lead2vibe

Leading a team in the age of AI

— it's still in your hands

Once your team is using AI regularly, the leadership task changes. The question is no longer whether people are using it, but how well.

AI makes it quick to produce documents that look complete but have not been properly thought through. Researchers call this "workslop". It tends to land with whoever has to review the work, which is often the manager.

The skills needed to manage this are ones most leaders already have: commissioning clearly, reviewing critically and setting expectations. This page sets out three risks to be aware of, five steps for managing them, and some shared language your team can use.

Start the 5 steps

Three risks to manage

Overload

AI can increase workload as well as reduce it. Researchers at the University of California, Berkeley, spent eight months studying one organisation and found staff taking on more tasks in parallel and spending more time reviewing AI output. Research by BCG found that 14% of AI users reported mental fatigue from overseeing AI, which the researchers called "brain fry". Those affected were more likely to consider leaving.

Checking burden

AI reduces the time it takes to write, but not the time it takes to check. Unless authors check their own work, that time passes to the reader. Research by Stanford University and BetterUp found that people spend nearly two hours, on average, dealing with each piece of workslop they receive. Managers carry much of this burden, because they are accountable for the outcome and spotting errors often requires experience.

Cognitive atrophy

Skills that are not used weaken over time. Research by Microsoft and Carnegie Mellon University, involving 319 knowledge workers, found that people who placed more trust in AI applied less critical thinking, while those confident in their own expertise applied more. The risk is greatest for less experienced staff, who may not yet have built the judgement that AI can appear to replace.

Step 1

The author is responsible for checking

Addresses: checking burden

More than half of employees surveyed by Stanford and BetterUp said they had knowingly passed on workslop, partly due to social pressure. Clear expectations remove that pressure.

Make it a standard that the author checks AI-assisted work before sharing it. If work arrives that needs substantial correction, it is reasonable to return it rather than fix it yourself.

Try this now

The next time you receive work that you would need to rewrite, return it and ask the author to check and revise it. Explain that this will be the standard approach.

Step 2

Form a view before using AI

Addresses: cognitive atrophy

A useful distinction is between using AI as a pilot or as a passenger. A pilot starts with their own view and uses AI to test and improve it. A passenger asks AI to produce the answer.

Pilot prompts share three habits: they lead with the author's own view, they ask for challenge rather than content, and they leave the author in control of the drafting.

Using AI as a pilot, not a passenger

Passenger prompt Pilot prompt
"Write a briefing on X." "Here is my view on X in three lines. What have I missed?"
"Summarise this report." "I think the main point of this report is Y. Am I right? What have I underplayed?"
"Write a recommendation." "I am recommending option A over option B. Set out the strongest case for option B."
"Improve this." "Where is this argument weakest? Do not rewrite it."
"Draft an email to the board." "Here is my draft. What will a sceptical board member challenge?"
"Give me some ideas." "Here are my three ideas. Which is weakest, and why?"
"Check this for errors." "List every factual claim in this so I can verify them."
"Write a strategy." "Here is my strategy on one page. Which assumptions could be wrong?"

Try this now

When you next commission a piece of work, ask for the author's own view in three lines before any further drafting.

Step 3

Keep documents short

Addresses: checking burden, overload

AI makes it easy to produce long documents. Length makes it harder to see whether the thinking is sound, and adds to the reader's workload.

Three simple expectations help:

  • State the decision or request in the first sentence.
  • Keep papers to one side of A4, with annexes only on request.
  • The author should be able to explain the paper in two minutes.

Try this now

Commission your next paper on one side of A4, and ask the author to talk you through it briefly before you read it.

Step 4

Make checking proportionate

Addresses: checking burden

Where AI-assisted work is not checked before it is shared, teams build up what we might call slop debt: checking and rework that has been deferred rather than avoided.

Some practical measures:

  • Ask authors to add a short note to each paper setting out what AI was used for, what they checked and what they are uncertain about.
  • In reviews, ask what the AI got wrong and how the author identified it.
  • Match the level of review to the level of risk.
  • Spot-check a small number of claims. If they do not stand up, return the paper.

Try this now

Add a three-line note on AI use and checking to your team's standard paper template.

Step 5

Protect time for thinking

Addresses: overload, cognitive atrophy

BCG's research found that performance tended to fall when people used more than three AI tools at once. Time spent without AI also matters for maintaining skills: judgement is built by exercising it.

Some practical measures:

  • Recognise quality and clarity rather than volume or level of AI use.
  • Limit the number of AI tools used in any one workflow.
  • Protect time free from meetings and AI tools for thinking and drafting.
  • Ask less experienced staff to draft some pieces without AI, so they build judgement.
  • Discuss a task before drafting it. A short conversation can save a long document.

Try this now

Set aside an hour this week for your team to think and plan without meetings or AI tools.

Shared language

Shared phrases can help teams adopt good habits and make it easier to raise concerns about quality.

  1. You send it, you own it.
  2. Is this workslop?
  3. Return to sender.
  4. Think first, then prompt.
  5. Bottom line first.
  6. Fits on a page.
  7. Tell me in two minutes.
  8. Show me your workings.
  9. What did the AI get wrong?
  10. Thinking time is real work.

Also useful: pilot or passenger, slop debt, and "bring me your view, not the AI's".

Glossary

Workslop

AI-generated work that appears complete but does not meaningfully advance the task. It shifts effort onto the person who receives it. The term was coined by researchers at Stanford University and BetterUp.

Back to where you were
Slop debt

The checking and rework that builds up when AI-assisted work is shared without being properly reviewed.

Back
Pilot or passenger

A pilot uses AI to test and develop their own thinking. A passenger relies on AI to do the thinking for them.

Back
Checking burden

The time and effort needed to verify AI-assisted work. It often falls on reviewers and managers rather than authors.

Back
Cognitive atrophy

The weakening of skills such as critical thinking and writing when they are regularly handed over to AI.

Back
Brain fry

Mental fatigue resulting from overseeing AI tools beyond a person's capacity. The term was used in research by BCG published in 2026.

Back

If you have not yet started using AI yourself, begin with the five vibe2lead steps. Back to vibe2lead

Sources

Some Harvard Business Review articles may require a free account to read in full.