Jays Computer

# Jays Computer > Engineering notes, build logs, and things I learned the hard way. Written by Jay Derinbogaz. Every page has a plain-markdown version at the same URL plus `.md` — for example `/blog/how-i-removed-greptile.md`. Prefer those: they are the source the HTML is built from, without navigation or styling. Posts are written in English and machine-translated into 8 more languages (de, fr, es, it, pt, nl, pl, tr), including the URL slug. Translations are labelled as such in their markdown header and link back to the English original; quote the English version in preference to a translation. ## Posts - [Agent-written pull requests turn CI into a metered dependency](https://jays.fyi/blog/agent-written-pull-requests-turn-ci-into-a-metered-dependency.md): Agentic development turns CI from a final check into a feedback loop. Runnerhut replaces the runner with one runs-on label. Measure the result. (2026.8.25, 6 min, tooling/pricing/ai) - [GitHub contributor lists are an early startup hiring pool](https://jays.fyi/blog/github-contributor-lists-are-an-early-startup-hiring-pool.md): Pedigree and incoming applications gave me weak signals, so I started sourcing early startup hires from open-source contributors and people I already knew. (2026.8.10, 7 min, startups/hiring/open-source) - [Per-seat billing turns a 60¢ code review into $4](https://jays.fyi/blog/how-i-removed-greptile.md): You are billed per developer and consume reviews per pull request. Here is that division, plus the pull request where our AI reviewer caught one defect in six. (2026.8.8, 10 min, tooling/code-review/pricing) - [Your performance review is a memory test](https://jays.fyi/blog/your-performance-review-is-a-memory-test.md): We scored 740 pull requests by severity, weighted each by how much the roadmap cared about the component, and published the leaderboard. I had the order wrong. (2026.8.8, 13 min, engineering-management/performance-reviews/ai/gitrank) - [A static blog that translates