AI Made Motion Cheap. It Never Touched Progress.

So activity goes up. Commits and drafts, generated artifacts of every kind, all up. And a team can watch that dashboard and feel like it's moving faster, because it is. It's just moving, not necessarily toward anything.

This isn't a uniform problem. It shows up hardest in exactly the kind of work Nathan's own numbers say is growing fastest: knowledge work without a clear finish line.

A developer fixing a bug has a test suite. Pass or fail is not a matter of opinion. Motion and progress collapse into the same thing, because the validation step is automated and instant.

A marketer drafting ten campaign angles has no such test suite. Neither does an analyst generating ten different takes on the same dataset. Someone still has to read all ten, decide which one is actually right, and defend that choice. AI made the drafting free. It left the deciding exactly as expensive as it always was.

That's the 20%-of-Codex-users-and-growing-3x-faster group Nathan mentioned. It's not that non-developers are misusing the tool. It's that they're the group most exposed to a trap that only bites when the finish line is fuzzy.

If activity volume is the wrong metric, the right one isn't complicated; it's just less flattering to look at.

Progress is whatever survives contact with the outcome you actually wanted. A pull request that passes review and ships. A forecast that still holds up next quarter.

None of those are counted by "messages sent" or "drafts generated." All of them require someone to have been prescriptive about the target before the AI ever started producing anything, exactly the discipline Nathan was describing as the part that got harder to skip and easier to skip at the same time.

Easier to skip, because motion feels like enough now. Harder to skip, because when a team full of cheap motion ships nothing that survives contact with reality, the gap shows up eventually, just later than it used to.

The tooling Nathan's team built made it possible to generate ten times as much. It didn't make it possible to want ten times as many different things. That gap is where the trap lives, and it's not going to show up on a usage dashboard.