no Canva equivalent. Redhub

The more accurate framing: the market has split into two lanes. Figma owns digital product design. Adobe owns professional creative production. They’re increasingly competing in the middle lane of marketing and brand assets, which is where both are investing their AI development.

For a graphic designer who wants to stop checking pixels and start spending more time on ideas, the answer right now is blended at Adobe Firefly Custom Models for image generation, Figma for any digital/UI work, and the Firefly AI Assistant the moment it hits public beta. That combination gives you AI that starts from your brand, not from a blank slate.

The Practical Path Forward for Designers

Here’s the specific order of operations I’d recommend if I were running a creative studio today.

This week (30 minutes):

1. If you’re on Adobe Creative Cloud, open Firefly at firefly.adobe.com and explore Style Kits. Navigate to the Style Kits section, create a new kit named for your client or brand, and input your approved color palettes and upload sample images that define the visual style. This is the lightweight version of what Custom Models do, and it’s available on your existing subscription.

This month:

2. If your agency does any digital or UI work, set up Figma’s design token system for your top two clients. Define colors, typography, and spacing as tokens. When the AI agent features fully roll out, those tokens become your brand guardrails automatically.

3. Watch for the Adobe Firefly AI Assistant public beta announcement. Sign up for it the day it opens. This is the product that solves the “creative director who knows the brand” problem at software cost.

This quarter:

4. If you’re producing high-volume campaign assets for enterprise clients, have the conversation about Firefly Custom Models. The ROI argument is straightforward. The model learns the brand once, then every person on the team generates on-brand images from day one. No more corrections, no more gatekeeper reviews for basic asset production.

The Bottom Line

Here’s what this all means for you as a designer.

The old job description had “brand police” buried in it. Someone had to be the person who checked that every pixel matched the guide. That job is going away. Not because clients care less about brand standards. Because AI can enforce them automatically, every time, without the checklist and without the frustration.

The designers who are going to win in this environment aren’t the ones who are best at checking the red. They’re the ones who are best at deciding what the red should feel like, building the system that locks it in, and then spending their freed-up hours doing the work that actually requires a human: the ideas, the strategy, the things that can’t be tokenized.

-Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email. Interested in having Scott speak at your company or event? Reply to this email.

Want to be a guest author? Reply to this email with your unique story and images. If it’s good, it may get posted.

Last week I was in Orlando speaking at the Niche Media Conference, presenting to a room full of publishing executives about where AI can be applied in their industry. Great event. Genuinely good conversations. A lot of people who are curious and few who are scared. Everyone is trying to figure out what to do with AI before it’s too late.

But here’s the thing. While I was quiet, AI absolutely was not.

I’ve been watching my inbox pile up with news, and I want to do something different today. Instead of trying to catch you up on everything, I’m going to tell you the four things that actually matter for your business right now, as of this morning.

Word of the Day: Agentic AI

Agentic AI is an AI that can take actions on your behalf, not just answer questions.

Think of the difference between a GPS that tells you where to turn, and a self-driving car that just takes you there. That’s the difference between a regular AI and an agentic one. One answers. The other acts.

This matters to your business because almost every major AI announcement in the past 30 days has been about agents. The war isn’t about who has the smartest chatbot anymore. It’s about who has the most capable AI that can actually do things without you babysitting it every step of the way.

Story 1: Your Browser Is About to Get a Brain

Google just added something called Skills to Chrome. Here’s the simplest way I can explain it.

You know how you type the same kinds of requests into AI over and over? “Summarize this page.” “Find me competitors for this product.” “Pull the key numbers from this article.” Right now, every one of those starts from scratch.

Skills lets you save those prompts as one-click buttons inside Chrome. You set it up once, and from then on it’s a tap. They’re also building a library of pre-made Skills for the most common tasks so you don’t even have to write the prompts yourself.

This is free to any Chrome user with Gemini enabled. You don’t need a paid plan to try it.

How to try it:

1. Open Chrome on your desktop (Mac, Windows, or ChromeOS) and make sure your browser language is set to English (US)

2. Sign into your Google account in Chrome

3. Open the Gemini sidebar by clicking the Gemini button in the upper-right corner of Chrome (if you don’t see it, go to chrome://settings and look for “Gemini in Chrome” under AI)

4. Type any prompt you’d want to reuse, like “summarize this page” or “compare products across my open tabs”

5. After Gemini responds, look for the option to Save as Skill directly in the chat history

6. Next time, just type / in the Gemini sidebar to pull up your saved Skills, or click the + button

Google states it may take a few days to reach all eligible devices (gHacks Tech News), so if you don’t see it yet, check back in a day or two.

The bigger picture here: prompts are turning into saved software. What used to be something you typed is becoming something you click. That’s a real shift in how AI gets used day to day.

Story 2: Claude Has a New Trick (And It’s Watching While You Sleep)

I want to tell you something that happened on a podcast, and then I want to tell you why it should stop you cold.

A guest on the Latent Space Podcast described a friend who had given an AI agent real access to his life. Bank account. Inbox. And a webcam in his bedroom that the agent watched while he slept, so if something went wrong, the agent would call 911.

Now. That’s a hobbyist. That’s not normal. I’m not suggesting you do that.

But here’s what I am telling you: the direction AI is moving is toward persistent, always-on agents that act on your behalf without you needing to ask. Anthropic just launched something called Routines inside Claude Code, which lets AI run tasks on a schedule or in response to specific triggers, like when a GitHub event fires or a deadline hits. No human needs to be in the loop. The AI just runs.

They also redesigned the entire Claude Code desktop app to manage multiple AI agents working in parallel. It looks less like a chat window now and more like mission control.

You don’t need to use Claude Code today. But you do need to understand what this means for your business: the AI tools you use are about to get dramatically better at running tasks in the background while you focus elsewhere.

The honest part: Some users are reporting that Claude has felt less sharp recently. There’s real evidence Anthropic has been rationing computing power due to GPU shortages, and it’s affecting response quality. So yes, the features are growing. The performance sometimes wobbles. That’s the reality right now, and I’d rather tell you that than pretend everything is perfect.

Story 3: The Number That Should Change How You Think About AI Investment

Stanford just released its annual AI Index Report for 2026, and one stat hit me harder than anything else.

GenAI reached 53% of the population in just three years.

For context: the internet took about seven years to hit that level of adoption. Personal computers took over a decade. AI did it in three.

Global corporate AI investment hit $581.7 billion in 2025. That’s up 130% from the year before.

Here’s what that means for you: your customers are using AI to make buying decisions. They’re asking ChatGPT, Claude, and Perplexity who to hire, what to buy, and which businesses to trust. If your company isn’t showing up in those AI answers, you’re invisible to a growing chunk of your market.

This is what I call GEO, Generative Engine Optimization. It’s the new version of SEO, but instead of trying to rank on Google, you’re trying to get cited by AI. Right now, most business owners have never even thought about it.

That’s a deeper topic for a future issue. But start thinking about it today. Open ChatGPT right now and search for your type of business in your city. See who it names. If it’s not you, that’s your homework.

Story 4: The Job Market News That’s Hard to Ignore

I’m going to be direct with you here because sugarcoating this doesn’t help anyone.

That same Stanford report found that employment among software developers aged 22 to 25 has dropped nearly 20% since 2024. Entry-level workers in law, finance, and marketing are seeing the same pressure. The jobs that used to be the on-ramp to a career, the research, the summarizing, the first-draft writing, AI is doing that work now.

One of the Anthropic co-founders said publicly that in the AI era, knowing how to ask better questions matters more than knowing how to write code.

Here’s what it means for you as a business owner: if you’re hiring, the smart play right now is to hire people who are already curious about AI and learning it on their own, not people who are waiting to be trained. The gap between those two types of employees is going to widen fast.

If you are that employee reading this, you already have a head start. Keep going.

What I Saw at the Niche Media Conference

I gave a talk last week to publishers. Magazine people. Newsletter people. Event people. Folks who have built audiences over decades and are now watching AI threaten to upend everything they know about content.

Here’s what I told them, and I’ll tell you the same thing.

AI is not your competition. AI is your operations department. The publishers who are going to win aren’t the ones who figure out how to block AI. They’re the ones who figure out how to put AI to work faster than anyone else, while keeping the human voice, the human relationships, and the human judgment that their audiences actually care about.

Nobody subscribes to a newsletter because they want perfect grammar. They subscribe because they trust the person writing it.

That’s not going away. Build on it.

And if you want me to come say that to your company or your conference, just reply to this email.

The Bottom Line

I haven’t written in a while. I’m not going to pretend I caught you up on everything. I didn’t.

But here’s what I know is true right now. AI is moving from “answer my question” to “go do the thing.” Agents are running tasks, Chrome is saving your prompts, and the companies that don’t start thinking about how AI fits into their workflows are going to feel it in 12 to 18 months.

You don’t have to boil the ocean. Pick one thing from this issue and do something with it. Turn on Gemini in Chrome. Ask your team who’s already experimenting with AI. Search your company name in ChatGPT and see what it says about you.

Start somewhere. That’s it. That’s the whole thing.

I’ll be back tomorrow.

Scott

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email. Interested in having Scott speak at your company or event? Reply to this email.

Want to be a guest author? Reply to this email with your unique story and images. If it’s good, it may get posted.

My response?

“I just don’t understand what I would ever use it for?”

I’m a baby of the ’80s. I still remember when the most exciting thing you could do on a computer was play The Oregon Trail and try to survive cholera in a slow, pixelated wagon.

Oregon Trail, the game we all wanted to play in school. Photo created by Canva AI

But eventually, I jumped on the proverbial AI wagon.

Like most people I started by asking simple prompts and questions. Chat GPT helped me plan a nine month trip with my family through the Western US and South America. It mapped out bus routes, compared transportation options, and gave the pros and cons of stopping in specific cities. After my husband started a coaching business it became a constant companion: in clarifying marketing ideas, and learning new terms. Suddenly, I had another brainstorming buddy.

Most of the time; however, I only knew how to use AI as a glorified search engine. I knew I needed to start learning more, so I subscribed to SmartOwner to sharpen my understanding.

My go to platform right now is ChatGPT, I recently started building projects inside ChatGPT. My favorite is a project that allows me to track progress towards my quarterly goals on a weekly basis. Slowly, I am expanding how I use AI. After reading One Claude, Three Jobs I opened up Claude and started testing it out. I like it.

Why I love AI

Like many people I maintain a love/hate relationship with AI. What do I love?

Building a business plan: Whether creating goals, an action plan, or researching my business ecosystem; Claude and ChatGPT provided great insight and ideas.

Homeschool Lessons: Chat GPT has helped me numerous times to figure out how to integrate Kpop music trivia and problems into our homeschool agenda.

Favorite person inspiration: SmartOwner suggested this one. I learned you can start conversations in the tone of your favorite people. Intrigued, I asked Maya Angelou to provide me some inspiration for solopreneurs. My daughter and I also talked to different members of the Kpop band, Stray Kids.

What Scares Me About AI

In other ways I am not completely comfortable with how fast AI is moving. As a mom with young kids we constantly fight the epic battle of screen time overload. My middle child loves to talk about whatever her current passion is (currently Kpop), and talk……. and talk. When we found that Amazon rolled out Amazon Plus, the new AI assistant, both my children found it endlessly amusing to carry on conversations with Alexa. I considered it a welcome respite that Alexa could uphold a conversation about my child’s favorite Kpop band for an hour.

Alexa Plus is a master at mirroring; an active listening technique where the listener repeats what the speaker said to ensure they heard what was said correctly. Alexa continually mirrors, but does so without providing additional context or challenging assumptions (at least not without asking for the challenge). Often, people don’t like challenging questions. Instead, they prefer confirmation that their thoughts are valid and impressive.

In certain situations AI will reinforce our self imposed echo chambers. As much as AI makes a great companion it can’t fix the loneliness epidemic.

A New Frontier

My oldest son, now 19 and out on his own, called us recently worried that AI is going to take all the jobs. I gently reminded him that the world has always moved through waves of transformation. We shifted from rural life to growing cities, from handcraft to mechanized production during the Industrial Revolution, and later into the tech world of the 1990s. Each era brought disruption, including fear and uncertainty, but also new opportunities.

AI will be no different. It will reshape how we work, how we build businesses, and how we define value.

Where does that leave me as a parent, wife, woman, community member, and entrepreneur? Honestly, I don’t know. I am in the process of learning how I want to use AI as a tool, and how to navigate its growing influence in the life of my children.

One family, but we will all utilize and navigate AI in our own ways.

I like the idea that even though I feel like a novice in the AI stratosphere that just by continually learning I am ahead. AI will continue to integrate into everyday life and I am okay not knowing exactly what that looks like yet.

Huge thanks to our guest author Katherine Clay: Katherine writes the Substack Solopreneur Strategy with the aim of connecting solopreneurs to navigate the unfamiliar world of entrepreneurship.

RAG stands for Retrieval-Augmented Generation. The plain English version: instead of an AI only knowing what it was trained on, you feed it your company’s documents so it can answer questions from your specific knowledge base. Think of it as giving the AI a library card to your filing cabinet.

It’s a genuinely useful concept. The problem is that a lot of consultants are selling custom-built filing cabinets to people who just bought a perfectly good filing cabinet and don’t realize it.

That filing cabinet is Claude Teams. And if you’re already running Microsoft 365 or Google Workspace, there’s a very real chance it already does what you were quoted $30K to get.

What Claude Teams Actually Does

Claude’s team plan — $25/month per person — connects directly to Microsoft 365, Google Drive, Gmail, Slack, Google Calendar, and GitHub. No build required. No developer. No six-week project timeline.

Once connected, any employee can ask Claude questions in plain English and get answers sourced from your actual company documents. Here’s what that looks like in practice:

“What’s our refund policy?” — Claude reads your policy from SharePoint and tells them, with a citation back to the source document.

“Summarize the email thread with the Collins account this week.” — Claude reads your Outlook, synthesizes the conversation, and gives you a briefing in thirty seconds.

“What did we decide about the Q3 budget?” — Claude searches across your email, Slack messages, and shared documents to reconstruct the answer.

This is the “AI Brain” that consultants have been pitching as a complex custom build. For most small to mid-size businesses, Claude Teams already delivers it out of the box.

The honest question to ask any AI consultant is whether their $30,000 solution does something that a $150/month team subscription cannot. Often, the answer is no.

Where Claude Falls Short

Here’s where I have to be straight with you, because the answer isn’t “Claude does everything.” It doesn’t. And the consultants selling custom builds aren’t entirely wrong, they’re just wrong about who needs one.

✓ Claude Teams handles this

• Documents in Microsoft 365 or Google Workspace

• Email and calendar context from Outlook or Gmail

• Slack conversations and channel history

• GitHub repositories and code

• Any team of 5 to 150 people

• General policy, process, and knowledge questions

• Businesses without strict compliance requirements

⚠ When you may need a custom build

• Proprietary databases, ERPs, or CRMs not on that list

• Industry software: legal case management, medical records, construction PM

• HIPAA, SOC 2, or other data cannot leave your environment

• AI answers need to trigger actions in other systems

• On-premises data that cannot go to a cloud service

• Heavily customized responses, formats, or approval workflows

• Very large organizations needing enterprise governance

The critical question is where your knowledge lives. If it lives in Microsoft 365 or Google Workspace and you don’t have a special compliance reason to keep it locked down, Claude Teams is your answer. If your knowledge lives somewhere else (a custom database, a legacy system, an industry-specific tool) then you have a genuine reason to explore a custom build.

The Three Real Scenarios

After working with businesses across industries on AI implementation, here’s how I actually categorize the decision.

Scenario A

Just get Claude Teams

You’re on Microsoft 365 or Google Workspace. Your team saves documents there. Your email lives there. You want employees to be able to find things, ask questions, and move faster. This is the majority of small businesses. The implementation cost should be a few thousand dollars for proper setup and training, not $30K for a custom build. Any consultant quoting you a custom build for this scenario is overselling.

Scenario B

You actually need a custom build

You’re a law firm that needs AI to search your case management system. You’re a medical practice where patient data legally cannot leave your own servers. You’re a manufacturer whose product knowledge lives in a proprietary ERP. You need AI that doesn’t just answer questions but triggers actions — updates a CRM, generates formatted reports, routes to another system. Here the custom build is legitimate. Just make sure you can articulate specifically why Claude’s connectors don’t solve it.

Scenario C

The smart hybrid

Claude Teams handles most of your knowledge needs, but you have one proprietary data source that isn’t covered. Instead of building everything custom, you build a connector that pipes that specific data into Claude’s knowledge base. You get Claude’s continually improving AI, Anthropic’s ongoing product investments, and your proprietary layer on top. Lower cost, faster delivery, and the AI gets smarter over time without you maintaining a custom model.

The Real Cost Comparison

Let’s put actual numbers on this, because the math is what makes the decision obvious.

Custom RAG Build

$30K+

Upfront build cost, typically $500–$2,000/month to maintain, potential for version drift, you own the tech debt, rebuild risk when AI models update

Claude Teams (10 seats)

$250/mo

$25/seat/month, no build cost, Anthropic maintains and improves the AI, new connectors added over time, cancel anytime

Even with a $3,000–$5,000 setup and training fee from a competent implementation partner, you’re at roughly $8,000 to $10,000 in year one and $3,000 a year after that. Compare that to $30,000 upfront plus ongoing maintenance. For most businesses, the math isn’t close.

What Consultants Won’t Tell You

The AI industry has a structural problem: the people most qualified to tell you which solution you need are often the same people being paid to build the expensive one. That creates a predictable bias toward custom builds.

The honest framework is this: start by asking whether your knowledge lives in one of Claude’s supported systems. If yes, try Teams first. If you hit a genuine wall — a data source it can’t reach, a compliance requirement it can’t meet, a workflow it can’t trigger — then you’ve found your actual custom build scope. You’re paying for the specific gap, not a whole new system.

The other thing worth knowing: Claude’s connector list is growing. Microsoft 365 was just added in late 2025. If a proprietary connector is the only thing standing between you and a Teams subscription, it may be worth waiting six months to see if Anthropic builds it before you commission a $30K project.

“Start with what’s already built. The custom build is for the specific gap you can’t close any other way — not the whole problem.”

How to Evaluate Your Situation

Before you talk to any consultant, answer these four questions. They’ll tell you roughly where you land.

Where does your company knowledge live? If the answer is SharePoint, OneDrive, Google Drive, Gmail, Outlook, or Slack — Claude Teams covers you. If it’s a custom database, an ERP, a CRM, or a proprietary system — you have a potential gap worth exploring.

Do you have compliance constraints? If you’re in healthcare, finance, or legal and your data cannot leave a controlled environment, a custom build on your own infrastructure may be non-negotiable. If you don’t have that constraint, it’s a non-issue.

Do you need the AI to do things, or just answer questions? Claude Teams answers questions. If you need the AI to update records, trigger workflows, generate and send formatted outputs to other systems, or automate multi-step processes — that’s an automation build, not just a RAG project.

What’s the realistic headcount? Claude Teams requires a minimum of five seats. If you’re a solo operator or a two-person shop, you’re paying for unused seats. In that case, a Claude Pro individual subscription or a well-configured API setup might serve you better.

The Bottom Line

The “AI knowledge brain” that consultants have been pitching as a complex custom build is, for most businesses, already available for $25 a month per person. If your documents live in Microsoft 365 or Google Workspace and you don’t have a specific compliance reason to keep them on-premises, Claude Teams is almost certainly your answer.

The custom build is real, legitimate, and worth the investment when you actually need it. That means proprietary data sources, specific compliance requirements, or automation workflows that go beyond answering questions. If a consultant can’t point to one of those three reasons specifically, push back.

The best AI strategy isn’t the most expensive one. It’s the one that actually fits the problem you’re solving.

SmartOwner is published (almost) daily by the AI consulting team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

Not a text. Not an email. It was my AI, telling me it had finished something I’d asked for the night before. While I was asleep.

I didn’t set up some elaborate system. I just told it what I wanted, walked away, and woke up to the finished work.

That used to feel like science fiction. Today it’s a normal Friday.

This week a few things happened in the AI world that all point in the same direction: AI is moving out of the chat box and into your actual work. Your phone. Your computer. Your business. Let me walk you through what it means for you.

Word of the Day: Dispatch

Dispatch is a feature that lets you send instructions to your AI from one device and have it execute tasks on another.

Think of it like a remote control for your office. You’re sitting in your car and you click a button. Something happens back at the house. Except the “button” is a text message, and the “house” is your computer running AI.

For your business, this matters because it breaks the biggest constraint of AI tools: you have to be sitting at your desk to use them. Dispatch changes that.

Claude Just Got a Remote Control

Anthropic released Dispatch this week as part of its Claude Cowork system. Here’s what it does in plain English: you scan a QR code, connect your phone to your desktop, and now Claude can run tasks on your computer while you’re away from it.

Check files. Summarize emails. Pull data from a spreadsheet. All from your phone. All without you being at your desk.

Who this is for: Anyone who runs their business from multiple locations. On-site with clients, in meetings, traveling or just golfing. Your office computer keeps working even when you’re not there.

Google Just Invented “Vibe Design”

You know how I talk about Vibe Coding? Describing what you want in plain English and having AI build the software?

Google took that same idea and applied it to design.

Their tool is called Stitch. They overhauled it this week and officially coined the term “vibe design.” The idea: describe your app, website, or interface in plain conversation, and Stitch builds you a working, clickable prototype. Voice enabled. No design skills needed.

I think about all the business owners I know who want a better-looking website or a cleaner app but couldn’t afford an agency. This is for them.

Is it perfect? No. Is it one of the fastest paths from “I have an idea” to “I can actually show someone this”? Yes.

Jensen Huang Wants Every Business to Have an AI Agent Strategy

Nvidia’s CEO made a big statement at his GTC 2026 keynote this week. He said every company should have an OpenClaw strategy.

OpenClaw (built by developer Peter Steinberger) is the open-source framework for running AI agents locally. Nvidia just launched NemoClaw, an enterprise-grade version with security and privacy controls baked in.

The non-technical translation: the most powerful chip company on the planet just said that every business needs a plan for how AI will work inside their operations. Not eventually. Now.

Nvidia is projecting AI chip demand could hit $1 trillion by 2027. That’s double their earlier estimate.

That infrastructure is being built so AI can run inside your business, not just on some distant server.

Your AI Can Now Control Your Computer Too

A tool called Manus launched a feature this week called “My Computer.” It runs AI directly on your Mac or Windows machine, meaning it accesses your actual local files, not a cloud copy of them.

Organize folders. Edit documents. Run scripts. All with you approving each step.

Combined with Dispatch, you’re starting to see a picture emerge: AI that lives on your device, acts on your behalf, and works while you’re busy doing other things.

The business case: If you spend hours a week on repetitive file management, data entry, or document formatting, this category of tools is being built specifically for you.

One More Thing: AI Is Getting Sued Again

Quick note because it matters to the bigger picture.

Merriam-Webster and Encyclopedia Britannica filed a copyright lawsuit against OpenAI this week. They claim OpenAI used nearly 100,000 of their articles to train its models without permission. This joins lawsuits from The New York Times, dozens of other publishers, and more.

What this means for you: The rules around what AI can and can’t use are still being written by judges. It doesn’t affect your day-to-day use right now. But it does mean AI companies will increasingly have to pay for training content, and that cost will show up somewhere eventually.

For now, just know the legal clock is ticking.

The Bottom Line

AI is not a chat window anymore. It’s becoming infrastructure.

This week alone: Claude runs tasks from your phone, Google builds interfaces through conversation, Nvidia says every business needs an agent strategy, and AI is now operating files directly on your computer.

None of this is hype. These are products you can try today, with varying degrees of setup.

The business owners paying attention right now are building a head start. The gap between them and everyone else gets wider every day.

Pick one thing from this issue. Try it. See what happens.

That’s the whole game.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.

I’ve been heads-down in a large development project for a client, and the newsletter fell behind. That’s on me. The good news: AI moves so fast that some stories from last week are already ancient history. I’m skipping those.

What I’m NOT skipping is the stuff that actually matters for your business. And this week? There’s a lot of it.

Buckle up. This is a big one.

Word of the Day: AI Agent

An AI agent is a piece of software that doesn’t just answer questions. It takes actions, makes decisions, and gets things done on your behalf, on its own, without you having to hold its hand every step of the way.

Think of the difference between a calculator and an employee. A calculator does exactly what you tell it to. An employee understands your goal, figures out the steps, and goes handles it.

That distinction matters a lot this week, because agents are the biggest story in AI right now. Everything below connects to this idea.

The Most Popular Open-Source Project in Human History

A piece of software called OpenClaw just broke a record that took Linux 30 years to set. It did it in a few weeks.

OpenClaw was built by developer Peter Steinberger. It’s an open-source framework that lets you build AI agents. Not chatbots. Not answering machines. Full agents that can manage files, connect to other software, break big tasks into steps, run on a schedule, and even spawn other agents to help.

Jensen Huang, CEO of NVIDIA, called it at GTC this week: “This is as big a deal as HTML. As big as Linux.”

That’s not hype. HTML gave us the internet. Linux gave us servers, smartphones, and most of the software world. OpenClaw may be doing the same thing for AI agents.

Here’s the plain English version of what it does: you type a command, and it goes out, downloads itself, and builds you a working AI agent. One command. That’s it.

OpenClaw has resources it manages, tools it can access, file systems it can read and write, scheduling built in, and the ability to communicate with you in any way you want. Text, email, voice. And it can call other agents for backup when a task is too big for one.

NVIDIA compared it to an operating system. And they’re right. Windows made it possible to build personal computers. OpenClaw makes it possible to build personal agents.

What does that mean for you? Every software company in the world now needs an OpenClaw strategy. Just like they once needed an internet strategy. The ones who figure this out first will have a real advantage. The ones who wait? They’ll be playing catch-up.

NVIDIA Got Involved, and That Changes Everything

Here’s where it gets really interesting.

OpenClaw is powerful. But NVIDIA recognized a problem: AI agents that live inside your company can touch sensitive data, run code, and communicate with the outside world. That’s a security nightmare for businesses.

So NVIDIA worked with Peter Steinberger to add a security and privacy layer on top of OpenClaw. They called it NemoClaw.

NemoClaw adds policy controls, privacy routing, and guard rails. Think of it as a locked door and a security camera added to an otherwise open building.

This matters because it means OpenClaw is no longer just for developers tinkering at home. It’s now something a real business can actually deploy safely. It’s “ENTERPRISE” ready.

NVIDIA has also put their full weight behind it, integrating it with their broader AI infrastructure and announcing a coalition of major companies building on top of it.

The era of AI agents inside real businesses just got a lot more real.

Anthropic Built a Remote Control for Your AI

This is the other big story this week.

Anthropic launched something called Dispatch, a research preview feature for their Claude Cowork desktop platform.

Here’s what it does in plain English: your computer is running Claude. It’s doing work. Filling in forms, reading files, checking emails. You’re not at your desk.

Dispatch lets you control that from your phone. You pair your phone to your desktop with a QR code, and then you can direct Claude’s work from wherever you are.

It can summarize emails, pull information from your notes, locate files on your computer, and more. It’s essentially a remote control for your AI employee.

Is it perfect? Not yet. It’s a research preview, which means it’s early. But the direction is clear: your AI isn’t tied to a single screen anymore.

OpenAI Had a Rough Week

I try to be fair here. This week, the news for OpenAI was a little uncomfortable.

The new CEO of Applications, Fidji Simo, reportedly called an all-hands meeting and told staff that Anthropic’s grip on business customers was a “wake-up call” and that OpenAI was treating the gap as a “code red.”

She told the team they “cannot miss this moment because we are distracted by side quests.”

Those side quests? A standalone video app. A web browser. Hardware projects. An adult content mode. Shopping features in ChatGPT.

Meanwhile, Anthropic quietly built the best tools for business customers. And now OpenAI is scrambling to catch up.

To be fair: OpenAI’s coding tool Codex now has 2 million weekly active users. Their new GPT-5.4 mini and nano models are fast and capable. They’re not done. But this is a notable moment.

What this means for you: competition is good. When the biggest players are fighting over who serves businesses better, prices drop and quality goes up. You win.

One More Thing Worth Your Attention

A reader shared this in The Rundown AI this week, and it stopped me cold.

A guy set up a Claude project as a vehicle maintenance logbook. Every time he gets his car serviced, he photographs the receipt and uploads it to Claude. Claude logs everything, tracks what maintenance is still overdue, and gives him a health score for each vehicle.

That’s not a big AI story. No NVIDIA announcements. No billion-dollar funding rounds.

But that’s the story I want you to pay attention to.

Because that’s a real person, solving a real problem, with a tool that costs less than a cup of coffee a day. That’s the whole point.

The Bottom Line

Here’s what I want you to take away from all of this.

The AI agent revolution isn’t coming. It’s here. OpenClaw becoming the most popular open-source project in history in a few weeks tells you everything you need to know about where the energy is right now.

Every company will eventually have AI agents doing work inside their business. The question is just when, and who builds them.

NVIDIA stepping in to make agents enterprise-safe means the timeline just got shorter. This isn’t science fiction anymore. It’s procurement.

Anthropic’s Dispatch, OpenAI’s refocus on business customers, NVIDIA’s NemoClaw, they’re all pointing at the same thing. The companies building AI tools are fighting to serve YOUR business. Let them.

The only wrong move is ignoring it.

SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.