The case against relying on AI for test generation right now

Unpopular opinion, but I think the current hype around AI test generation is leading teams to invest in workflows that don't scale. **The fundamental problem:** Good tests encode understanding of what the system should do. AI doesn't have that understanding — it has pattern matching on your code and tests. The tests it generates test what the code does, not what it should do. **What this means in practice:** - AI-generated tests rarely catch the bugs that actually matter (requirement mismatches, incorrect business logic) - They do catch syntactic and structural bugs, which your type checker and linter already catch - They create a false sense of coverage **Where I think it actually helps:** Boilerplate reduction. If you've already decided what to test and how, AI can write the scaffolding faster. That's legitimately useful. But that's different from AI deciding what to test. Interested in pushback — especially from people who've caught real production bugs with AI-generated tests.