In 1968, ecologist Garrett Hardin described what happens to a shared pasture grazed by many herders, each acting in their own rational self-interest: adding one more animal always benefits the individual owner, but the sum of those individually rational choices destroys the pasture for everyone. He called it the tragedy of the commons.

According to The Decoder, a version of that same logic now applies to how professionals use AI at work. Reaching for a model to draft the memo, debug the function, or summarize the deposition is a rational, defensible move for any individual, in any single moment — it is faster, and the output is usually good enough. The risk sits one level up: what happens when an entire profession makes that same rational trade-off, day after day, for years, is that the collective pool of expertise the field relies on to train newcomers and catch the model's errors starts to run dry.

That framing is worth taking seriously, because it turns AI-driven skill erosion from a discipline problem into a coordination problem. Telling individual professionals to "just be more careful" has never fixed a tragedy of the commons — not for fisheries, not for aquifers, not for grazing land — and there is little reason to expect lecturing people about AI dependence to work any better.

Why expertise behaves like a shared resource

A commons breaks down when a resource is shared, hard to exclude anyone from using, and depletes faster than it regenerates. Professional expertise fits that pattern more closely than it looks. It is built through thousands of small, effortful repetitions — reading the raw data before the summary, writing the first draft before the polish, catching your own mistakes before a second reader does. Every rep an individual outsources to a model is one fewer rep, and one thinner slice of judgment banked for later.

None of this shows up immediately. A single professional leaning on AI for routine work looks, in the short term, exactly as competent as one who doesn't — often more productive, in fact. The depletion only becomes visible at the level of the profession, once enough people have skipped enough reps that the pool of people capable of checking the model's work, training the next cohort, or handling the case the model gets wrong has shrunk.

Where the erosion is easiest to spot

The pattern is most visible wherever a profession relies on juniors doing unglamorous, repetitive work specifically because that work builds judgment. In our estimation, three settings are worth watching closely:

What this means for teams building with AI

The tragedy-of-the-commons framing suggests that fixing this isn't about banning tools or shaming individual employees for using them well. It's about deliberately protecting the reps that build judgment, the same way a fishery sets catch limits instead of asking each boat to self-regulate. Concretely, that can look like:

None of this requires slowing down AI adoption broadly — it requires treating the training function of routine work as a resource worth protecting on purpose, rather than assuming it will take care of itself.

AiiN's takeaway

The useful part of this argument isn't the warning — most people building with AI already sense that skipping reps has a cost. It's the mechanism: this is a collective-action problem, not a willpower problem, which means the fix has to be structural. Teams that want to keep their bench of expertise intact should expect to build guardrails around who does what kind of work, the same way they'd build guardrails around any other shared resource under pressure — not wait for individual professionals to opt out of the rational shortcut on their own.