QA × AI insights
Notes and analyses shared on LinkedIn — each card opens the full article.
- AI & QAA detection AI's right to be wrongA detection AI will get things wrong. The useful question is where it is allowed to, and to write down beforehand, as blocking thresholds, what it has no right to miss.Read the article →
- QA practiceFrom no rule to one rule too manyAn agent loses its chest and writes "never again". A regex rejects a legitimate identifier. Both rules came from an observation, not a specification — and only one of the two ever changed its mind.Read the article →
- AI & QA"The button doesn't work": rebuilding what someone already knewA one-line report costs ten to thirty minutes to rebuild, and the lost context never comes back whole. I tooled that rebuilding — the tool started out by getting itself wrong three times.Read the article →
- QA practiceThe QA who arrives at refinement is already too lateWhen the QA function does not exist yet, nobody has decided when QA enters the cycle. I chose framing over refinement — what that changes, what it costs, and what it does not fix.Read the article →
- AI & QADeterministic first, AI where it earns its placeSeven tools, the same architecture decision taken seven times: forbid the LLM from settling anything. Where I draw the line, what the rule cost me, and the case where it is not enough.Read the article →
- Test reliabilityFlaky tests: who decides a test is unstable?A deterministic score on three signals, a four-run floor, and an "unknown" answer when confidence is missing. What I learned building the diagnosis — and why the LLM decides nothing in it.Read the article →
- Security & QA"Strong" passwords: what if your acceptance criteria dated back to 2003?NIST has inverted the rules inherited from 2003: composition banned, rotation forbidden, length first. How many test suites still validate the old model?Read the article →
- AI & QAAI in QA: start with the simple tasksReviewing a user story against the DOR, writing documentation: the fastest ROI on AI is not where people look for it.Read the article →