What gets measured gets gamed
On AI adoption metrics, productivity theater, and warning your best people for the wrong reasons
At brunch this Easter, between hunting for eggs and eating too much chocolate, a friend who works in recruitment at a big tech company told me a story. One person on the team had gotten a warning for not using AI. The same person also happened to be the highest performer on what their actual job was.
The reason was one most companies can relate to: management needed to implement AI, fast. Their course of action was to make LinkedIn courses mandatory, start tracking of everyone’s AI usage, and make this part of their performance reviews. Unsurprisingly, this also caused some people to use their personal accounts instead, to avoid being monitored, and if they left, they could take their own AI, loaded with their personal context, with them. Not the first time I heard a version of this story, and I’m sure it won’t be the last.
In the same spirit, tokenmaxxing is a thing now. At some companies, like Meta and OpenAI, burning a lot of tokens (basically the price of requests) can give you a placement on the internal leaderboard. Not sure how that makes you better at your job, but yey for you, I guess.
* * *
There’s something called Goodhart’s Law, which states that when a measure becomes a target, it ceases to be a good measure. It creates an incentive to game the system instead of focusing on the underlying goal. I think that’s something worth reflecting on. What behaviors are your targets actually driving?
To me, it feels a lot like a version of productivity theater. Busy work, performing actions that look and feel productive, but having little to no real impact on your business.
Using AI has made me more productive. It has also made me feel more productive, even when I’m not actually producing anything of value. I know this pattern well — I’ve been trying new productivity tools and systems as a way of avoiding real work for years, but as I wrote here, I eventually saw through my own bullshit and have worked hard on rehabilitating myself.
The output you’re able to deliver in seconds is intoxicating, though. You can do so much, vibe-code multiple websites at the same time, have an agent write you custom Harry Potter fanfiction, and build a relationship with your perfectly agreeable AI companion. No wonder people are getting AI burnout.
* * *
Years ago, I implemented time reporting at an agency. There’s a common temptation to track everything: client meetings, team meetings, client email, all the way down to toilet breaks. Instead of being granular and doing what everyone else did, I considered what actually mattered; understanding how much time was spent on each client project and identifying overtime. The team was all type A people, great at what they did, but also prone to working past midnight and responding to every client email within minutes. And with most people working remotely, a brewing burnout wasn’t as obvious as it would have been in an office.
So we created a simple structure. You could report on client projects or internal time. That’s it. No requirements for billable hours, no detailed tracking of what you did with the time. It took some time to get everyone into the habit, but eventually, it gave us the insights we needed. What clients were profitable, where did we spend more than we earned, and what team members needed to adjust their workloads. It was used to make better decisions, not to argue about details.
* * *
I get that companies need everyone onboard the AI train to avoid being left behind. But if your best performer gets a warning while the person gaming the leaderboard gets a promotion, I'd say you're not measuring what matters. You're measuring what's easy to track.
✨ If you’re curious what I do beyond writing here, I’m a project and operations consultant, helping teams restore their sanity with better ways of working. More at sannastefansson.com