Your agent says it's done. Excellent checks.
AI agents can now produce more work than anyone can carefully review, and asking another model whether the first one did a good job does not close that gap. Excellent treats every “done” as a claim and goes to look: the repo, the tests, the build and the browser. What holds up leaves a receipt. What does not goes back to the agent with the reason.
I am also its first customer. About 12,300 commits landed in Excellent’s main repository between June and November 2026, and agents made roughly 9 in 10 of them. Running that much agent work is how the product learned where agents go wrong: tasks marked done whose work never reached main, workers stopped halfway and recorded as successes, an agent that quietly built a workaround for a tool it had been refused.
Excellent runs on your machine, beside the agents you already use, and keeps the evidence there. I made the brand too: the seven-ray star, a scanline typeface drawn on a six-row grid, and a world of ordinary objects that each check something.
- What I do
- Founder. Product, engineering, research and brand
- The rules
- The agent does not grade itself. A check that cannot run stays visible. History is append-only
- License
- Fair Source
