AI for Managers: A Practical Guide to Leading with AI

2026-10-24 · Tommaso Maria Ricci

Managers are the people companies count on to make AI work, and they are also the people companies are least prepared to support. In a Gartner survey of 114 HR leaders, only 8% said their managers have the skills to use AI effectively. At the same time, Gallup's latest global data shows manager engagement falling to 22% in 2025, down from 27% a year earlier. AI for managers is no longer an optional skill set. It is the job description being rewritten in real time, often without anyone telling the manager.

This guide is for people who run teams: department heads, team leads, operations managers, finance and sales managers, and the founders who still manage people directly. It covers how to use AI as a manager in your own work, how to lead a team that uses it, what to delegate and what to keep, how to measure whether it is working, and where the real risks sit.

I have built and run several companies, and I have managed teams through every wave of "this technology changes everything." Most waves changed less than promised. This one is different in one specific way: it changes what a manager's hour is worth. The rest of this guide is about spending that hour well.

Why AI for managers is a different problem from AI for employees

Most AI rollouts are designed for individual contributors. Give people a tool, run a training session, measure usage. That model breaks at the manager level for three reasons.

First, managers have a double job. They have to use AI in their own work and, at the same time, decide how their team uses it. An analyst who misuses a tool makes one bad report. A manager who sets the wrong rules makes twenty.

Second, the manager's work is mostly judgment, coordination, and conversation. These are exactly the areas where current AI is uneven. The Harvard Business School field experiment with Boston Consulting Group, run on 758 consultants, found that people using GPT-4 completed 12.2% more tasks, 25.1% faster, with higher quality, on tasks inside the AI frontier. On a task chosen to sit outside that frontier, a complex business problem, the AI users were 19% less likely to reach the correct answer. The authors called it a jagged technological frontier. Managers spend their days on that jagged edge.

Third, managers carry the accountability. When an AI-assisted decision goes wrong, nobody blames the model. They ask who approved it.

The adoption gap is already here

The data shows managers moving faster than their teams. BCG's AI at Work 2025 survey, covering 10,635 employees in 11 countries, found that regular use among managers reached 78%, up from 64% in 2023. Among frontline employees it was stuck at 51%. Only about one in four frontline employees reported getting strong support from their leaders on AI, and the share feeling positive about it rose from 15% to 55% when that support existed.

Read that carefully. The single biggest lever on whether a team adopts AI is not the tool. It is the manager.

What changes in a manager's job when AI arrives

It helps to break the manager's job into its real components and look at each one honestly. Roughly, a manager does five things:

Plans the work: priorities, resources, timelines.
Coordinates the work: meetings, updates, handoffs, follow-ups.
Develops people: feedback, coaching, reviews, hiring.
Makes decisions: trade-offs, escalations, approvals.
Represents the team: reporting upward, negotiating sideways.

AI does not touch these equally.

Coordination shrinks the most

Coordination is the most automatable part of management. Meeting summaries, status reports, follow-up emails, project updates, first drafts of plans: all of this is text processing, and text processing is what large language models do best.

Microsoft's 2025 Work Trend Index, based on 31,000 workers in 31 markets, found that 80% of the global workforce, leaders included, say they lack the time or energy to do their work. The same report counts roughly 275 interruptions a day during core hours, one every two minutes. If AI gives a manager back even a fraction of that coordination time, the gain is real.

Planning and decisions change shape, not size

AI can prepare a plan, model scenarios, and surface data a manager would never have time to pull together. It does not remove the decision. What changes is the input: instead of deciding on three slides and a gut feeling, a manager can decide on a structured comparison of options. The risk is that the structure looks so convincing that the manager stops questioning it.

People development becomes more important, not less

Feedback, coaching, and trust are where AI helps least directly and matters most indirectly. When half of a team's routine output is AI-assisted, the manager's job shifts toward judging quality, developing judgment in others, and keeping people engaged through a change that many find threatening. With manager engagement at 22% globally according to Gallup, this is the part most companies are neglecting.

Representation gets harder

When everyone can produce a polished report in ten minutes, polished reports stop being a signal. Managers who used to stand out by presenting well now stand out by presenting the right thing: the decision that matters, the risk nobody else saw, the number that does not add up.

How to use AI as a manager: the daily workflow

This is the practical core. Here is how to use AI as a manager across a normal week, organized by the work you actually do. None of this requires a specific vendor. It requires a company-approved tool, clear data rules, and the habit of checking output.

Before meetings: preparation in minutes

Before a one-on-one or a project review, use AI to pull together what you need:

A summary of the last three meeting notes with this person or project.
Open action items and who owns them.
A list of questions you should ask, based on what changed since last time.

The value is not the summary itself. It is walking into the meeting with your attention on the person instead of on remembering what was said two weeks ago.

During and after meetings: capture and follow-up

Meeting assistants can produce a transcript, a summary, and a list of decisions and actions. Two rules make this work:

Announce it. Everyone in the meeting should know it is being recorded and summarized. Some conversations, such as performance concerns or personal issues, should never be recorded.
Edit before sending. The summary is a draft. You are responsible for what it says about who agreed to what.

Writing: from blank page to editor

Managers write constantly: emails, proposals, job descriptions, project plans, policies, feedback. AI changes your role from writer to editor. Give it the context, the audience, and the outcome you want, then rewrite the draft in your own voice. The fastest managers I know use AI to produce three versions of a difficult message and pick the one closest to what they would say on a good day.

Analysis: asking better questions of your data

You do not need to be an analyst to ask AI to compare this quarter against last quarter, spot the outlier in a sales pipeline, or summarize customer complaints by theme. The rule here is simple: AI can find the pattern, you have to explain it. If you cannot explain why a number moved, you do not understand it well enough to act on it.

Decisions: a structured second opinion

For important decisions, use AI as a sparring partner, not an oracle:

Ask it to argue against the option you prefer.
Ask what information would change the decision.
Ask what the second-order effects might be in six months.

This is where the jagged frontier matters most. The HBS study showed that on complex business problems AI can confidently lead people to the wrong answer. Use it to widen your thinking, never to close it.

Reporting upward: shorter and sharper

Use AI to condense your team's work into a one-page update for leadership: three wins, three risks, one decision needed. Then check every number against the source. A manager who forwards an AI summary with a wrong figure loses credibility that takes months to rebuild.

Leading a team that uses AI

Using AI yourself is the easy half. Leading a team through it is where most managers struggle, and where the BCG data says the real leverage sits.

Set the rules before the tools

Before anyone on your team uses AI for real work, they need three answers:

Which tools are approved? Company-licensed tools with data protection, not free consumer accounts.
What data can go in? Customer data, financials, employee information, contracts: each category needs a clear yes, no, or "only in the approved tool."
Who checks the output? Every AI-assisted deliverable has a named human reviewer.

If your company does not have an AI acceptable use policy, push for one. If it does, make sure your team has actually read it. BCG found that 54% of employees say they would use AI tools their company has not authorized. That is the shadow AI problem, and it lands on the manager's desk first.

Model the behavior

Teams copy what managers do, not what managers say. If you use AI openly, share what worked and what failed, and show how you check its output, your team will do the same. If you use it quietly and present the results as your own unaided work, your team will learn that AI use is something to hide.

A practical habit: once a week, in a team meeting, spend five minutes on "one thing AI helped with and one thing it got wrong." It normalizes use, spreads good techniques, and keeps everyone alert to errors.

Redesign the work, not just the tasks

The biggest mistake is to bolt AI onto the existing process. A team that used to spend two days on a weekly report and now spends two hours has not gained anything unless the freed time goes somewhere useful. Decide in advance where it goes: more customer calls, deeper analysis, faster response times, or simply less overtime.

This is what the research calls redesigning workflows, and it is where the value actually appears. Managers who start now will be the ones doing the redesign rather than receiving it.

Train for judgment, not for prompts

Prompt tricks change every few months. Judgment does not. BCG found that only 36% of employees feel their AI training is sufficient, and that people with more than five hours of training are much more likely to be regular users. The training that matters most covers:

Recognizing when AI output is wrong or incomplete.
Knowing which tasks to keep away from AI.
Explaining AI-assisted work to a customer or a manager.
Handling data safely.

For a deeper framework, see the guide on AI training for employees.

Address the fear directly

People worry about their jobs, and they are not wrong to think about it. Microsoft found that 33% of leaders are considering headcount reductions. Pretending the concern does not exist destroys trust. What works better is honesty about what you know and what you do not:

Say clearly which tasks are changing.
Say what new skills the team will need.
Say what you will do to help people build them.
Do not promise what you cannot guarantee.

Change management for AI is its own discipline. The AI change management framework covers it in depth.

One-on-ones, feedback, and coaching when work is AI-assisted

The one-on-one is where management actually happens, and AI changes it in ways most managers have not thought through.

The quality question becomes central

When a team member hands you a report, you used to judge two things at once: the effort and the result. With AI in the loop, effort becomes invisible and almost irrelevant. What matters is whether the person understood the problem, checked the facts, and made the right calls. That changes the questions you ask:

What did you check, and how?
Where did you disagree with the first draft?
What would make this conclusion wrong?

These questions reward judgment instead of polish. They also make it obvious, quickly, when someone has passed along unchecked output.

Feedback on AI use is now part of feedback

Good managers already give feedback on how people work, not just on what they deliver. AI use is now part of how people work. Notice who uses it well, who avoids it, and who over-relies on it, and talk about it the way you would talk about any other skill. A junior employee who uses AI to learn faster deserves recognition. One who uses it to skip learning needs a conversation.

Use AI to prepare, never to replace the conversation

AI is useful before a difficult conversation: it can help you structure your points, anticipate reactions, and find clearer words. It is not useful during it. People can tell when feedback has been scripted by a machine, and nothing erodes trust faster. The NBER research on support agents found the largest gains among less experienced workers, which suggests AI can speed up how junior people learn. Coaching is what turns that speed into competence.

Protect the time you saved for people

If AI saves you four hours a week of coordination, put at least one of those hours back into your team: longer one-on-ones, more frequent check-ins with new hires, career conversations you never had time for. With engagement among both managers and employees falling, this may be the highest-return use of AI in management.

What managers should delegate to AI and what they should keep

This is the question I get most often from founders and managers. A useful way to answer it is to sort your work by two criteria: how reversible a mistake is, and how much the task depends on trust between people.

Delegate freely: low risk, low trust

Meeting summaries and action lists (with a quick edit).
First drafts of routine emails and documents.
Data pulls and simple comparisons.
Scheduling and agenda preparation.
Research summaries on topics you will verify.

Delegate with review: higher risk, low trust

Financial analysis and forecasts.
Customer-facing proposals and contracts.
Policy documents and process descriptions.
Job descriptions and interview question sets.
Performance data summaries.

Here AI does the first 70% and a person owns the final 30%, with a documented check.

Keep: high trust, regardless of risk

Delivering difficult feedback.
Performance ratings and promotion decisions.
Hiring and firing decisions.
Conflict resolution between team members.
Conversations about personal circumstances.

AI can help you prepare for these conversations. It should not have them for you, and it should not make the call. Performance decisions in particular carry legal and ethical weight. The guide on AI in performance management explains where AI supports the process and where it must stay out.

Watch for the regulatory line

In the European Union, the AI Act classifies AI systems used for recruitment, for decisions on promotion or termination, and for monitoring and evaluating workers' performance as high-risk, with obligations that include human oversight. In New York City, Local Law 144 already requires bias audits for automated employment decision tools. If you manage people in these jurisdictions, check with your legal or HR team before using any AI tool that influences decisions about individuals.

The manager of the future manages humans and agents

AI agents are systems that do not just answer questions but carry out multi-step tasks: updating records, sending follow-ups, routing requests, preparing reports. Microsoft's report describes a future where employees become "agent bosses," building and supervising agents. In its global data, 36% of leaders expect managing agents to be part of their job within five years.

The reality today is more modest. BCG found that 13% of organizations have agents integrated into broader workflows, 56% are experimenting or running pilots, and 31% have not started. But the direction matters for managers now, because the skills transfer.

What managing an agent actually involves

Managing an agent looks surprisingly like managing a new hire:

Define the task clearly. Inputs, outputs, quality standard, deadline.
Set the limits. What the agent can do on its own and what needs approval.
Check the work. Especially at the start, review a sample of everything it produces.
Handle the exceptions. Decide who intervenes when the agent gets stuck or gets it wrong.
Measure the result. Time saved, errors made, rework required.

The managers who are good at delegation to people tend to be good at delegation to agents. Those who micromanage people tend to either micromanage agents into uselessness or, worse, trust them blindly. For the governance side, see the guide on AI agent governance.

Will AI flatten middle management?

Gartner predicted in October 2024 that, through 2026, one in five organizations would use AI to flatten their structure, eliminating more than half of current middle management positions. That is a forecast, not a measured result, and it applies to a minority of companies. The more likely outcome for most is wider spans of control: fewer managers, each with more people, and AI handling part of the reporting and coordination.

The practical implication is clear. The managers who survive a flattening are the ones whose value sits in judgment, development, and decisions. The ones whose value sits in assembling status reports are exposed.

Measuring whether AI is actually helping your team

Without measurement, AI adoption becomes a feeling. Some people love it, some hate it, and nobody knows whether the team is better off. A manager needs a small set of indicators, tracked before and after.

Four metrics that matter

Cycle time on recurring work. How long does the weekly report, the monthly close, the proposal, or the customer response take? Measure the baseline first.
Quality and rework. How often does work come back for corrections? AI that saves time but doubles rework is a loss.
Where the freed time goes. Track it explicitly. If saved hours disappear into more meetings, nothing has been gained.
Team sentiment. A short pulse survey every quarter: do people feel AI helps them, do they feel supported, do they feel their job is at risk?

Benchmarks to keep in mind

Controlled studies give a sense of what is possible. The NBER study by Brynjolfsson, Li and Raymond on 5,179 customer support agents found a 14% average productivity gain, rising to 34% for novice and less-skilled workers, with little effect on the most experienced. The HBS experiment found 25% faster work on tasks inside the frontier.

Two lessons follow. Gains are real but not magical: think in tens of percent, not multiples. And the biggest gains often go to junior people, which means AI can be a powerful tool for developing your team, if you use it to teach rather than just to produce.

Connect to business outcomes

Time saved only matters if it turns into something the business values. In my own companies, this is where AI paid back. At a sports distribution company, using AI in marketing contributed to a 30% increase in sales, because the time saved on content production went into better targeting and faster campaign cycles. At a hotel, revenue grew from 9 to 10 million euros, partly because managers stopped spending hours on manual reporting and started spending them on pricing and guest experience. At a medical center, capacity rose 20% when scheduling and administrative coordination were reorganized. In each case the tool mattered less than the decision about where the freed hours would go.

If you want help defining these metrics for your own team, or deciding where AI fits in your management structure, a consultation request through the site is the most direct way to start the conversation.

Common mistakes managers make with AI

These are the patterns I see most often, in companies of every size.

Waiting for the company to decide. Many managers hold back until there is an official program. Meanwhile their team is already using AI, often with unapproved tools. Act within the rules you have, and push for clearer ones.
Using AI only for yourself. A manager who saves five hours a week personally but leaves the team on its own has captured a fraction of the value.
Measuring usage instead of results. Logins and prompts tell you nothing about whether work got better.
Trusting polished output. Fluent text is not correct text. The more convincing it looks, the more carefully you need to check it.
Feeding sensitive data into the wrong tools. Employee reviews, salary data, and customer contracts in a consumer chatbot are a compliance incident waiting to happen.
Delegating judgment. Using AI to decide who gets promoted, who gets a raise, or which customer gets priority without a human owning the decision.
Ignoring the people side. Rolling out tools without addressing fear, workload, or career paths. This is how adoption stalls.
Not redesigning the work. Speeding up a broken process just produces broken output faster.

Self-assessment: how ready are you to manage with AI?

Score yourself honestly on each statement: 0 if it is not true, 1 if it is partly true, 2 if it is fully true.

Your own use

I use an approved AI tool at least several times a week for real work.
I know which types of company data I can and cannot put into AI tools.
I check AI output against sources before using it in decisions or reports.
I use AI to challenge my thinking on important decisions, not just to confirm it.

Your team

My team knows which AI tools are approved and which are not.
Every AI-assisted deliverable on my team has a named human reviewer.
We regularly share what works and what fails with AI.
I have talked openly with my team about how AI will change their roles.

Your processes

I have measured the baseline time and quality of our main recurring work.
We have redesigned at least one process around AI, not just added a tool.
I know where the time saved by AI is going.
People decisions, such as ratings, promotions, and hiring, always have a human owner.

Reading your score

0 to 8: Starting point. You are likely using AI informally or not at all. Begin with your own workflow and the basic rules for your team.
9 to 16: Building. You have the habits but not yet the system. Focus on measurement and process redesign.
17 to 24: Leading. You are ahead of most managers. Your next step is spreading the approach to peers and preparing for agents.

A 30/60/90-day roadmap for managers

This roadmap assumes you have access to at least one company-approved AI tool. If you do not, step one is getting one approved.

Days 1 to 30: your own workflow and the rules

Week 1: Read your company's AI policy. If there is none, write down the three rules you will apply to your team (approved tools, data limits, human review) and share them with HR or IT.
Week 2: Pick three recurring tasks of your own, such as meeting preparation, weekly reporting, and email drafting, and use AI on them every time.
Week 3: Measure. How much time did you save, and how many errors did you catch?
Week 4: Talk to your team. Explain the rules, share what you learned, and ask what they are already doing with AI. You will probably be surprised.

Days 31 to 60: the team and one process

Week 5: Choose one team process that is recurring, text-heavy, and low-risk. Measure its current cycle time and rework rate.
Week 6: Redesign it with the team. Decide which steps AI handles, who reviews, and where the freed time goes.
Week 7: Run it. Hold the weekly "one win, one failure" slot in your team meeting.
Week 8: Measure again against the baseline and run a short sentiment check.

Days 61 to 90: scale and prepare

Weeks 9 and 10: Extend to a second process, ideally one with higher value, such as customer proposals or financial analysis, with stronger review.
Week 11: Write a one-page summary for leadership: what changed, what it saved, what it cost, what you recommend next.
Week 12: Identify one task that could become an agent workflow in the next six months, and define its limits on paper before anyone builds it.

At the end of 90 days you should have measured results on two processes, a team that uses AI openly and safely, and a clear proposal for what comes next. That is far more than most managers will have.

AI for managers in specific functions

The principles are the same everywhere, but the high-value uses differ by function. A few examples:

Sales managers

Pipeline review is the obvious starting point: AI can flag deals that have gone quiet, summarize account history before a call, and draft follow-ups. The judgment that stays human is forecasting commitment and coaching reps on difficult deals.

Finance managers

Variance analysis, report drafting, and reconciliation support are strong uses. Anything that goes into official financial statements needs documented review, because errors here have legal consequences.

Operations managers

Scheduling, shift planning, incident summaries, and supplier communication are good candidates. Many operations teams also have the cleanest structured data, which makes AI analysis more reliable.

HR and people managers

Drafting job descriptions, summarizing engagement survey comments, and preparing for difficult conversations are useful. Decisions about individuals stay with people, for both ethical and regulatory reasons. The guide on AI for HR professionals goes deeper, and AI workforce planning covers the capacity side.

Founders who still manage directly

In small companies the founder is often the manager of everyone. AI is especially valuable here because there is no layer of middle management to absorb coordination work. The risk is also higher, because there is nobody to check the founder's AI-assisted decisions. Build the review step in on purpose, even if the reviewer is a trusted advisor outside the company.

The bottom line

The manager's job is not disappearing. It is being sorted. The parts that are coordination and assembly are moving to machines. The parts that are judgment, development, and decision are becoming more valuable, because they are the parts that remain scarce.

The data points in one direction. Managers are adopting AI faster than their teams, yet companies rarely equip them to lead that adoption. Teams adopt AI when their managers support them, and stall when they do not. And the productivity gains, while real, depend on redesigning work rather than just adding tools.

The managers who will do well are the ones who start now, within clear rules, measure honestly, and spend the time they save on the work only they can do. If you are working through this in your own company and want an outside view on where to begin, how to set the rules, or how to measure results, request a consultation and we can look at your team's situation together.

FAQ

How do I use AI as a manager without losing control of quality?

To use AI as a manager without losing quality control, treat every AI output as a draft that a named person reviews before it is used. Start with low-risk tasks such as meeting summaries and email drafts, check numbers against their sources, and keep decisions about people, money, and customers with a human owner. Measure rework rates before and after: if corrections go up, the process needs more review, not more automation.

What tasks should managers delegate to AI first?

Managers should start with recurring, text-heavy, low-risk tasks: meeting preparation and summaries, status reports, first drafts of routine emails and documents, and simple data comparisons. These tasks are frequent enough to save meaningful time and low-risk enough that mistakes are easy to catch. Tasks involving trust, such as feedback, performance ratings, hiring, and conflict, should stay with the manager even when AI helps with preparation.

Will AI replace middle managers?

AI is more likely to change middle management than to eliminate it. Gartner predicted that through 2026 one in five organizations would use AI to flatten their structure, which means most organizations will not. The more common outcome is wider spans of control, with each manager overseeing more people while AI handles part of the reporting and coordination. Managers whose value lies in judgment, people development, and decisions are far less exposed than those whose main work is assembling information.

How can a manager help a team adopt AI?

The strongest lever is visible support from the manager. BCG found that the share of employees feeling positive about AI rose from 15% to 55% when they had strong leadership support. In practice that means setting clear rules on tools and data, using AI openly yourself, sharing successes and failures in team meetings, providing training on judgment rather than prompt tricks, and talking honestly about how roles will change.

Is it legal to use AI for performance reviews and hiring decisions?

It depends on the jurisdiction and how the tool is used. In the European Union, the AI Act classifies AI systems used for recruitment, promotion or termination decisions, and the evaluation of workers' performance as high-risk, with requirements including human oversight. New York City requires bias audits for automated employment decision tools. Managers should involve HR and legal before using AI that influences decisions about individuals, and a human should always own the final decision.

How do I measure whether AI is improving my team's performance?

Measure four things before and after introducing AI: cycle time on recurring work, quality and rework rates, where the saved time goes, and team sentiment through a short quarterly survey. Compare against a baseline you recorded before the change. Controlled studies suggest realistic gains in the range of 14% to 25% on suitable tasks, with larger gains for less experienced workers, so treat claims of multiples with skepticism.

What skills do managers need to work with AI agents?

Managing AI agents requires many of the same skills as managing people: defining tasks clearly, setting limits on what can be done without approval, checking work, handling exceptions, and measuring results. Microsoft found that 36% of leaders expect managing agents to become part of their job within five years. Managers who already delegate well to people are in the best position to delegate to agents, provided they keep clear accountability for the outcome.