Posts

• meta · 2026-06-28 The real cost of coordination files wasn't disk — it was agent context pollution We moved cross-repo coordination from files to GitHub Issues. Disk wasn't the problem. The real cost: every grep/ls pulled those files into the agent context window and burned tokens — a tax an agent pays every turn.
• infra · 2026-06-28 I set helm upgrade --timeout 10m — so why did it hang for 25 minutes? A CI helm deploy hung 25 minutes and locked the release. Why didn’t --timeout 10m kill it? helm’s --timeout bounds only the wait/hook phase, not the apiserver hang before it. The real backstop was a CI job timeout.
• meta · 2026-06-28 My review tool could approve nothing — then it caught me redteam's one-off review could never return APPROVED — a dead gate. After I fixed it, the tool caught a flaw I'd just defended as an 'improvement.' An author's certainty can't cross another provider's review.
• meta · 2026-06-27 Build in studying — learning while you build, how non-engineers make vertical AI Vertical AI shrank the distance from idea to shipping, not skill acquisition. Build with AI and even domain learning speeds up — a non-traditional career becomes raw material, though not everyone makes it.
• meta · 2026-06-26 Spend fewer tokens, spend more — both companies are half right 'Token efficiency' and 'tokenmaxxing' look opposite, but both measure token volume — the AI era's overtime hours. The real axis is productivity on the bottleneck: set a generous cap, then judge results, not token count.
• meta · 2026-06-24 The model I trusted most led me into a dead end Four stages of migrating a photo-preprocessing pipeline. The model I trusted most recommended a dead end; the answer was in none of the models but in moving between them. Automating that cross-check became redteam.
• meta · 2026-06-22 The heart of loop engineering is verification — and the verifier must be a different model Prompt → context → harness → loop engineering. The 2026 axis is the loop; its core is the verifier. A model judges its own work poorly — verify with a different agent (canon); this post argues for a different provider.
• meta · 2026-06-22 Who does AI replace? Job levels are dead — what's left is the filter 'AI cuts juniors first' is the wrong axis. Survival isn't a level — it's knowing what, how, and why to ask AI, and seeing when its easy answer is wrong for your problem. And AI tends to just agree with you.
• meta · 2026-06-21 How to reject a feature — an AI pair cut its own roadmap, on the record Deciding not to build a feature is engineering too. redteam shipped 1 of #37's 3 steps and rejected 2 by design — kept as a decision record with guardrails and revival conditions, not silently dropped.
• meta · 2026-06-21 Using AI to write your résumé isn't cheating — so what do you read instead? Is using AI to write your résumé cheating? People have always padded theirs. A single résumé is forgeable no matter who wrote it — so don't judge by it alone; read the timestamped record instead, and hire by that.