Commit Counts Are Not Throughput
A thousand commits in a day is a real number. It just measures the wrong thing, and buying AI coding tools on it is a mistake you'll pay for later.
A commit count is an input, not an outcome, and the whole "100x productivity" conversation keeps mistaking one for the other. The most honest version of the claim I've seen puts the tell right in the open: one practitioner's commit count went from about ten a day to nearly a thousand on his busiest day. That's a real number. It just doesn't measure what you think you're buying.
The gap is between individual commit velocity and team throughput, defect rate, and whether anyone can maintain the result in six months. Those are different variables, and only the second set shows up on a P&L. "We're going to take it from three days to fifteen minutes" beats a commit multiplier every time, because the first one names an outcome and the second one names activity.
A thousand commits a day is personal productivity theater until someone proves the codebase still holds together.
This is why vendor studies feel so clean and land so poorly. If you sell the tool, you measure the variable that shows the biggest gain, not the one the buyer should care about. Generation speed is easy to measure and flattering. Maintainability and rework are hard to measure and unflattering. Guess which one makes the slide.
And the underlying mechanism doesn't change: AI accelerates whatever you already do. Fast generation on top of a weak review path doesn't give you a faster team, it gives you more code arriving faster than anyone can reason about it. That's the vibe-coding-to-production failure, rebranded as a procurement win.
I don't know where the honest multiple lands for a real engineering org. But if the number on the contract is commits, you're optimizing the one variable that can go up while delivery gets worse.
Sources
Want to talk about this?
Get in touch