You can lead a team whose members know more about AI by learning enough to understand their work, giving their expertise room to influence decisions and keeping the responsibilities that belong to you. Ask a colleague to teach you one permitted workflow. Try it yourself, explain what still needs checking and agree who can resolve questions beyond either person's knowledge.
You do not have to become the team's best AI user. You do need to understand what you are asking people to rely on. A confident demonstration cannot settle a data permission, a customer commitment or an unfamiliar technical risk.
The wider people and change guide covers support across the organization. This article focuses on your own learning and the boundary between delegating expertise and handing over an unresolved management decision.
Identify what you need to understand
Start with the workflow your team actually uses. You may need help understanding an AI tool while already knowing the supplier rules, customer expectations or operational constraints that determine whether its output is useful. Those are different kinds of knowledge.
A 2026 interview study by Qing Xia and colleagues gives a small but useful example. An HR officer at an agricultural institution described using AI extensively while checking the relevant subject matter with a supervisor who was less inclined to use it. The study interviewed 19 knowledge workers who already used generative AI regularly. It records their experiences, rather than testing whether a particular management method improved results.
The account shows why “knows more about AI” needs a specific meaning. Someone may know how to get a useful draft while needing a colleague's judgment about its content. Neither person's expertise automatically covers the whole task.
For a workflow you supervise, work toward being able to explain:
- The purpose. What should the finished work help someone do, and why is AI being used here?
- The information path. What goes into the tool, what is allowed and where the output goes next?
- The limit. What can go wrong, and what would reveal that the result needs correction or expert help?
- The responsibility. Who checks the work, who can authorize its use and what happens while an important question remains open?
These are practical learning goals, not a proficiency score. If you cannot yet explain a consequential part, name that gap and arrange support before relying on it.
The Chartered Management Institute's June 2026 report on AI leadership argues for managers to develop practical familiarity rather than outsource all understanding to technical departments. Its recommendations combine professional and advisory perspectives; they do not establish a universal amount of training or prove that tool use alone makes a manager capable.
Learn one workflow from the colleague who understands it
Ask for a learning exchange around an ordinary task, with a clear stopping point. Let the colleague know what you want to understand and ask whether they can help within their workload. If teaching a direct manager feels awkward, offer a suitable peer, trainer or group arrangement instead.
Suppose Sam Taylor manages purchasing at a commercial bakery. Jamie Lee, a purchasing coordinator, knows an approved AI tool better than Sam does. Jamie uses it to draft routine supplier availability inquiries from a public flour specification and an approved message template. Orders still go through the bakery's existing purchasing process.
Sam asks Jamie to show one inquiry from input to checked draft. They use permitted practice material, excluding private prices and customer information. The purpose is for Sam to understand the method and its limits, rather than ask Jamie for a polished demonstration.
Try the exchange in this order:
- Let the colleague explain the choices. Ask what they supply, what they leave out and which part of the draft they inspect most carefully. Include a plausible mistake or an attempt that needed correction.
- Take a turn yourself. With the same permitted practice material, try the task while the colleague observes. Say where you are unsure. Use notes and any accessible working methods you need.
- Explain the boundary back. Describe what the draft can be used for, which details need checking and which decisions remain outside the workflow. Ask the colleague to correct your understanding.
- Record the unresolved question. Name the person who can answer it and what the team will do meanwhile. An unanswered question is a result of the session, not something to hide to make it look successful.
For Sam, a useful outcome is being able to explain that the message asks about availability; it does not approve a flour substitution or place an order. If the draft introduces a different product, Sam needs to recognize that the original task has changed. Jamie's fluency with the tool does not grant either colleague authority to approve a new specification.
One successful attempt does not establish competence across every AI use. Choose the next learning task from a real responsibility you still cannot explain. Use the regular manager review for team learning afterward, without turning every meeting into a lesson for you.
Delegate a contribution with an explicit boundary
After the exchange, agree what the colleague can decide and what remains with you or another owner. Avoid making the most enthusiastic user responsible for every tool question, permission request and finished result.
For the bakery's availability inquiry, the arrangement could look like this:
| Decision | Contribution to delegate | Responsibility to keep clear |
|---|---|---|
How to prepare the draft | Jamie can improve the instructions and explain the method. | Sam confirms the task and permitted practice; organizational tool and data rules still apply. |
Whether the inquiry is usable | The purchasing colleague checks the requested product and message against the approved specification and template. | Sam arranges qualified review and follows the existing sending approval. |
Whether a different flour is acceptable | A qualified quality owner answers the specification question. | Neither Sam nor Jamie treats the AI suggestion as approval. The substitution stays outside this inquiry. |
Whether to place an order | Purchasing follows the established order process. | A drafting experiment does not create authority for automatic orders. |
This makes delegation concrete without requiring Sam to prescribe every tool interaction. If the question is about integration, access or system behavior that the team cannot evaluate, prepare a bounded request for technical specialist support. If the gap is an organizational permission, take it to that owner.
When neither person can validate a consequential claim, keep it unresolved and use an appropriate alternative. Asking another AI system for agreement does not supply the missing authority or source evidence. For a fuller approach to checking one deliverable, use the guide to evaluating AI-assisted work.
Make disagreement useful and teaching sustainable
Your colleague needs room to tell you that your proposed method will not work. In a public Reddit discussion, a Reddit user described a manager feeding their completed work into AI and asking them to follow the generated alternative. They said it did not fit the actual project and, in a reply, the client's brand. They also clarified that the manager had knowledge of the field and that they used AI themselves. This is an anonymous account with no independently available work samples.
The concern is more specific than a disagreement over whether to use AI. A generated alternative was becoming an instruction without settling the requirements. As a manager, ask which requirement your suggestion meets or misses. Give the colleague a way to show the conflict using the actual approved materials. Change your view when the evidence warrants it.
A learning arrangement also needs an honest workload agreement:
- Agree the time. Include preparation and follow-up, and decide what moves if teaching competes with delivery.
- Define the support boundary. One learning session does not make the colleague your continuing help desk or the team's default reviewer.
- Credit the contribution. Acknowledge the person's teaching and expertise appropriately; ask what they are comfortable sharing from the session.
- Keep the purpose clear. Explain what will be recorded and how the exchange relates to ordinary work assessment. Do not promise absolute confidentiality that you cannot provide.
You still manage the employee, so saying “there is no hierarchy here” does not remove the difference in authority. Make it possible to decline the teaching arrangement or suggest another route without turning that choice into a judgment about their commitment to AI.
Questions about leading a more AI-fluent team
Does a manager need to learn to code?
A manager does not need coding skills for every AI-assisted workflow. The knowledge required follows the work and its consequences. For routine drafting, understand the permitted inputs, checks and approval boundary. For connected systems or automation, arrange qualified technical support when you cannot assess the behavior. Use the workflow learning goals to identify your own gap before choosing training.
Should the youngest employee teach the manager?
Choose a teacher for relevant demonstrated knowledge, willingness and available time, rather than age or seniority. Someone who uses AI often may still be unable to explain the checks for your workflow. Agree a bounded exchange and offer another learning arrangement if teaching a direct manager feels uncomfortable. The bakery example shows the manager taking a turn and checking their understanding.
What if the employee and manager disagree about an AI result?
Identify the disputed requirement and the evidence or authority that can settle it. A manager's preference, an employee's confidence and an AI-generated alternative are not interchangeable with that evidence. Check the approved source material together; involve the relevant owner when the question exceeds your knowledge. Keep the affected use on hold if the consequential detail remains unresolved, with an appropriate way to finish necessary work.
Can a manager delegate all AI decisions to a power user?
A manager can delegate specific work to a capable colleague with agreed time and boundaries. That does not automatically transfer operational accountability, organizational permissions or specialist approval. Record which decisions the colleague can make and which need another owner. Use the delegation comparison to separate drafting, checking, specification approval and ordering before expanding the arrangement.
