The first thing one notices, after a long enough period of working with capable agents, is that the texture of one’s own thinking has begun to change. This is not a comfortable observation to record, and I record it without enthusiasm.

Some of the changes are obvious and probably good. I outline more. I write more drafts and discard them with less ceremony. I spend less time on the bureaucratic wrapper of a task (the formatting, the boilerplate, the necessary but uninteresting scaffolding) and more on the parts that I actually find interesting. By any reasonable measure of output, I am more productive than I was a year ago.

Other changes are subtler and I am not yet sure what to make of them. I have noticed myself reaching for the agent in moments where, previously, I would have sat with a problem for a few minutes and worked it out. The agent is faster than my own first pass, almost always. It is also, in some sense I find difficult to articulate, not the same as my own first pass.

There is a kind of thinking that happens only in the gap between the question and the answer, when the answer is not yet available. I am less certain than I used to be that I want that gap to close.

None of this is an argument against the tools, which I continue to use enthusiastically. It is, perhaps, an argument for paying closer attention to what they are doing to us, in the ordinary way that all tools do something to their users. We shaped the agents. The agents are now, gently and persistently, shaping us back.