The check is now the slow step
The slow part of most work used to be making it. Someone had to write the report or build the model, and reading it afterwards was quick by comparison.
That has reversed. AI now produces the draft almost at once. The person who has to check it and sign it still reads at the old speed. So the check has become the slow step, and work queues up in front of the people who have to read it.
The temptation is to let the speed of the machine decide how fast everyone works. People read less, or they pass the output on unread. That temptation does not remove the need to check. I still move too fast myself, and I default to results.
But a team that lets the output decide how fast it works has made a decision without noticing that it made one. It has decided that the AI's speed matters more than the team's check. Nobody wrote that decision down, and nobody owns it.
What the research says about people working with AI
Michelle Vaccaro, Abdullah Almaatouq and Thomas Malone published a review of 106 experiments in 2024. It compared people working with AI against people working alone and against AI working alone.
On average, people with AI did better than people alone. But the pair did worse than the better of the two working on its own, and the shortfall was largest on decision tasks.
The comparison does not establish that adding a person always improves on AI alone. Putting a person next to an AI does not, by itself, give you the better result of the two. The reported shortfall was largest on decision tasks. The study does not say how leaders decide.
How often people accept AI output without checking
The second piece of evidence is about habit. KPMG and the University of Melbourne surveyed more than 48,000 people in 47 countries between November 2024 and January 2025. Among employees who use AI at work, 42% said they had relied on AI output without evaluating its accuracy sometimes or more often.
Put the two studies side by side. The pair of person and AI falls short on decisions. Many employees who use AI at work say they have relied on its output without evaluating its accuracy. Neither study establishes how faster output changes a team's checking. The practical question remains where the check sits before the output is used.
AI now drafts and proposes conclusions. Sounding right is a style, not a signal.
The two decisions a leader makes
So what does a leader decide? Two decisions concern the team's work: where checking happens and who owns the outcome.
The first is where a person's check sits before an output leaves the team. Not every output needs the same check. A draft note to a colleague can go out after a quick read. A figure that goes into a board paper needs someone who opens the source and confirms it before it goes. The check should be in proportion to what rides on the output. It should sit before the output goes out, not after a customer or a regulator finds the mistake. The team can still use the speed everywhere else. What the leader decides is where the speed stops and a person reads.
The second decision is who owns the outcome. When something goes wrong, "the AI said so" cannot be the answer. A named person signed the output off, and that person owns what happened next, including the part the AI got wrong. The team needs to know who checks the output and who is responsible for correcting it.
AI can draft the plan. It cannot decide whether the plan is right. That second part stays with a person, and the leader decides which person.
I learnt that a result is not the same as progress. At Electrolux, at twenty-nine, I delivered 241 per cent growth and was dismissed within the year. A team that sends out AI output faster than it can check it is producing results. Whether it is making progress depends on those two decisions.
You can be right about the technology and wrong about the humans. And the humans will always win.
That leaves a question for leaders whose teams receive AI output faster than they can read it. Do you know how your people are deciding with AI?
What I will not claim
Neither study establishes that slowing AI output makes decisions better on its own. An output that nobody reads is no better for arriving late. What matters is where the person's check sits and who stands behind the result.
The studies are cited for their reported findings, not as evidence about my clients.
Nothing here measures anyone.
I give keynotes, briefings and workshops on this.