Let’s talk about the AI assistant that’s supposed to make you faster — and the report it’s writing about you while it does.
There is this moment in every office that rolls out an AI copilot. The tool drafts the email, summarises the meeting, writes the first version of the report. People feel clever. The machine feels like leverage.
And then the manager gets a dashboard.
Not a dashboard of outcomes. A dashboard of usage. Who’s adopting the tool. How often. Which prompts get accepted. Which suggestions get rejected. Which employees are “resistant” because they keep editing the output instead of taking it.
Same keystroke, three directions at once. It helps you do the job. It feeds the model. And it writes a line in the file your employer keeps about you. One interaction, three owners.
That’s the pattern worth naming, because it’s being sold as pure productivity. It isn’t. It’s a measurement regime wearing a helpful face.
Now the incentive part, because nobody here is evil. The vendor sells to the company, not to you. So the product’s real features are the ones the buyer wants: adoption metrics, activity signals, a clean way to tell the board the AI investment is being used. The buyer wants visibility. The user wants help. Same tool, two different customers, and only one of them pays.
You know where this ends. The dashboard becomes the job. Once there’s a score, people optimise for the score. They stop editing the output, because editing shows up as friction. They accept the generated version, because acceptance is the behaviour the system rewards. Work starts serving the metric — the oldest corruption in management, now running at keystroke resolution.
And here’s what almost nobody checks. The vendor sits underneath both sides. Your employer sees your usage. The vendor sees your employer’s entire workforce, aggregated — every prompt, every correction, every hesitation. The vendor is not neutral infrastructure. The vendor is the only actor in the room who can see the whole game.
So the operator lesson, for anyone deploying this stuff and anyone working inside it.
Before you roll out a copilot, ask four questions and write down the answers. Who owns the telemetry? Who can see it? How long does it live? And does it train the model? If the answer to the first three is “the vendor” and the fourth is “yes”, you haven’t bought a productivity tool. You’ve bought a surveillance system that happens to draft emails.
If you’re an employee in a company running one of these, the question changes. The thing to ask for is the same thing you’d ask for any personnel file: show me what the machine says about me. See the report. Read the score. Correct the record.
That should be the rule of this entire era. Any system that judges a human must be visible to that human. Not just under regulation — in the product itself. The right to see the report the machine writes about you is the new frontier of workplace consent, and it’s the difference between a tool that helps and a tool that rules.
The machine that makes you faster shouldn’t get to decide what your work is worth. That’s a human call — and it’s the one we keep giving away.