Notes on production AI, the integration substrate around it, and the engineering discipline behind operational software that gets used day after day.
The test for agent autonomy isn't whether work repeats. It's whether correctness can be checked, harm prevented, errors reversed, and authority bounded.
Read the post →FMCSA's broker transparency rule rewrites 49 CFR 371.3: electronic records, 48-hour production, no more waivers. What ready actually looks like.
Read the post →Egypt's goal was erased for a shirt pull a full pitch away. Why fans cried 'FIFA corruption' within minutes, and what VAR's design actually broke.
Read the post →Why Procore and Sage 300 CRE fall out of sync: mapping mismatches, middleware failures, and workflow rules. Diagnosis guide from an independent studio.
Read the post →An independent assessment of the Procore API for two-way data synchronization: what it covers, where the edges are, and how to build sync that doesn't drift.
Read the post →For fifty years we told our tools exactly what to do. Now we tell them what we think and they infer the rest. A field guide to the shift from instruction to judgment.
Read the post →The Garden Grove blast radius explained: the ~1-mile evacuation zone, which cities and schools were affected, and why the May 2026 chemical tank couldn't be fixed.
Read the post →What happened in the Garden Grove methyl methacrylate tank emergency — why a faulty valve left no safe option, and the inspection gap underneath it.
Read the post →Five years of architecting a complex SaaS platform, and the slow disappearance of the part of me that wasn't useful to a Jira board.
Read the post →The AI panel bolted onto traditional dashboards has aged poorly. What's replacing it, and what mid-market operators should ask their software vendors next.
Read the post →A Claude integrator builds the operational substrate around Claude — eval suites, audit logs, integrations — so the model does real work safely.
Read the post →