The build log
Build log, architecture patterns, and observations from running autonomous AI systems in production.
claude --resume <sid> worked in a terminal and returned No conversation found when a web app ran the exact same command, on the same machine, against a session file that demonstrably existed.
Building a conversation archive across four provider repos is the easy part. The useful part was sending the spec to Manus and watching it find gaps in every assumption.
Elapsed time, concurrent process time, human attention, and verified outcomes are different measures. Once work happens in parallel, a useful record must preserve provenance, uncertainty, review state, and the limits of what any productivity metric can claim.
We asked an agent to set up a scheduled job that would keep building out our agent-log archive.
AI made ticket analysis nearly free, which exposed the cost of a read-only backlog page. Inline verbs turned tab-thrash into one-click decisions.
A morning IPv6 tweak made localhost resolve to ::1 and every request hang for ten seconds. Timed curl probes found the dead hop in a three-hop proxy chain.
Why an AI-facing archive portal shipped as a narrow, auditable read-only path, and why write access waits until the key model has earned it.
We went in expecting to wire event registration onto the existing cart. One grep later, the premise was dead and we were designing from scratch. The new cart shipped the same day.
The ticket said STAGING/LIVE. It had a fix version, 3.349, and a real commit with a real diff.
Pro-tier bot rules never touched a residential proxy pushing 224k requests a day. JS Detections moved the fight to a capability it couldn't fake.
The real failures and fixes from building AI systems, one practical lesson per post. Get the next one in your inbox.