Generative AI can make a document look finished before the thinking behind it is finished. The prose is polished. The headings are in place. The summary sounds confident. Yet the recipient may discover that the facts are uncertain, the recommendation is generic, or the writer has not decided what the document is actually asking them to do.
The term “workslop” has emerged for AI-generated work that appears complete but lacks the substance needed to move a task forward. I find the term useful because it reveals something that productivity measures often miss: AI slop does not eliminate effort. It moves effort from the creator to the recipient.
The hidden transfer of work
Imagine receiving a long proposal assembled with AI. It repeats background everyone knows, uses unfamiliar terminology, and offers five plausible recommendations without choosing among them. To respond, you must determine what is true, reconstruct the author's intent, locate the actual decision, and possibly rewrite the work.
The sender saved an hour. Three colleagues may each lose half an hour trying to understand the result. The visible activity went up; the organization's progress went down.
There is a human cost as well. Receiving low-effort output can feel disrespectful, especially when the recipient is expected to supply the judgment the sender omitted. Repeated often enough, this can erode trust: Did my colleague check this? Do they understand it? Is the polished language concealing uncertainty?
This is a management problem, not a grammar problem
Telling employees to “use AI responsibly” is too vague. So is banning AI-generated text. The same tools that produce workslop can help someone question an assumption, clarify a difficult passage, compare options, or prepare more thoughtfully for a meeting.
The distinction is not whether AI touched the work. The distinction is whether a person still owns its meaning, quality, and consequences.
AI can assist with the labor of expression. It cannot transfer accountability for the message.
A practical standard for teams
I would begin with a simple rule: do not send AI-assisted work until you can explain it, defend it, and state what you need from the recipient. From there, teams can adopt a few useful habits.
- Name the purpose. Is this meant to inform, recommend, request a decision, or begin a discussion?
- Verify consequential claims. Check facts, quotations, calculations, citations, and policy interpretations.
- Add the missing judgment. Do not submit a menu of options when your role requires a recommendation.
- Edit for the recipient. Remove generic context, repetition, and language that does not sound like you.
- Make uncertainty visible. Separate what is known, inferred, and still unresolved.
- Match effort to consequence. A brainstorming note and an executive recommendation should not receive the same review.
What leaders should change
Managers should be careful not to create the incentives that produce workslop. If leaders reward document volume, rapid responses, and visible AI adoption without measuring usefulness, employees will optimize for those signals.
Better expectations focus on outcomes: Did this clarify the decision? Did it reduce uncertainty? Did the recipient know what to do next? Was the information reliable? Did AI help the employee contribute better judgment, or merely more material?
Leaders can also model the behavior they want. A short note that states a real point of view is often more valuable than a perfectly formatted briefing that avoids one.
What I am still considering
We should not turn every use of AI into a disclosure ritual. Teams need trust and room to develop new working habits. But they also need a shared understanding that pressing “generate” is the beginning of responsibility, not the end.
The most promising organizations may be those that use AI to raise the quality of thinking while protecting the attention of everyone downstream. How should your team distinguish helpful acceleration from work that merely transfers effort to someone else?