Let’s talk about the AI rollout that reported ninety percent adoption — and couldn’t tell you if a single job got done better.
I keep seeing the same press release. “We rolled out AI to our whole team. Adoption hit ninety percent in the first quarter. Estimated hours saved: thousands.”
Every one of those releases is quoting the same two numbers. Seats activated. Prompts logged.
And almost none of them can answer the question that actually matters: did the work get better? Did the invoice go out faster? Did the rework rate drop? Did the customer wait a day less for an answer?
Because here’s the uncomfortable truth about the enterprise AI rollout. The number that gets reported up the chain is the one that’s cheap to count.
Usage is cheap to count. A login is a login. A prompt is a prompt. You can put that on a slide in an afternoon.
Outcome is expensive to count. It means measuring the work before AI touched it, then measuring it after, and having the discipline to admit when nothing actually changed. Almost nobody budgets for that.
So the incentive does what incentives always do. When the visible metric is adoption, everyone optimizes adoption.
The manager tells the team to use the tool so the number looks good. The team pastes a few prompts in to hit the target. The CIO reports the adoption to the board. The board nods. And somewhere in that tidy chain, the actual purpose of the tool — doing the job better — becomes optional.
Now follow the human. Not the exec. The person who was already doing the work.
They’re not a user in this story anymore. They’re a row in a dashboard. Their job security quietly becomes tied to whether they can show they’ve been “activated” — not whether their work is good.
That’s the part the press releases skip. The rollout that’s measured by adoption doesn’t make work better. It makes people perform being productive for a system that only reads one number.
I built software long enough to know the tell. It’s not the dashboard. It’s the before.
Ask anyone running a rollout: what was the baseline? How long did that task take in May, before the tool? What was the error rate? What did a customer actually wait for?
If they can’t answer, they haven’t adopted AI. They’ve staged it.
And the cost of staging it isn’t neutral. Because the company that fakes its way through the first rollout — that convinces itself the tool works because the numbers are green — that company builds its next decisions on a lie. It lays off people based on “hours saved” that were never saved. It expands a tool that never worked. It trust-falls into the machine on a dashboard that measured nothing.
The fix isn’t harder tech. It’s a harder question, asked before the rollout starts: what is this supposed to make faster or better or cheaper, and what’s the number that proves it?
Not seats. Not prompts. The invoice. The rework. The wait. The thing a real customer would feel.
Because a tool that makes a broken workflow faster doesn’t give you innovation. It gives you faster waste — and a board that believes it’s innovating because someone gamed the adoption number.
We’re still human. So let’s stop measuring whether people touched the machine, and start measuring whether the machine touched the work.
You’ve got this.
⚠️ File-mutation verifier: 1 file(s) were NOT modified this turn despite any wording above that may suggest otherwise. Run git status or read_file to confirm.
• /tmp/__noop__ — [patch] Failed to read file: /private/tmp/noop