Many tech leaders are pushed to use AI to do more with fewer people. They are trying to adapt, use AI efficiently, and improve their team’s knowledge of AI. This shift comes with many challenges.
One awkward situation arises when a tech leader assigns a task to an employee and gets back something they could have generated themselves in under a minute. Something the employee clearly didn’t think through. When the leader tries to discuss it, it becomes evident that the employee doesn’t understand the topic and has no opinion of their own. The work was outsourced entirely to AI.
Yet if the leader complains, the employee might be confused: aren’t they supposed to use AI?
Let’s face it: a truly AI-native developer won’t make a mistake like that, because they know how to use AI properly. The problem starts with developers who don’t know how. For them, leaders now need to define what inaccurate use of AI looks like.
The value of human knowledge and judgment
When a worker researches a new topic, they learn about it — enough to draw their own conclusions. They build their own expertise and the company’s knowledge. Of course, those conclusions won’t always be the right ones for the project or the business, but they give the employee a position to defend, and the leader and colleagues something to build on.
That’s why an employee who does the research and gets it wrong can be more valuable than one who skips it and can’t back up what the AI says. The first is more likely to notice when something isn’t working, and has acquired the knowledge to understand why. The second has no way to tell (remember, AI can reach the wrong conclusions too).
As entrepreneur Mark Cuban said on The Diary of a CEO podcast, the major LLMs won’t all give you the same answer. You still need the skill to ask the right question, compare what each source says, “and come to a conclusion that you think is in your best interest.” Even when AI can write a business plan and build the agents to run it, he said, “I’ve got to be the one to say, this is right, this is wrong.”

“Workslop” is a leadership problem
Researchers Kate Niederhoffer, chief scientist at workforce development platform BetterUp, and Jeff Hancock, professor of communication at Stanford, coined the term “workslop.” They define it as AI-generated work that looks polished but doesn’t advance the task, and note that a key aspect of workslop is shifting the burden of cognitive work onto the recipient.
In a survey of 1,150 U.S. full-time workers, 53% admitted they had sent workslop. In a podcast, Hancock says the real figure is probably higher.
But the researchers evaluate that workslop isn’t always the result of laziness. Workslop “is more of a symptom that there’s a problem in the organization,” says Hancock. “And so if that’s true, then it’s a leadership problem.”
The professor sees two main reasons behind workslop: “One is AI mandates that are really general, ‘Hey, you need to use AI. We’ve just spent all this money. You better use it.’ And the second is, ‘Because we have given you all this AI, you should be able to do more work.’”
Hancock connects the issue to the “J curve” of new technologies. Productivity dips at first, then rises as people learn to use the tool well. Workslop is part of that initial decline before the gains arrive.
The cost of workslop
Workslop comes with a tangible productivity cost. According to the researchers, participants who received workslop said it took about two hours to deal with it: first, they have to detect it; then they have to decide what to do — talk to the employee and ask them to redo it, or redo it themselves. For the company, that typically means two hours of work wasted.
Using participants’ self-reported salaries, the researchers estimated that, for a company of 10,000 employees, workslop costs USD 9 million a year.
What should leaders do?
Here are 9 tips to help leaders fight workslop.
Be specific
“Use AI more” is vague. Instead, you can say, for example, “Use AI to learn the subject, draft first-pass summaries, then verify them” — or anything that applies to your project’s needs.
Establish ownership and accountability
AI is more than welcome, but the author owns the content. Say it explicitly, because many employees, particularly the junior ones, assume that if AI produced something, AI is responsible for it. It doesn’t work that way. If you drive a car and hit someone, you are still responsible. The car is a tool, and so is AI. Whoever uses it is responsible for the result.
A single line in your AI guidelines can set the expectation: “AI is welcome, but you are accountable for everything you submit under your name. If you can’t explain it, don’t send it.”
Ask for the reasoning
A leader can ask questions to understand the reasoning behind the output:
- Which part of this could go wrong?
- What would change your recommendation?
- What did the AI get wrong or leave out, and how did you check?
- Which source would you trust most here, and why?
An employee who did the thinking should be able to answer these questions.
Train people on how to use AI well
Many were never shown. Cover how to verify claims, how to spot problems, and how to give the AI enough accurate context without overloading it. Too little context makes the output generic; too much noise degrades its quality.
Encourage skepticism, too. Employees can and should push back with follow-up questions like “What are the alternatives?” or “What’s the strongest argument against this?” AI tends to agree with whoever is asking, which is a risk. They can check elsewhere. Stack Overflow is still there.
Show what good looks like
You can compare two outcomes of the same task. Say a developer is asked to compare two tools:
- Bad version: The developer pastes an AI summary with a recommendation. When asked why, the answer is “because AI said so.”
- Good version: The developer uses AI to learn the subject, identify evaluation criteria, and summarize documentation. Then they test both tools against a real use case, flag claims they can’t verify, and make a recommendation based on their conclusions. (The same approach can work across a team: several people can use AI collaboratively, each checking and building on the others’ findings.)
Match the task to the expertise
Would you call someone a doctor because they can diagnose with AI? Of course not. They still need the expertise to tell whether the AI-generated diagnosis is correct.
Similarly, if someone without a proper coding background is asked to solve a development task with AI, they won’t be able to tell whether the result is correct or risky.
Assign tasks to people who can validate the result, and if work does go outside someone’s field, have an expert review it. Think of the developer as an architect: they lead the work and delegate tasks to AI agents selectively.
Create the conditions for honest answers
Psychological safety can help employees speak up or admit uncertainty instead of hiding behind AI-generated work. Employees tend to work best when they feel confident that their team is capable and that feedback will be given constructively. That also gives them the confidence to give feedback constructively if they receive workslop.
Lead by example
Be open about how you use AI yourself. Say what you use it for, where it lets you down, what you double-check. Employees copy what they see. If your own habits contradict your guidelines, the policy won’t stick.
Give people time to adapt
AI saves time. Just don’t automatically raise workload or output expectations immediately, because rushed people may skip the thinking.
Subscribe to our newsletter
Join 30,000+ HR professionals and tech leaders: subscribe to the TechTalents Insights newsletter and get our best articles and interviews, completely free.

