Testing, QA, and shipping fast without breaking things.
Claude Code writes tests that debug rather than guard, mocks what it should exercise, and asserts what the code does instead of what it should do. The four failure modes, and the fixes.
Read the post →Nua's QA philosophy: AI testing that behaves like your real users, learns your product from Notion, Linear, and Figma, and runs hands-off so developers keep shipping. User-based, context-aware, exploratory.
Read the post →Six findings from peer-reviewed and large-scale studies on AI-generated tests: over-mocking, tautological oracles, coverage without targeting, and the 29.6% of passing patches that were wrong.
Read the post →From proof-testing bridges with locomotives to the 1947 moth to fuzz testing born in a thunderstorm to AI agents that use your app like a person. Five named eras, one unchanged job: prove it holds before the public arrives.
Read the post →AI writes most new code, and verification did not keep up. DORA found throughput up and stability down. Why QA is the real constraint in agentic coding, and what has to change.
Read the post →