Twenty years in AI here. QA is not dying. What is dying is the specific flavor of QA where a person manually clicks through a checklist of features and reports whether they work. That has been dying for 10 years and AI is just accelerating it.

What is actually happening is that the QA function is being absorbed into engineering (shift left) and also evolving into something harder. Both of these are true at the same time and that is what makes this confusing.

The part most people in this thread are missing is what testing looks like when AI is generating the code or when the product itself uses AI. Traditional QA assumes deterministic behavior: you give the same input, you get the same output, you compare it to the spec. AI breaks that assumption. Models produce non deterministic output. Code generated by LLMs has different failure modes than handwritten code (subtle off by one logic, invented API calls, confident looking code that does the wrong thing). You cannot test this with the old playbook.

The teams I have seen handle this well have someone whose job is essentially adversarial validation: fuzzing AI outputs, building regression suites around edge cases the model handles badly, monitoring for drift over time, catching hallucinated function calls before they hit production. That person used to be called QA. Now they might be called an ML engineer or an evaluation engineer or just 'the person who makes sure this does not blow up'. The title is changing but the work is if anything harder and more important than before.

If you are in QA right now, the move is to learn what it means to test non deterministic systems. Build fluency with evaluation frameworks, property based testing, statistical assertions (does the model get the right answer 95% of the time, not 100% of the time). That skill set is not going away. The person who can certify that an AI feature actually works reliably is going to be worth more in two years, not less.