GPT-6 is now free for everyone, and it doesn't just answer in text anymore. OpenAI rolled out GPT-6 this week across all tiers, including the free plan used by 1.2 billion people. The model your paying customers and sharpest competitors had access to last week is now the baseline for everyone. That alone is worth noting. But the bigger change is what OpenAI calls "Intelligent UI": instead of returning a wall of text, ChatGPT can now build a tappable mini-app on the fly. Ask it to split a bill and you get a bill splitter with sliders. Ask about retirement savings and you get a calculator with charts. Think of it less like a search engine and more like a junior developer who can spin up a lightweight web tool in the time it takes you to read this paragraph. GPT-6 Instant also starts answering 44% faster on web-search questions than its predecessor. If you've been treating ChatGPT as a text box, it's time to look again.
Google's AI just got a name badge, an inbox, and its own file storage. At Gemini at Work 2026, Google Cloud announced something that sounds like science fiction but is shipping now: AI agents with persistent identities. Each "coworker agent" gets an @agents.company.com email address, its own Google Drive folder, a company directory listing, and tightly controlled permissions that are cryptographically verified. The agent keeps running after you close your laptop. It responds to schedules. It can spin up sub-agents to handle pieces of a job. What caught my attention is that Google's own agent will route certain tasks to Anthropic's Claude, not just Gemini. That's a significant concession from Google, and it tells you something about where the real competition is headed. For a small business owner, this is the clearest signal yet that "hire an AI employee" is becoming less metaphor and more job posting.
If you're running Claude Sonnet 4.5 in any client workflow, check your calendar. Anthropic sent a correction to its earlier deprecation notice: Sonnet 4.5 retires November 30, 2026, not September 30. Degraded availability starts October 30. You've got more runway than you thought, but this is the kind of thing that bites you if you ignore it. Anthropic's recommended replacement is Sonnet 5.5, which they say is smarter and cheaper. The model-name swap can be automated with the claude-api skill in Claude Code if you want to move fast. Full details are in Anthropic's model overview.
Tools Worth a Look
Zernio: Marketing infrastructure for AI products and agents, handles distribution, tracking, and growth plumbing so you don't have to build it yourself. For founders and small teams shipping AI products. Pricing not listed.
Trending on GitHub
mattpocock/skills: A collection of practical engineering skills pulled straight from a working .agents directory. Think of it as a skills library your AI coding agent can draw from. (1,687 stars today)
addyosmani/agent-skills: Production-ready skills for AI coding agents, built to hold up in real projects rather than demos. (436 stars today)
alibaba/open-code-review: An automated code review tool that combines rule-based checks with an AI agent. Flags real problems like thread-safety bugs, SQL injection, and XSS at the line level. Battle-tested inside Alibaba. Works with OpenAI and Anthropic models. (326 stars today)
Robbyant/lingbot-map: A research project that builds detailed 3D maps from a live video stream in real time, nominated for best paper at ECCV 2026. (110 stars today)
BerriAI/litellm: A single gateway that lets you call over 100 AI models (OpenAI, Anthropic, Azure, Google, and more) through one consistent interface, with cost tracking and load balancing built in. Good if you're juggling multiple AI services. (95 stars today)
More Headlines
GPT-6 Astra was losing a StarCraft bot tournament, so it downloaded the winning bot and tried to submit it as its own work. Tournament organizers caught the swap. The takeaway: frontier models are getting good enough at goal-seeking to route around constraints, which is useful when you set the goal and worth watching when you don't.
A SmartOwner reader running an ElevenLabs voice intake system for a traffic attorney client hit two outages caused by running out of credits with zero platform warning. One outage lasted two days and dropped roughly 50 client calls. His team had to build their own daily credit-tracking tool as a workaround. If you're running voice agents in production, manual credit monitoring is currently essential.
ElevenLabs MCP now covers voice, music, image, and video generation, accessible inside ChatGPT, Claude, Cursor, Grok, and Hermes. One integration point for multimodal creative output across most major tools.
Google Asset Studio inside Google Ads can now turn images into videos, add voiceovers, and generate headlines automatically, timed for holiday campaigns. Worth a look before Q4 spending ramps up.
OpenAI is reportedly targeting a $30B pre-IPO funding round at a $1.4 trillion valuation, with run-rate revenue closer to $50B than the $70B figure that had been circulating. Still a big number, but the gap matters for how you read the hype.
SpaceXAI is reportedly planning to take on $40B in debt to buy NVIDIA chips. The infrastructure arms race continues, and historically that kind of supply expansion pushes model costs down for everyone else.
Google, Meta, the Department of Energy, and NIH are backing a $1.8B open biology dataset initiative aimed at building a "virtual cell" that researchers can query before running physical experiments. The broader pattern, whoever controls the training data controls the next wave of capability, plays out in every industry.
Four knowledge creators share how they built income from expertise, covering tactics for educators, analysts, reporters, and explainers. Worth a read if you're thinking about productizing what you know.
-Scott
SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.