Owning technical decisions in the age of AI

Lately, I've been noticing a rising trend where a technical decision needs to be made, and the proposal leads with the sentence: "This was suggested by Claude," followed by a list of bullet points. I don't know about you, but framing a proposal this way makes me feel like we've almost completely handed the thinking and responsibility behind the decision over to the model.

Claude, like any other LLM, is a tool. I use it, you probably use it, most engineers I know use it. But it's just one tool in your belt that you can use to research a topic. We all should know by now that models can hallucinate, and then we have a problem if we simply outsource our thinking to the model.

The model is not a co-signer

If you need to make a technical proposal, sure, go ahead and use Claude to help you get there. But the fact that you used it is not the interesting part, and honestly I am not that interested in it. What I want to know is whether the proposal holds up.

There is an important difference between using Claude to develop an idea and using it to justify that idea. Claude is a great tool to help you run experiments and find options, even identify gaps in your thinking. However, “Claude suggested it” is not evidence. This can feel like it gives you a backup, some safety net in case the idea turns out to be wrong. It doesn't, and in the end you are accountable for what you wrote there.

The same goes for code

It goes without saying that this applies to the code you push to the repo as well. Sure, use AI to write it, nobody is going to stop you. I do the same, and companies such as Google already write 75% of their code using AI at this point. But once the code lands in the repo with your name on the commit, you are accountable for it.