Why answers converge

A language model writes the next word by choosing among the words most likely to come next, given everything before. Asked a design question, it produces the answer that most often followed questions like yours in what it learned from. That is, by construction, the common answer: often a good one, rarely the one that fits an unusual constraint.

The evidence on sameness

In a writing experiment, people who could use story ideas from a generative AI wrote stories that readers rated as more creative and better written, especially among less creative writers. But those stories were also more similar to each other than stories written without AI ideas (Doshi and Hauser 2024). Each person did better; the group became more alike.

Code works the same way. If every team asks "how do I make this endpoint faster?", they all get "add a cache". When the real problem is a missing index or a slow image resize inside the request, a field full of caches is a field full of the same blind spot.

What to do with this

Not "never use the first answer". Map it instead: know what the default is for a kind of question, so you can tell when you are about to give it, and check whether your problem is the ordinary case. The next two lessons do exactly that with thirty answers.