People tell me that #AI code is fine, because you can run automatic tests. But tests can only tell you if code is doing the thing you want it to do.

To know what it SHOULD do, we used to have requirements. But now requirements are themselves vibe-coded slopotypes.

People were hoping that this would get them higher velocity, but testing the requirements in production only produces waste and risk.

To test requirements safely, you need a culture of critique.

productpicnic.beehiiv.com/p/ch…

#LLM #UX

Checking an LLM's work is a systemic, not an individual, problem.

Only a culture of design critique can prevent sloppy AI-generated problem definitions.

^{Pavel Samsonov (The Product Picnic)}