Engineering Blog
Practical insights on Playwright, self-healing tests, AI-driven QA, and scaling engineering teams. Real lessons from 20+ years in quality engineering.
150 posts
A correct AI answer can hide a broken trace. Use a synthetic request and a deliberately stripped context header to test whether your debugging evidence survives service boundaries.
Read articleA judge that must choose pass or fail will turn uncertainty into false confidence. Add an abstain state, calibrate it, and test the cases where the evaluator should refuse to decide.
Read article
JSON parsing is not schema testing. For AI features, QA needs to verify types, allowed values, relationships, and what the model was actually permitted to claim.
Read article
An agent can produce a polished answer while calling the wrong tool, using the wrong arguments, or exceeding its authority. Here is a practical test boundary.
Read article
A fallback can keep an AI workflow available while quietly changing the question it answers. Test the degraded path as a contract, not as a successful HTTP response.
Read article
The most expensive engineering mistakes in my systems were not hard bugs. They were reasonable implementations of questions nobody had defined.
Read articleA signal says that conditions are interesting. It does not say that the portfolio should trade. The missing layer is the decision record between detection and execution.
Read articlePage 1 of 22
Get weekly insights on AI-driven QA, engineering leadership, and automation strategies.
No spam, ever. Unsubscribe anytime.
Built with Vite, React, and Tailwind CSS.
View Source