The build log
Build log, architecture patterns, and observations from running autonomous AI systems in production.
One environment variable that gives you no-downtime production rollback, keyless local dev, and deterministic tests for any LLM-backed feature. Build it on day one, not after the incident.
Why optimistic UI breaks when a write triggers model inference: show an honest Analyzing state and invalidate the cache instead of faking the result.
AI assistants kept editing date-stamped build output instead of the src files. A .cursor/rules glob encodes the correction once so every session inherits it.
A four-line redirect fix shipped as a hotfix while a 760-line refactor rode the preview branch. Route changes by blast radius, not diff size or the release calendar.
How we replaced string-soup pages with a fluent DOM builder on Page.prototype: no framework, no big-bang rewrite, one page at a time.
A bad ID reached an internal-error screen because a late redirect ran inside the failure path. The fix replaced brittle URL assembly with a safe internal handoff and documented the remaining fallback issue.
A copy button broke with no deploy because clipboard.js and the browser both claimed window.Clipboard. A three-level fallback fixed it without a version bump.
A legacy web application had no bundler or asset manifest, so it versioned compiled JavaScript and CSS with date-stamped filenames. The convention changes the asset URL with a minimal, auditable release step.
Two days after we bootstrapped a new app, we stopped asking agents to read a 3,500-token README.
A payment gateway callback fed a raw POST field straight into a SQL WHERE clause. How trusted-caller framing kills input validation, plus the five-line boundary fix.
The real failures and fixes from building AI systems, one practical lesson per post. Get the next one in your inbox.