A hacker, a few thousand dollars, and 100+ companies breached. A cybersecurity researcher watched in real time as a Chinese-speaking hacker used AI agents to break into websites belonging to more than 100 companies, including John Deere, SharkNinja, and Cummins. The cost to run the whole operation was a few thousand dollars. That's the part that should stop you cold. Targeted cyberattacks used to require expensive, skilled operators who had to pick their victims carefully. AI just turned that into a commodity, like buying a mailing list and hitting "send." If you've been telling yourself your business is too small to bother with, that assumption is now dangerously out of date. Check your site's security posture this week. Not next quarter.
The singularity, apparently, has started. Jack Clark, co-founder of Anthropic, told IMF and World Bank economists that we are already in it. His evidence is concrete: AI benchmark scores on expert-level questions jumped from 8% to 67.7% since 2025. That's not a trend line. That's a vertical wall. He also flagged "very disruptive" job impacts coming and said there are only "a small number of years" to act. He projects a possible $44 trillion U.S. economy by 2030. Maybe that number lands, maybe it doesn't. But the practical signal for owners is narrower: the window to build AI into your operations before your competitors do is closing faster than most forecasts suggested. Read his full remarks here.
Notion's AI agents now have their own computer and can work all day without you. Notion agents can now run on a dedicated cloud machine for up to 24 hours straight. Hand off a multi-step research project, a documentation rewrite, a competitive analysis, whatever you'd normally delegate to a junior hire. Come back to finished output. If your team already lives in Notion, this is a meaningful upgrade, you don't need to bolt on a separate automation tool to get that kind of hands-off work done. The details are here.
Tools Worth a Look
Wabi 2.0: A shared AI agent that lives inside group chats, builds apps, updates calendars, and coordinates travel for the whole team, not just one user at a time. Good for small teams already coordinating inside group threads.
Scrawl: Draw and talk over your screen to describe a bug or task, then hand it straight to your AI agent. Saves owners who work with developers from writing up long issue descriptions.
Inflight: Collects feedback on a prototype or design and automatically turns it into structured next steps for Claude. Cuts the manual "okay, what did everyone actually say" synthesis step.
OpenUI 1.0: Adds buttons, forms, and structured inputs to AI agents so you're not stuck with a plain text box. For owners building or customizing AI tools who want a polished interface without heavy dev work.
Idealab.ai: Bill Gross's database of 1.2 million startup ideas, with $100M available for founders who want to build them. Worth a look if you're considering a new product line or a pivot.
Trending on GitHub
vllm-project/vllm: A fast, memory-efficient engine for running large AI models, think of it as a high-performance server rack for serving your own AI, built for speed at scale. (441 stars today)
DietrichGebert/ponytail: Makes your AI agent default to the simplest possible solution rather than over-engineering everything. The pitch: the best code is the code you never had to write. (2,079 stars today)
supabase/supabase: A full database platform built on Postgres, gives developers a dedicated database for web, mobile, and AI apps without the usual setup headache. (504 stars today)
twentyhq/twenty: An open-source CRM built as an alternative to Salesforce, designed from the ground up to work with AI. (112 stars today)
shiyu-coder/Kronos: An AI model trained specifically on financial market data, think of it as a language model that speaks in price patterns and market signals instead of plain text. (575 stars today)
Also Worth Knowing
Supabase is changing how it handles API access for new tables on October 30, if you're running any Supabase-backed tools or automations, new tables will need an explicit permission grant or your app hits a "permission denied" error in production. Existing tables are fine. Check your upcoming migrations now.
ChatGPT can now take over your Chrome tabs, autonomous browser control just landed for users, meaning you can point it at a task and let it click through pages on your behalf.
GPT-Live-1.1 shipped with new voices and better accent rendering across ChatGPT, Codex, and the dots ambient agents, API access coming soon, and the quality bar for voice-based customer interactions just moved up again.
Meta released Muse Spark Open, its open-weights frontier model, now available for developers to run and customize.
The US government banned DeepSeek from federal civilian agencies in the latest escalation around Chinese AI tools, worth tracking if you work with government clients or handle sensitive data.
Google's ad-tech antitrust ruling is coming, but don't plan your revenue around it, analysts expect publisher yields to stay mostly flat in the short term even as Google is required to make behavioral changes.
EMARKETER is pairing star analysts with proprietary data as its AI strategy, a B2B playbook that's working and worth understanding if you're in a knowledge or content business.
Disney is reportedly in talks to acquire Character AI assets for $20M after the company filed for bankruptcy, context for anyone watching where the AI companion space is heading.
Inception's Mercury 3 is now leading benchmark tests on speed in the API market, relevant if you're building or using latency-sensitive AI tools.
-Scott
SmartOwner is published (almost) daily by the team at DigitalTreehouse. Want AI consulting or automations for your business? Reply to this email.