Claude, Please Reconstruct What I Did Today
Using an AI to reverse-engineer your own work history is a perfectly normal thing to do.
Having Claude reconstruct your git history from dirty diffs is still the most relatable developer workflow post on this entire blog.
The canonical developer workflow involves committing work as you go — small, atomic, descriptive. Meaningful history. A log someone could read and understand.
I have never done this once in my life.
What I do is ship a bunch of fixes across a bunch of projects, tab around for six hours, and then surface at some point with a pile of dirty working trees and no memory of what changed or why. The diff is correct. The story is gone.
So today I had Claude go through the diffs and write the commits for me — not one big slop commit, but actual bespoke messages per file, per logical change, reading the context and inferring intent from the code itself. It went through everything. Found the bug fix in the auth layer, the config tweak I'd forgotten about, the three-line change that was actually the whole point of the afternoon.
This is archaeologically correct. The commits it wrote are better than the ones I would have written in the moment, because it has the benefit of seeing the whole shape of the change rather than whatever I was thinking about when I made it.
There's something slightly disturbing about this — not in an AI-is-taking-over way, but in the way it implies that the narrative of "what happened" in a codebase is a story that gets told after the fact anyway. I didn't lose the work. I just hadn't explained it yet. Claude explained it for me, from the evidence.
Git history as forensics. Every commit message a reconstruction.
This has always been true. I just have a better detective now.
Counterpoints
Push back, extend the argument, or sharpen it. New counterpoints go through review before they show up here.
No approved counterpoints yet.