Hi, I’m Vinesh. I build AI agents, and then I work very hard not to trust them.
That tension is what this site is about. An agent that can read your mail, call your tools and spend your money should have to prove a few things before anyone lets it act. Working out what, and then finding out the hard way where our own answers were wrong, turns out to be a full-time hobby.
Three kinds of things live here:
- The doctrine. A long series on zero trust for agentic systems: what an agent must prove, to whom, and when. Start at the Spotlight.
- The confessions. Green Is a Claim: every time our own dashboards, tests and logs said yes and the world said no. The doctrine was the easy part.
- The bench. Labs: small katas and codelabs you run on your own machine. Build it. Break it. Fix it.
A few house rules you’ll notice:
- A claim should point at something you can run or check.
- The failing test is usually more interesting than the passing one.
- When a post is wrong, it gets corrected in public, not quietly deleted.
Want to compare notes, argue with a post, or build something together? Write to the
webmaster. I hope you leave with a smile, and then pass it on. 
Designed, built and maintained by Vinesh somewhere in the Pacific Northwest, usually under a cloud (the rainy kind and the computing kind). A seeker, making the most of now.