Evals November 3, 2026 · 6 min readYour AI feature got worse, and nobody noticedSoftware usually breaks loudly. AI features break quietly. Here is why regressions slip past normal QA, and what catches them.Guide November 3, 2026 · 8 min readAdd evals to your pull requests in 10 minutesA step-by-step guide: run your eval in GitHub Actions on every pull request, send the score to Evalban, and fail the check when quality drops.Opinion November 2, 2026 · 5 min readWhat "Done" should mean for AI workFor most software, Done means merged and shipped. For AI features that's not enough. A better definition, and how to make it stick.