Leading a team through AI is mostly a leadership job, with a thin layer of technology on top. The tools are the easy part. What changes is how you set direction, how you judge work you did not do yourself, and how you keep people confident while their jobs shift under them. Get those three right and the technology follows. Get them wrong and no tool will save you.
I run a one-day course on exactly this, and the thing that surprises leaders most is how little of the day is about AI. Here is what I have seen change, and what to do about it.
You have to decide, earlier and more often
AI compresses the gap between an idea and a first version. A draft report, a working prototype, a rough analysis, all arrive in minutes rather than days. The bottleneck moves to you. Which of these do we pursue, which do we bin, and on what basis.
Leaders who thrive here get comfortable making more calls with less ceremony. They set clear criteria in advance, so the team can filter before anything reaches them. Those who struggle try to review everything and drown.
In practice this means writing down, once, what a good idea looks like in your business. Who it serves, what it costs, what it would need to be true, what we would stop doing to make room. Then the team scores against that before anything lands on your desk. Your job becomes the final call on the few that clear the bar, rather than the first read on all of them.
You judge the output, not the effort
For most of your career, effort was a proxy for quality. Someone who spent a week on it probably did it properly. That link is gone. A polished-looking piece of work can take four minutes and be wrong in ways that are hard to spot.
So the leadership skill shifts to inspection. Can you tell good from plausible. Do you know where the tool tends to fail in your domain. Do your people know to check the numbers, the citations and the edge cases rather than trusting the fluency. This is where quality lives now, and most teams have not been taught it.
A simple habit helps. For any AI-assisted piece of work that matters, ask the person who produced it three questions. What did you check. What could be wrong. Where did the facts come from. If they cannot answer, the work is not finished, however good it looks. Make that the norm and the quality problem largely disappears, because people start checking before they are asked.
You own the confidence of the room
People are worried, even the ones who say they are not. PwC’s latest survey of 4,454 chief executives found 56% had seen neither revenue growth nor cost savings from AI in the past year. That gap between the hype and the results lands on your team as anxiety and cynicism in equal measure. Half the room thinks the tool will take their job, the other half thinks it is a waste of time, and both halves are watching you for a signal.
The honest move is to say what you know and what you do not. Which roles will change, which will not, what you expect people to learn, and what you will do to help. Vague reassurance makes it worse. Specific, visible investment in people’s skills makes it better.
What that sounds like in a team meeting. ‘We are going to change how we handle the weekly reporting and the first-draft proposals. Those two things. Nobody’s role is going. The hours we save go into client work. Everyone gets a half-day course on using the tools safely, and I am doing the leadership one. If you are worried, come and talk to me, and I will tell you what I actually know.’ That is specific enough to believe.
You redesign the work, not just the tools
McKinsey’s 2026 research found that the organisations seeing real returns were far more likely to have fundamentally redesigned how work flows, rather than dropping AI into the old process. Nearly three quarters of the high performers had done so.
That is a leadership task. Nobody else in the business has the standing to change how the work is actually done. If you leave the process alone and add a tool, you get the same output slightly faster and a lot of frustration.
Take a proposal process. The old version has a salesperson writing a 20-page document over three days, a manager reviewing it, and a director signing it off. Dropping AI in means the salesperson writes it in three hours and everything else stays the same. Redesigning it means the salesperson writes a tight brief, the tool produces a first draft against a house template, the salesperson spends their time on the three sections that actually win deals, and the manager reviews only those. The second version is faster and better. The first is only faster.
You set the guardrails before you need them
Teams will use AI with or without a policy. The question is whether they do it safely. A short, plain set of rules on what data can go where, what must be checked by a person, and who is accountable for what the tool produces, removes most of the risk and most of the fear.
The rules fit on a page. Which tools are approved. Which categories of data may go into them, and which may never. What must be reviewed by a named person before it reaches a customer, a regulator or the board. Who owns each AI use in the business. That page, taught once and enforced, is most of AI governance for a mid-sized firm. Our AI governance work is built around exactly that, light enough to actually follow.
The common mistakes
Delegating it to IT. The decisions are about work, people and risk, which are yours.
Waiting for the perfect policy. Staff are already using the tools. A page of rules today beats a framework next year.
Measuring usage instead of results. Rewarding people for using AI gets you more AI, not better work. Reward the outcome and let people choose the tools.
Pretending nothing changes. Some roles do change. Saying so, specifically and early, is kinder and more credible than reassurance that turns out to be wrong.
Where to start
Pick one process your team runs every week, redesign it with AI in the loop and a person accountable for the result, and run it for a month. Measure the hours and the quality honestly. You will learn more about leading through AI in that month than in any amount of reading, and our piece on measuring whether AI is paying off shows how to keep the measurement honest.
We run a one-day, CPD-certified Leading through AI course for leadership teams that want to do this properly, and an AI opportunity review if you want help choosing where to begin.
Frequently asked questions
What is the biggest change for leaders in the AI era?
You spend less time producing and more time deciding and checking. The skill shifts from doing the work to judging work you did not do, quickly and well.
How do I keep my team from being afraid of AI?
Be specific. Say which roles change, what you expect people to learn, and what support you will give. Specific, visible investment in skills calms a team far better than general reassurance.
Should I let my team use AI before we have a policy?
They already are. A short, clear policy on data, checking and accountability makes that use safe. Waiting for a perfect policy leaves you exposed in the meantime.
How do I check AI-assisted work without doing it all myself?
Ask three questions of whoever produced it. What did you check, what could be wrong, where did the facts come from. If they cannot answer, it is not finished. The habit spreads quickly once it is expected.
Do I need to understand the technology?
Enough to know where the tools fail in your domain, and no more. The leadership decisions are about work, people and risk, which you already understand.
Why do most AI projects fail to pay off?
Usually because the work was left unchanged and a tool was added on top. The organisations seeing returns redesign the process around AI and put a person in charge of the outcome.