The Reviewer Who Actually Read the Code Found What Everyone Else Missed
The release was one “safe to ship” vote away from landing with a save path that did not save what the UI claimed.
The build log
Build log, architecture patterns, and observations from running autonomous AI systems in production.
The release was one “safe to ship” vote away from landing with a save path that did not save what the UI claimed.
At desktop width, the tournament navigation disappeared. The controls were still in the page. They were just not where we thought the breakpoint logic would put them.
On July 18, five of our ten clone cells looked unavailable even though the tickets tied to them were already finished or closed.
At 11:58 AM, five agent sessions were still live when I needed to advance main in the shared coordinator checkout.
The static scanner gave one Windows agent configuration a D, 57/100, after it found 24 issues and marked 14 of them high severity.
I kept saving goals to my agent's memory and they kept vanishing by the next session. The cause wasn't a flaky model. It was a silent truncation cap, and the fix is a pattern anyone running long-lived AI should steal.
One agent's browser is a config change. A fleet's browser is infrastructure: a locked slot pool, an isolated binary, tiled windows, and sticky logins.
At 11:06 PM on June 29, I had a post whose core failure was a 200-line or 25KB context cap, and I still did not let it ship after one AI said SHIP.
Three hours and 40 minutes into a tournament Teams-tab prototype, I had a working branch that deserved to be deleted.
Two builds shared one version number. I found it after a release I had just verified, and the collision meant the release tag could no longer answer the one question I needed it to answer: which build is act...
The real failures and fixes from building AI systems, one practical lesson per post. Get the next one in your inbox.