Companies Rolled Out AI to Kill Busywork. Employees Invented Performative Productivity Instead.

There is a specific behavior happening in every knowledge-work org right now, and almost nobody has put a name on it.

An employee opens a document they have already read. They paste it into Claude. They ask for a summary. Then a follow-up. Then the counterarguments. Then they ask it to rebuild the whole thing as a table. Forty minutes later they have a five-page artifact, an afternoon of visible activity, and a Slack status that stayed green the entire time.

Nothing moved. That was the point.

It is the AI-era descendant of the mouse jiggler, with one upgrade. The mouse jiggler produced nothing, which made it easy to catch. This produces a deliverable, and the deliverable looks like work.

Performative productivity is the use of AI to generate visible evidence of effort rather than to complete work. The output is real. The activity is real. The purpose is to be seen rather than used. It differs from AI "workslop," which gets inflicted on a colleague. Performative productivity has no recipient at all. It exists to be counted.

The incentive was designed on purpose

Blaming employees here gets the causation backwards. The behavior follows the metric, and the metric was a choice.

Over the past two years most enterprises rolled out AI with a mandate attached and a number that was easy to instrument: adoption. Seats activated, queries run, tokens consumed. Research from BetterUp Labs and Stanford's Social Media Lab, published in HBR as "Why People Create AI 'Workslop'", found that 41% of surveyed employees said leadership encouraged them toward AI without providing real guidance on how to use it. Use it, figure it out, we'll check the dashboard.

The returns never showed up to justify the pressure. The same team's original workslop study points to MIT Media Lab research finding that 95% of organizations saw no measurable return on their generative AI investment, a gap that has only gotten harder to explain to a CFO since.

When the metric is usage, usage is what you get. And when layoffs sit in the ambient air, as they have for this entire rollout, the rational move for a worried employee is not to finish faster. Finishing faster and going quiet is how you look replaceable. The rational move is to produce more visible evidence of effort per hour. Some employees skip the performance entirely and go straight to resistance: a 2026 survey of 2,400 workers found 29% actively sabotaging their employer's AI strategy.

The tool sold as the cure for busywork turned out to be the cheapest way to manufacture it.

Workslop has a receiver. This doesn't.

The BetterUp and Stanford work described a specific failure: AI output that looks finished, doesn't advance the task, and lands on a colleague who then has to fix it. Surveying 1,150 US full-time employees, they put the encounter rate at 41%, with roughly two hours of rework per instance. They estimated the invisible tax at $186 per employee per month, which works out to more than $9 million a year at a company of 10,000 people.

Performative productivity is the reflexive version. The audience is the calendar, the activity log, and whoever reads the weekly update. The artifact never gets passed downstream, because it was never meant to be used.

That makes it harder to catch. Workslop eventually surfaces, since the person receiving it has to deal with it. Performative productivity has no receiver, so nobody rejects it. It accumulates in Drive folders, gets summarized into decks, and quietly becomes the material later decisions rest on. Oxford's Matthias Holweg and Babson's Thomas Davenport call the downstream effect "knowledge decay": the erosion of an organization's shared understanding as low-quality output compounds through a process chain until workers stop trusting internal documents at all. This is one of the ways it gets in the door.

The behavior is old. The volume is new.

None of this is novel. Visier's research on performative work found that close to half of an organization spends roughly 1.25 days a week on work that shows up rather than work employees themselves called meaningful. Slack and Qualtrics surveyed more than 18,000 desk workers and put the global average at about 32% of working time spent appearing busy.

What changed is the marginal cost. Faking a day of analysis used to require spending the day. The performance had a ceiling because the props were expensive to build. That ceiling is gone. One person with a chat window can now generate the documentary evidence of a five-person workstream before lunch.

Marketing is the most exposed function in the building

Every department has a version of this. Marketing has the worst one, for a structural reason. Our visible metric has always been output volume.

Nobody in a QBR asks how many decisions a competitive brief changed. They ask how many briefs got produced. Persona docs, messaging frameworks, content calendars, brand voice guidelines, the audit of the audit: plenty of these were performative long before generative AI arrived. What AI removed was the last thing keeping them honest, which was that they took a week to build and somebody had to feel that week.

So the function can now produce unlimited proof of work and almost no proof of value, in the same year the CFO started asking which one he is paying for. That is an uncomfortable place to stand while the Fortune 500 deletes the CMO title.

The performance is already failing

The strategy has a flaw that should retire it on its own. It doesn't work.

The same research found that obviously AI-padded output damages how the sender is perceived. Around half of those surveyed rated colleagues who sent it as less creative, less capable and less reliable than before. Forty-two percent found them less trustworthy. Thirty-seven percent found them less intelligent.

Colleagues can tell. Unedited AI prose is now the most recognizable register in corporate life: the three-part structure, the confident hedge, the summary that restates its input at greater length. Anyone can spot it, which means the artifact built to signal diligence signals the opposite to everyone who opens it. Elena Verna made a version of this argument about AI confidence theater eroding trust at the brand level. The same mechanic runs inside the building, between coworkers, every day.

What leaders should change

The instinct will be to police the tool, with usage audits, approved prompt libraries and an AI governance council. That repeats the original error, which was measuring activity in the first place.

Three changes do more.

Change what gets reported. Stop asking teams what they produced and start asking what they decided, what they killed, and what moved as a result. An artifact that changed no decision shouldn't appear in an update at all.

Put a human name on the judgment behind every deliverable. Not who generated it, but who will defend it in a room. Once ownership is explicit, volume falls and quality rises without a policy.

Delete the standing deliverables. Most orgs carry four or five recurring documents nobody reads. Those are the ones getting inflated right now, because they are the safest place to look busy. Cancelling them costs nothing and removes the stage.

Performative productivity is a feedback problem, not an integrity problem. People optimize for what they are graded on, and most companies still grade effort. AI didn't introduce that flaw. It made it cheap enough to finally see.