By Andreas Pettersson, Founder, Leaders ADAPT
If I gave you five interns fresh out of college, could you train them to do something useful on your team? If yes, that is what AI for managers is. If no, no tool will fix it, because the problem is not the software. It is that you have never had to explain your job clearly enough for someone else to do part of it.
Most of what is written on this topic is a tool tour: five apps, ten prompts, a screenshot. This is the operating manual for how managers can use AI: what you hand over and what stays yours, the one on one and delegation uses, the mistakes that burn people, and a weekly routine that fits inside the job you already have.
Quick answer: AI for managers works best as delegation, not as a search engine. A manager hands AI the drafting, summarizing, sourcing and tracking parts of the job, keeps every decision, every piece of feedback and every final check, and runs a short weekly routine: prepare one on ones from transcripts, turn repeated prompts into reusable instructions, and review what went out. Start with one task that saves an hour this week.
What does AI for managers mean in practice?
AI for managers means using AI tools to take over the drafting, summarizing and tracking parts of a manager's job so the manager's time goes to judgment, coaching and decisions.
The word that matters is delegation. Treat the tool like a person you brief, not a box you search. Brief it as you would an intern taking dictation: the context, the constraint, what good looks like, what to do if unsure.
Short commands produce generic output; full briefs produce work you can use. If you still type into it like it is Google, start with using AI as an apprentice, not a search engine.
One more reframe. AI does not remove the manager's job. It removes the part that was never management: status chasing, note typing, reformatting. What is left is the part most managers were too busy to do.
What should a manager delegate to AI, and what stays yours?
The rule I use with the leaders I advise fits in one sentence. AI can own the sourcing, the shortlisting and the first draft of almost any workflow, and a human check by the accountable person stays in place until something goes out the door. Catching an error after the proposal has been sent means redoing the cycle, so the check sits before the customer-facing step.
| Hand to AI | Keep for yourself | Why |
|---|---|---|
| First drafts: updates, agendas, job descriptions, process docs | The final read and the send | Your name is on it, not the tool's |
| Meeting notes, summaries, action item extraction | Deciding which action items matter | The list is data; the ranking is judgment |
| Sourcing and shortlisting: candidates, vendors, options | The choice | Choosing is the job |
| Drafting feedback in the shape you want to deliver it | Delivering it, in person, privately | Feedback is a relationship act |
| Tracking commitments and deadlines | The conversation when one slips | Accountability needs a face |
| Answering "what do we know about X" from your own files | Answering "what should we do" | The model recombines the past; you author what is next |
The shape most work takes is 10-80-10. Ten percent is yours at the front: the brief, the context from the meeting, the judgment about what matters. Eighty percent is the tool's middle: the draft, the sort, the summary. The last ten percent is yours again: the check, the edit, the decision to send.
The four calls that never move to the machine, the question you choose to ask, the future you author, the accountability your team can feel and the final call when your gut disagrees with the model, are on what not to delegate to AI.
How can managers use AI for one on one meetings?
I keep a coach folder. Every session with my own coaches and my mastermind goes in as a transcript, and I ask the AI for the step-by-step instructions buried in the conversation, then check whether I did what I agreed to. Same for my sales calls: exported, analyzed, then an AI coach that rates each call and says "don't do this, do that instead."
Your one on ones deserve the same treatment, in three moves.
Before: prepare from the last transcript, not from memory
Record the meeting, with the person's knowledge. Before the next one, hand the AI the last transcript and ask for three things: commitments made by each side, anything raised and left unresolved, and any change in how the person sounded.
That last one matters more than it looks. Ask people for an energy number at the open, one to ten. It is not the number that matters; forcing someone to put down a number identifies the gap, and a number that slides over four meetings is a pattern you will never notice from memory.
During: put the tool away
The meeting is the fifteen minutes every other week where your report gets you, not a screen. One question I hand managers for the person above them works for the people below too: "What keeps you up on a Friday night, and what is coming down the pipe that I can help you attack before it is too late?" The purpose, cadence and structure are on the one on one meeting guide; none of it needs AI in the room.
After: draft the feedback in the right shape, then deliver it yourself
When something needs correcting, have the AI draft it in the shape I teach: describe the situation objectively, name the specific behavior and its complication, state the impact as the leader, then say clearly what you want changed. Read the draft, cut anything that sounds like a machine wrote it, and deliver it privately, never in front of the team. The tool gives the structure. You give the relationship.
How can managers use AI for delegation?
Delegate the outcome, never the task. State the result you want, the feeling you want the recipient to have, and the deadline, then let the person choose the route. That rule predates AI, and it is why AI makes delegation easier rather than harder: an outcome you can state clearly is one you can also hand partly to a machine.
Three practices make it work with AI in the mix.
Cap the escalations. For someone newly handling higher stakes calls, do not choose between full autonomy and none. Give them a capped number of discretionary escalations, roughly one in nine, and pair the cap with a short debrief twice a week that asks why a case was handled that way, not only whether the outcome was right. AI drafts the case summary; the debrief is you.
Run the two interns test. Ask each report: if you had two interns you could ask to do anything, what would you have them do? Now assume I said AI instead of interns. That single question surfaces the automatable work on a team, turns a nervous person into a participant, and finds the one person who already wants this, who is your rollout.
Turn prompts into skills. A prompt is a one-time ask; a skill is a reusable teammate that improves. The moment anyone on the team copy-pastes a prompt a second time, turn it into a saved instruction the whole team can run. That is how a veteran's way of doing the job transfers to a new hire without a course, and why a shared library beats a shared chat history.
The company level view of how decision rights split between people and machines is on delegation in the age of AI; which tools and data are allowed in the first place belongs on the company's AI governance framework page.
The manager's weekly AI routine
Ten minutes a day, thirty on Monday, one note taker, one strong model.
| When | What the AI does | What you do |
|---|---|---|
| Daily, 5 minutes | Asks you three journal questions by voice and files the answers | Answer honestly; the questions get sharper as it learns your team |
| Monday, 30 minutes | Pulls open commitments and unresolved items from last week's transcripts into one page | Rank them; decide what you will chase and what you will let go |
| Before each one on one | Summarizes the last conversation and flags energy changes | Choose the one topic that matters this time |
| After each meeting | Extracts decisions, owners and dates | Correct anything it got wrong before it goes anywhere |
| Friday, 15 minutes | Lists what went out under your name this week | Spot-check two items end to end; note any prompt you ran twice and save it as a skill |
Journaling is the part managers skip and the part that compounds. Three questions a day, answered by voice, is how the tool stops giving generic management advice and starts giving advice about your team's performance. After a few weeks, ask it the gap question: identify everything you do not know about my team that you would need to advise me daily, and list the questions. Then answer them.
What mistakes burn managers who use AI?
Three, none technical.
Making it too big. If you make it too big and complex, people will not use it, including you. The simpler and stupider the first version, the higher the chance of a quick win, and the win has to arrive within an hour of work or the habit will not stick.
Pretending. A member of one of my groups put it plainly: "I don't want to feel like an idiot when I'm asked about it. Give me enough so I feel like I understand what's going on." The temptation is to let an AI summary stand in for understanding, and it works until the first follow-up. Use the tool to close your knowledge gaps, never to hide them.
Sending unchecked. The 10-80-10 shape collapses the moment the human review at the end gets skipped. An AI-written document that goes out with the tool's mistakes in it costs you more credibility than never having used the tool, and the same goes for agents that act without a person clicking. Friday's spot-check exists for this reason.
Will AI replace managers?
Gartner predicted in October 2024 that through 2026, 20 percent of organizations would use AI to flatten their structure and eliminate more than half of their current middle management positions. I built AI products before ChatGPT existed, and I take that number seriously without agreeing with the conclusion managers draw from it.
The manager who gets flattened is the one whose job was the routing layer: collecting status, forwarding it upward, relaying decisions downward. Communication and knowledge sharing in an organization of more than twenty people is the largest slowdown there is, and AI removes that layer well. What it cannot do is the "keep" column: coach a person through a bad quarter, own an outcome, make the call on incomplete data.
Slack is not idle time. It is unused judgment capacity. The hours AI returns to you are only worth something if you spend them on judgment, and that choice decides which kind of manager you are in 2027. The page on leading with AI as a conductor makes the longer argument, and the AI for CEOs playbook is the version to hand the person above you.
AI for managers FAQ
How should a manager start using AI at work?
AI for managers starts with one recurring task that costs at least an hour a week, such as preparing one on one notes or drafting the weekly update, handed to the tool for two weeks. Brief it fully, check every output before use, and save the instruction once it works. Add a second task only after the first has stuck. Starting with a tool rollout instead of a single task is the most common reason managers stop.
What are the best AI tools for managers?
AI for people managers comes down to three categories rather than a long list: a meeting note taker that produces transcripts, one strong general model such as ChatGPT or Claude with a company account, and a place to store reusable instructions and files the model can read. AI tools for managers earn their place when they connect to the manager's own material; a tool that cannot see the team's transcripts produces generic advice.
Can AI replace managers?
AI replaces the routing part of management: collecting status, summarizing it, relaying decisions and chasing deadlines. It does not replace judgment under incomplete information, coaching a person through a difficult period, owning an outcome in front of a team, or delivering feedback that changes behavior. Managers whose job was mostly routing are exposed; managers who spend the recovered hours on judgment and people gain value.
How can managers use AI for one on one meetings?
Use it before and after the meeting, not during it. Before, it summarizes the last transcript, lists commitments from both sides and flags unresolved items or a change in the person's energy. After, it extracts decisions, owners and dates for the manager to correct. During the meeting the tool stays closed so the report gets full attention, and recording happens only with the report's knowledge.
Should managers tell their team they use AI?
Yes. Say which tasks AI drafts, which it will never see, and that every output is checked by a person before use. Transparency prevents two problems: a team that assumes feedback or decisions came from a machine, and a team that quietly uses unapproved tools because the manager never set a standard. Recording meetings for transcripts requires the report's knowledge and consent every time.
How much time does AI save a manager?
It depends on how much of the week was routing work, and no general figure is reliable. One documented example from a company Andreas Pettersson advised: leadership meetings cut from 60 to 30 minutes, with the decisions still made by the people in the room. The recovered time only counts as a saving if it is reinvested in coaching, delegation and decisions rather than absorbed as slack.
Start with the interns, not the tool
You do not need a course for any of this. AI for managers needs one task, one full brief, one check before it goes out, and the honesty to notice whether the hour came back.
Then ask your team the two interns question and listen to what they would hand over. The answers are your first quarter's plan, and the person who answers fastest is the one who will carry it.
The one on one and delegation system this manual sits on
The routine above assumes you already run one on ones that are not status updates and delegate outcomes rather than tasks. If not yet, The 5-Minute Leader is the system I wrote for managers with no spare hour: the three minute one on one template the transcript routine feeds, the delegation checks that go with the outcome rule above, a four week implementation plan, and one part I leave out of every description because it only makes sense once the first three are running.
Andreas Pettersson was a tech CEO for 10 years and one of Canon's youngest CEOs, running an AI company before ChatGPT existed. He founded Leaders ADAPT and wrote The 5-Minute Leader.
