Call me cynical, but I’m starting to think most companies don’t actually want to use AI well. They just want to be able to say they use it — so they don’t look like they’ve been left behind.
Recently, I read an Ars Technica piece about Amazon describing an internal phenomenon called tokenmaxxing: employees pushed to use AI tools as much as possible to drive adoption metrics up.
Not because it actually helps. Not because it improves the work. But because someone above them is watching those numbers. And honestly, it’s one of the most predictable things that could have happened.
Optimising the KPI, Not the Outcome
When you turn AI usage into a KPI, people don’t optimise for results.
They optimise for the KPI — and that produces absurd behaviour: using ChatGPT for things that would take thirty seconds, generating pointless documents, routing trivial tasks through AI agents just to prove they’re “adopting AI” and protect their jobs.
And then we’re told AI was supposed to reduce pointless work.
Digital Bureaucracy
Instead, in many companies it’s creating a new layer of digital bureaucracy — one that carries an enormous computing cost, which then conveniently becomes the justification for cutting headcount to fund the inference bill.
Corporate theatre. A race to the bottom.
Consumption vs. Outcome
Using a lot of AI without a clear strategy doesn’t automatically create value. It usually just adds complexity.
I use AI every day and I think it has a genuine, significant impact. But the difference between smart companies and panicking ones shows up entirely in outcomes — not in consumption metrics or usage behaviour.
And the moment you start rewarding consumption, you inevitably end up with people farming tokens instead of solving problems.
When a Metric Becomes the Goal
It’s the same old story: when a metric becomes the objective, it stops being useful. And in my view, over the next few years we’re going to see an enormous amount of completely fake AI adoption — with the real problem being that most people will realise it far too late.