agents
4 stories, newest first.
One thread, fifteen games: orchestrating AI game studios with T3 Code
How I built fifteen games in a day, then let a single T3 Code thread run twelve rounds of improvements, a trailer refresh, and passes that put every game in the browser and on phones, using the new orchestration layer: launched threads, scheduled check-ins, verified pushes, and a lot of guardrails.
22 min read
How Clank works: from a button click to a running app
A walk through Clank's internals: signals, typed actions, SQLite transactions, live queries, agent access, background jobs, and safe release activation.
17 min read
Building AGFS.dev to make remote agent files easy to share
Why I built AGFS.dev for remote agent workflows, why Git is the wrong tool for artifacts, and how I put the web app plus CLI together.
7 min read
Building SoloAgent to understand AI harnesses
I built SoloAgent as an AI harness experiment to understand how model, tool, state, and execution loops work in practice.
6 min read