Everyone is excited about AI agents writing code.
Almost nobody is talking about how to verify that the code actually works.
I spent the last few weeks building with agents in production. The code looked correct. It passed the tests the agent also wrote. Then real users showed up, and three things broke that none of those tests caught.
The gap is not in code generation.
The gap is in the verification layer, and almost nobody is building that layer.
I think the engineers who figure this out will build systems that survive production. The rest will keep building impressive demos.
If you want the deeper version of this idea, these two articles are the foundation:
I'd genuinely love to know what you think.
Top comments (0)