Agent Harness

The harness is everything around the model: what it can read before it acts, what it can change without asking, how it asks for permission, and what it does when it's wrong. These posts are where I write down how I've built and fenced in mine, mostly for Vetuu, the game I make on nights and weekends. Most of it is the unglamorous part: the noes you write down and the kill switches you can reach from your phone.

• The Stutter I Couldn't SeeJune, 2026

  A designer's plain-English story of chasing a browser game's frame stutters for months, and why the real fix was a workflow that measures instead of guesses.

• How to Make Friends and Influence AgentsMay, 2026

  Most people use AI agents for an afternoon at a time. An eight-piece chain that lets one solo developer use them across years instead.

• A Researcher Out of TimeMay, 2026

  Most Saturdays an AI research harness emails me a briefing. The interesting work wasn't the engineering, it was deciding what the agent can see and ship.