What this entry covers ## How eval is used in this channel's material In the captured material, evals (evaluation suites) appear alongside traditional tests as the safety layer that holds an agent-driven coding workflow together. A two-person team at Every — Kieran Klaassen and Nityesh Agarwal — runs Claude Code as a full engineering partner rather than an autocomplete, and they explicitly credit tests and evals with keeping that workflow reliable in production. [^11bdf6ed-1223-5cba-afde-fa9b4ce88712@284] [^320b90ca-d366-5cb1-9dbb-89b66cd5893b@256] ### Why evals matter when agents write the code When humans stop typing most of the code and instead plan, orchestrate, and verify, the verification layer becomes the main safeguard. The material frames tests and evals as what still holds everything together once agents build, review, and iterate on real software. [^11bdf6ed-1223-5cba-afde-fa9b4ce88712@284] [^320b90ca-d366-5cb1-9dbb-89b66cd5893b@256] - Evals are paired with tests as the mechanism that keeps an agent-heavy workflow trustworthy. - They exist so humans can shift from typing code to planning, orchestrating, and verifying. - The teams in the material run agents on production code, so evals are treated as production practice, not an optional extra.