AI in QA without fake confidence
I use AI to accelerate API and performance work. I do not let a model decide that a release is safe.
AI is useful in QA when the task is repetitive scaffolding: drafting Bruno collections, shaping k6 scenarios, summarising logs, or proposing edge cases you then verify. It is dangerous when people treat a fluent answer as evidence.
On recent engagements I have used Cursor and ChatGPT to speed REST API automation and performance checks. The rule is simple: the assistant drafts, the human owns the oracle. Pass/fail still comes from observables — status codes, timings, data integrity, release gates.
I keep AI away from silent authority. No model output becomes a gate by itself. If confidence is low, or the domain is policy-sensitive, a person reviews. That is the same design I expect from any product that mixes automation with judgement.
The practical win is cycle time. You spend less time wiring boilerplate and more time on the paths that actually break users. The practical risk is invented certainty — pretty reports that never touched the system under test.
Use AI like a junior pair that types fast and occasionally hallucinates. Pair it. Do not promote it to release manager.
Marius Ene
Senior Functional QA · user & business focus