AI is no longer something your organisation is experimenting with. It is being used in your team right now — sometimes with your knowledge, sometimes without it. The question for managers in 2026 is not whether to engage with AI, but how to lead a team through it thoughtfully. This guide sets out the most important things managers need to understand, and the concrete actions that follow from each.
1 Understand what is actually happening on your team
The first thing many managers discover when they look closely is that AI adoption in their team is already further along than they realised — and less organised than they would like. Research tracking employee behaviour found that around 27% of workers admit to using AI tools at work without their employer's knowledge or explicit approval, with the most common reasons being unclear guidelines or uncertainty about how it would be perceived, not deliberate policy violations. This is what is increasingly called shadow AI, and it represents a real risk: sensitive data flowing into systems with no oversight, inconsistent outputs being used without critical review, and no shared standard for what good looks like.
The starting point for any manager is therefore an honest audit. Not to police or punish, but to understand the actual landscape: what tools are people already using, for which tasks, and with what level of judgment applied to the outputs. That understanding is the foundation for everything that follows.
It helps to have a simple way of sorting what you find. AI tools in knowledge-work settings typically do one or more of four things:
- Content generation: drafting emails, documents, reports, presentations, code
- Information processing: summarising meetings, analysing data, extracting insights from long documents
- Task automation: scheduling, routing requests, triggering workflows, updating records
- Decision support: surfacing options, flagging risks, generating recommendations
The category a tool falls into tells you a great deal about where your attention is needed. Task automation on low-stakes internal processes genuinely just saves time. Decision support — anything shaping a judgment about a person, a client or a commitment — is where human oversight matters most, and where the absence of it does the most damage.
Run a brief, blame-free team conversation — in a one-to-one or a team meeting — specifically about AI. Ask what tools people are already using, what they find helpful, and where they feel uncertain about whether something is appropriate. Frame it as curiosity rather than audit. As you go, note which of the four categories each use falls into: that tells you where to concentrate your oversight and where to simply let people get on with it.
2 Your support for AI is the strongest predictor of your team's success with it
Gallup's 2025 State of the Global Workplace report identified a striking finding: employees who believe their manager actively supports their team's use of AI are 8.7 times more likely to say AI has genuinely transformed how much work gets done in their organisation. The manager's posture — engaged, curious, and willing to model AI use themselves — is a more powerful determinant of team adoption than almost any other factor, including the quality of the tools themselves.
This matters because many managers are still on the sidelines. They may be waiting for clearer guidance from above, or quietly uncertain about how to engage with tools they do not fully understand. But the research is consistent: an indifferent manager actively undermines the investment an organisation is making in AI. You do not need to become a technical expert to lead well here. You do need to demonstrate genuine engagement.
Pick one task you do regularly — drafting a status update, preparing talking points for a meeting, summarising notes from a team retrospective — and use an AI tool to do a first draft. Then edit it, apply your own judgment, and notice where the output was useful and where it needed significant correction. Share that experience with your team, including what surprised you and what you had to fix. The act of doing this openly is worth more than any policy document.
LeadWise includes weekly coaching on AI literacy for managers — covering how to lead AI adoption conversations, set team norms, and use AI tools for coaching and development — with 24/7 AI coaching to support you as you apply these skills.
Download the app3 Set clear norms before your team needs them
One of the most common management failures around AI is waiting until something goes wrong before establishing any shared expectations. By that point, bad habits are already embedded, and the conversation about norms becomes a reactive one tied to a specific incident rather than a constructive one tied to shared principles.
MIT Sloan research on AI in enterprise settings recommends that leaders assemble a cross-functional view of acceptable use, provide hands-on guidance for identifying appropriate use cases, and offer a limited set of approved tools so employees can make informed choices. For most managers, the practical version of this is much simpler: a one-page team guide that covers which tools are approved for what purposes, what kinds of data should never be input into external AI systems, and how outputs should be verified before they are used or shared. The specifics matter less than the fact that they are explicit and shared.
The part of this that most often fails is data privacy, and it fails for a predictable reason: it gets handled as a one-off training session rather than an ongoing practice. Many AI tools process significant amounts of personal and sensitive information, and the tools your team uses will change faster than any annual training cycle can keep up with. The managers who get this right make it routine — a short, standing conversation attached to each new tool rather than a compliance box ticked once a year.
Draft a short team AI guide — one page is enough — and co-create it with your team rather than issuing it top-down. Include three things: which tools are approved for use in your team's work; what categories of information should never be entered into an external AI system (client data, personal information, confidential financials); and how outputs should be treated (as drafts that require human review, not final deliverables). Then make it standard practice that whenever someone introduces a new tool to the team, you spend two minutes on what data it handles and where that data goes. Revisiting the guide quarterly keeps it current as both the tools and your team's use evolve.
4 Build genuine AI literacy, not just tool access
There is a significant difference between a team that uses AI tools and a team that uses them well. Using them well means more than knowing how to operate the interface. It means knowing how to frame a question and supply the right context, how to evaluate an output critically, how to iterate when the first attempt is not good enough, and how to recognise the characteristic failure modes of a given system — the confident-sounding answer that is subtly wrong, the summary that drops the one caveat that mattered.
This is what prompt literacy actually means in practice. It is not a technical skill; it is a critical thinking skill applied to a new kind of tool, and it is becoming as fundamental for managers as knowing how to run an effective meeting or write a clear email. You do not need to understand how large language models are built. You do need to understand enough about how they behave to make informed judgments about when to trust their output and when to apply your own thinking instead.
As a manager you are responsible for your team developing this, for two reasons. The first is effectiveness: the gap in output quality between critical and uncritical AI use is large and widening. The second is risk. Uncritical AI use is how inaccurate work reaches a client, and the reputational cost of that lands on the team, not on the tool.
Run a short team session where everyone uses an AI tool on the same real work task, then compares outputs side by side. What worked well? What needed heavy editing? What did the AI miss that a human would have caught? The comparison is what makes this valuable — seeing how differently the same tool responds to different framing teaches prompt literacy faster than any training document, and it surfaces the informal expertise already sitting in your team.
5 Use AI to become a better manager — not just a more efficient one
The most obvious applications of AI for managers are efficiency gains: faster meeting summaries, quicker first drafts, less time on administrative tasks. These are real and worth pursuing. But the more significant opportunity is using AI to become a more effective people manager — more prepared for difficult conversations, more consistent in giving feedback, more deliberate in developing your team.
AI tools can help managers think through how to approach a sensitive one-to-one, identify questions to ask in a development conversation, or prepare for a performance review. The AI does not replace the human judgment required in those conversations — that remains entirely with the manager. But it can reduce the cognitive load of preparation, which in practice means those conversations happen more often and are better structured when they do.
Before your next difficult one-to-one — a performance conversation, a piece of critical feedback, a conversation about someone's development — use an AI tool to prepare. Describe the situation (without identifying information) and ask the tool to help you think through how to open the conversation, what the employee's perspective might be, and what outcomes you are aiming for. The preparation process itself is valuable: it forces clarity about your own thinking before you are in the room. The conversation remains human; the preparation is simply better.
6 Address the anxiety in your team directly
Not everyone on your team is enthusiastic about AI, and some of that scepticism is well-founded. McKinsey's 2025 workplace AI report found that while a slim majority of employees are AI optimists, a significant minority — around 41% — are more apprehensive and will need active support to engage productively. Concerns about job security, about inaccuracy, and about the ethics of AI outputs are all legitimate and deserve to be taken seriously rather than dismissed.
A 2025 study from researchers at Harvard, Hebrew University, and the University of Texas found that people consistently rate the same responses as more empathetic when they believe a human wrote them rather than an AI — even when the text is identical. This points to something important: the perceived authenticity and human connection in management matters deeply to employees, and that perception can be damaged if AI is felt to be replacing rather than supporting genuine human judgment. Managers who are transparent about how and why they use AI — and who remain visibly present and human in their leadership — protect against this.
The practical consequence is that some interactions should stay deliberately unmediated. Your one-to-ones are the clearest example. Let a summary tool run if it is useful, but do not let it stand in for your own listening, and do not arrive with AI-drafted talking points you have not thought through yourself. The value of a one-to-one was never the information exchanged; it is the relationship built, and that is precisely the thing no tool can do on your behalf.
Invite the concerns directly. In a team meeting or one-to-ones, name the topic: "I know AI is changing a lot about how we work, and I want to make space to talk about what that's bringing up for people." Listen without immediately reassuring. Some concerns will surface things worth acting on — tasks that genuinely are changing, skills that need developing, or fears about job security that deserve honest answers. Others will reduce simply by being heard. The manager who avoids this conversation cedes the narrative to rumour; the one who opens it becomes the person people trust to navigate it.
7 Redefine what good work looks like in your team
As AI absorbs more of the routine and structural elements of knowledge work — first drafts, data summaries, scheduling, status reporting — the definition of high-quality work in most teams is shifting. The question is no longer just "did this get done?" but "what did the human add?" Judgment, context, ethical consideration, the knowledge of a specific client relationship or organisational history that no AI system has access to — these are the things that remain distinctively human and that are growing in value.
Asana's experience, shared publicly, describes the shift managers face: moving from encouraging AI adoption to actively redesigning how workflows are structured around AI as an integral component — reconsidering job scopes, redefining quality standards, and identifying which human skills become more valuable. This is genuinely managerial work, and it requires deliberate attention rather than drifting into it by default.
There is a risk on the other side of this that is easy to miss because it develops slowly. If AI writes most of your team's first drafts, do people stay sharp at writing? If AI summarises every meeting, does anyone still develop the skill of listening and synthesising in real time? Skill atrophy is not a hypothetical concern — it is a predictable consequence of delegating a capability consistently enough for long enough. The skills most at risk are often the ones underpinning the human contribution you are trying to make more valuable, which makes this a problem for managers to counteract deliberately rather than notice in hindsight.
For each major type of work your team produces, ask: if an AI handled the structural elements of this, what would excellent human contribution look like on top of that? The answers will vary — it might be domain expertise, stakeholder relationship knowledge, creative interpretation, or quality judgment. Once you are clear on the answer, make it explicit with your team. Then guard against the atrophy risk by periodically rotating which tasks are done with AI assistance and which are done unaided. Teams that can still work both ways are considerably more resilient than those that have quietly become dependent on one mode.
8 Keep humans in the loop on decisions that affect people
As AI tools are increasingly used in performance management, feedback synthesis, and workload tracking, a critical line needs to be maintained: AI should inform managerial judgment, not replace it. The risk is not that AI makes wrong decisions — it is that managers delegate consequential decisions to AI outputs without applying the human context, relationship knowledge, and ethical consideration that those decisions require.
Research on AI in performance management is clear on this point: AI can aggregate inputs, identify patterns, and surface summaries — but managers must retain final judgment on anything that affects a person's role, development, or standing in the organisation. Employees who feel assessed by an algorithm rather than by a manager who knows them experience a significant erosion of trust that is hard to rebuild. The efficiency gain from AI-assisted performance processes is only real if the human judgment layer is visibly and genuinely present.
Bias is the sharpest version of this problem. An AI system is only as fair as the data it was trained on and the design choices made in building it, and neither of those is visible to you as a user. Bias can produce unfair outcomes in recruitment screening, performance assessment and workload allocation — and those outcomes carry your name, because you are the manager who acted on them. This is also the practical reason a plausible output deserves more scepticism than an obviously poor one: a recommendation that reads as reasonable is the one most likely to be accepted without the scrutiny it needs.
Establish a personal rule: any AI output that will influence how you manage or assess a specific person gets reviewed with a simple question — "What does my direct knowledge of this person tell me that this output does not capture?" If the answer is "nothing," that is worth pausing on. Your knowledge of the individual — their context, their circumstances, their trajectory — is exactly what makes your judgment valuable. AI can support that judgment, but it cannot replicate it, and treating its outputs as verdicts rather than hypotheses is where things go wrong.
9 Know your regulatory obligations — and where the technology is heading
Two things have changed since most organisations wrote their first AI guidance, and both sit squarely in a manager's remit. The first is agentic AI. The tools of 2024 largely answered questions; the tools of 2026 increasingly take sequences of actions on their own — researching, drafting, sending, scheduling, updating records. That is a meaningful shift rather than an incremental one. When a tool only produces a draft, human review is a natural checkpoint in the workflow. When a tool executes, the checkpoint has to be designed in deliberately, or it simply does not exist. Knowing which of your team's tools can act rather than only advise is now part of understanding your own operating environment.
The second is regulation. The EU AI Act entered into force in 2024 and is being phased in through 2026 and beyond. It classifies AI systems by risk level and places specific obligations on organisations using high-risk systems — including in recruitment, performance evaluation and workplace monitoring, which is to say precisely the areas where managers make decisions about people. You do not need to be a legal expert. You do need to know whether your organisation uses AI in any of those areas, and to understand your own responsibilities in that context — particularly around being transparent with your team when AI is informing decisions that affect them.
If your organisation uses AI tools in hiring, performance management or workforce planning, ask your HR or legal team for a short briefing on how the EU AI Act applies to them and what it requires of you. Then do a separate pass on your team's own tools with one question: can any of these take an action without a person approving it first? For every one where the answer is yes, decide explicitly what the human checkpoint is. Being informed on both fronts protects you, your team and your organisation — and demonstrates the kind of responsible leadership that builds long-term credibility.
The bottom line
Managing in the age of AI does not require becoming a technologist. It requires the same things that have always made management effective — clarity, genuine engagement with your people, the willingness to model the behaviour you want to see — applied in a context where the tools are more powerful, more autonomous, and more regulated than they were even a year ago. The specifics you need to be clear about have expanded: what your team is actually using, what good use looks like, where the human checkpoint sits, and what the law expects of you. The underlying job has not.
The managers who navigate this well are not those who wait for a comprehensive framework to arrive from the top. They are those who engage with their team honestly, set clear shared expectations, build real literacy rather than just handing out tool access, and keep their own human judgment firmly in the decisions that matter most.