Reinvention Roadmap
Workslop has become a recurring theme in our FlipWork Sprint sessions with executives.
The frustration usually sounds familiar: someone sends over a document that looks polished, organized, and convincing. Then the person receiving it starts pulling at the thread and discovers how much work remains. The assumptions need to be examined, the numbers need to be verified, and the recommendation is buried. What looked like a completed handoff turns out to be the beginning of someone else’s work.
This is cognitive debt: thinking deferred by one person that eventually has to be done by someone else.
The person who created the work often feels genuinely more productive. A report that once took two hours now takes twenty minutes, and from where they sit, AI delivered exactly what was promised. But if a colleague then spends an hour making sense of it, correcting it, or reconstructing the reasoning behind it, the organization hasn’t eliminated that work; it has just moved it.
BetterUp Labs and Stanford’s Social Media Lab put numbers behind this. In a study of 1,150 full-time U.S. desk workers, 40% said they had received AI-generated work that looked polished but lacked the substance to move anything forward. Recipients spent nearly two hours dealing with each instance. At scale, BetterUp estimates the cost reaches roughly $9 million annually for an organization of 10,000 people.
But what the numbers reveal is how we define productivity in the first place.
Most of the conversation about AI at work still focuses on production: how much faster can we draft the proposal, how quickly can we summarize the research, how many presentations can a team now create. None of those measures give us the full picture.
A piece of work rarely ends with the person who created it. Someone has to review it, communicate it, or put it into action. The quality of that handoff matters as much as the speed of production.
And this is where workslop becomes more than a problem with inaccurate AI output. A document can contain perfectly accurate information and still create cognitive debt: filled with relevant data but offering no interpretation. Nothing is technically wrong, yet the recipient still has to complete the intellectual work.
We’ve always had mediocre memos and poorly considered presentations; AI didn’t invent unfinished thinking. What’s changed is how easy it is to make unfinished thinking look finished.
Until recently, producing a sophisticated twenty-page report required enough effort that the artifact itself conveyed something about the work behind it. That relationship is weakening.
We can now generate the structure, language, formatting, and apparent confidence of professional work in minutes. Which means polish tells us less than it once did. The more important questions are whether someone understood the material, exercised judgment, tested the claims, and decided what the work means in this particular context.
Follow-up research from BetterUp and Stanford points to something leaders often miss: workslop is more likely when people are already overloaded, when leaders are encouraging AI adoption without establishing clear expectations, and when teams lack clarity on what responsible use actually looks like. When employees hear both “use AI” and “you should now be able to produce more,” the appearance of productivity can start to substitute for useful work.
Leaders need to examine the incentives they are creating. If we celebrate faster turnaround and higher output without noticing the rework accumulating downstream, we may be optimizing one person’s productivity at the expense of everyone else’s.
A better measure would follow the work farther.
How much total human effort did it take to get from the initial task to a useful outcome? How often did the work come back? How much clarification did it require? Did another meeting become necessary because the recommendation was unclear? Those costs are harder to see than the twenty minutes saved at the beginning, but they are no less real.
The opportunity with AI is to remove the work that doesn’t require a person’s best thinking so they have more capacity for the work that does. The organizations that learn to make that distinction will get far more from AI than those that simply produce more.
A faster first draft may be useful, but it’s an incomplete measure of productivity. Follow the work through the organization and account for the clarification, verification, and revisions it creates before declaring the efficiency gain real.
As AI gets better at producing polished artifacts, appearance becomes a weaker proxy for quality. The better test is whether the recipient can understand the reasoning, trust the important claims, and take the next step without having to reconstruct the sender’s thinking.
Encouraging people to use AI is easy. Establishing clear responsibility for the work produced with it is harder, and far more important. Teams need to know what standard a piece of work must meet before it moves downstream.
Choose the kinds of work where workslop would create meaningful consequences and agree on what must accompany them. For an analysis, that might include the conclusion, the evidence, the assumptions, and the specific decision or action required from the recipient. The standard should be simple: Could someone use this without having to finish my thinking for me?
You probably don’t need a new productivity dashboard. Start with your managers and ask where work repeatedly comes back for clarification, or where meetings are being added because documents don’t make the decision clear. Those patterns will show you where the apparent efficiency of AI is being offset by downstream effort.
Leaders have to model the handoff they expect from everyone else: read the work, decide what you think, explain why you’re sharing it, and identify what you want the recipient to do with it. AI can accelerate the preparation, but seniority shouldn’t become permission to outsource your unfinished thinking to the team.
Before sending consequential AI-assisted work, add these five lines at the top of the document:
My recommendation: What I believe we should do.
What matters most: The considerations driving that recommendation.
What I verified: The claims I checked rather than relying on the generated output.
What remains uncertain: What still needs investigation or validation.
What I need from you: The decision or action I’m asking the recipient to take.
This takes a few minutes. It can save the next person considerably more.
The Hidden Cost of Workslop by BetterUp Labs and Stanford Social Media Lab
The original research is worth reading because it moves the conversation beyond whether AI-generated work is good or bad. It examines what happens to the recipient and shows how a seemingly small shortcut creates lost time, frustration, and reputational damage elsewhere in the organization. Roughly half of people who receive workslop view the sender as less creative, capable, and reliable afterward. That’s a trust problem, not just a productivity issue.
HBR IdeaCast: The Hidden Causes of AI Workslop and How to Fix Them
This conversation with BetterUp chief scientist Kate Niederhoffer and Stanford professor Jeff Hancock looks upstream at the organizational conditions that make workslop more likely. The discussion complicates the easy conclusion that employees simply need to use AI more responsibly. Expectations, workload, trust, and leadership signals all shape how people use these tools. That’s an important distinction for any leader trying to change the behavior without examining the system producing it.
Where is AI moving unfinished thinking to someone else in your organization?
Until next time...stay curious!
Cheers,
Nikki
© 2026 NikkiBarua.com
Privacy Policy