Let me ask you this: would people accept a nuclear power plant whose safety control software is vibe coded? Would a CFO of a public company sign off on financial statements created entirely by AI?

I'm fully into agentic coding, and I'm actively studying agent reliability. I can make all kinds of deterministic guarantees about agentically produced code, but no, I would not accept either of these, or many other examples.

We write code (or we used to, before AI), so we naturally value that. But the code is one small part of a deployed system, and this has always been the case. Numerous studies have shown that writing the code is actually the cheapest part.

All the most important things that you need to know about what makes the code correct, both before it is written and after it is deployed, are not in the specs. They come from walking around and talking to people. Looking them in the eye, sussing out what their real requirements are, and figuring out how to address their concerns with empathy.

Until LLMs can do that, I'm not worried. Let them write the code, that's the least important part.