The build log
Build log, architecture patterns, and observations from running autonomous AI systems in production.
A diagnostic page returned the literal word `undefined` instead of a client value. The lesson was not just a language quirk: copies of shared logic had drifted away from the defensive helper.
A geo-block WAF rule that allowed US and CA silently locked out an entire US territory. How Cloudflare's ISO 3166-1 country codes work, the five territory codes to add, and a small lesson about what you can and can't claim when verifying a fix.
I clicked into a single sampled Cloudflare security event to see who my new bot protection was actually catching, and the top blocked source was a documented Googlebot address. That is Googlebot.
Phase 1b fixed the central abstraction. The pre-flight audit before flipping nameservers found eight more direct REMOTE_ADDR reads. Six got the same fix, one was dead code, and one had to stay broken because the safe alternative would have been worse.
I built a health dashboard precisely so cron jobs could not rot in silence. The cron job rotted in silence anyway, because the dashboard was watching the wrong thing.
Five firewall rules stopped a proxy swarm. The real bug was a log watcher covering one of three IIS sites, which kept the attack invisible.
A small project inventory stopped quiet repositories and leftover automation from disappearing from view. The important choice was simple: unclassified work had to show up as a decision, not an omission.
A suggested classification should stay editable until a person confirms it. Separating analysis from persistence turns an AI mistake into a correction opportunity instead of a bad record.
A preview that skips a real precondition can produce a reassuring number and still fail at the moment of change. The solution is to share all decision logic and separate only the final side effect.
"Where do we stand on the migration?" I opened the epic to answer that, and its own phase table said Not started across every phase. Phase 0, Phase A, all of it. Not started.
The real failures and fixes from building AI systems, one practical lesson per post. Get the next one in your inbox.