Sergo's blog
Latest Article · · 2 min read

Deterministic Eval Harnesses for Coding Agents

Why flaky LLM benchmarks fail in CI, and how to isolate tool-use state transitions behind replayable fixtures so agent regressions surface in seconds.