A project lead arrives with an impressive plan. The milestones are neatly sequenced, the risks have sensible labels and the summary reads well. An AI assistant helped prepare it.
The manager could rewrite the plan, approve it or ask for another version. A more useful first move may be to discover which choices the employee has actually examined.
That is the role of the manager as coach in the AI era: helping someone think through a real work decision while keeping expectations, authority and support clear. It does not mean answering every request with a question. Sometimes the employee needs information, a demonstration or a decision that only the manager can make.
AI adds a reason to choose that response carefully. A polished first answer can make a useful conversation possible. It can also make an unresolved problem look finished.
Coaching is a way of managing a conversation
For this article, managerial coaching means a developmental conversation in which the manager helps an employee examine a goal, consider options and decide what to do next. It sits alongside instruction, feedback and direction.
The distinction matters because a manager has responsibilities that an independent coach may not share. The manager sets priorities, allocates resources and evaluates work. Asking someone to find their own answer does not remove those responsibilities.
Nor does coaching require an employee to rediscover information the manager could simply provide. If a project has a fixed deadline, say so. If a procedure is mandatory, explain it. If the employee lacks essential knowledge, arrange teaching or expert help.
Coaching becomes useful where there is room for the person to make and understand a choice. The conversation should make that room visible.
What the evidence supports—and what it does not
A controlled study published in January 2020 examined a coaching-based leadership program involving 41 executives and middle managers at one Spanish automotive company. The program combined workshops and individual coaching, and the researchers reported improvements in coaching skills and work-related outcomes.
Important limits accompany that result. The groups were not randomly assigned, the sample was small and the intervention combined several elements. It predates widespread workplace generative AI. It provides a reason to take manager development seriously, rather than proof that a particular question improves an AI-assisted team's performance.
More recent research also cautions against treating human and AI coaching as interchangeable. In an exploratory study published on 15 October 2026, Forshaw, Hale and Longmire studied eight managers at a single consulting firm who experienced human video coaching and text-based AI coaching. Participants preferred the human format and responded differently to its reflective questions.
That study concerned managers receiving coaching, not managers coaching their staff. Its small, voluntary sample and differences between the formats prevent a general causal comparison of human and AI coaching effectiveness.
The OECD's June 2026 skills brief offers broader context: managerial capabilities, problem-solving and interpreting data remain relevant alongside technical AI skills. It is a synthesis of workplace evidence, not a trial of a coaching method.
The practical conclusion is modest. Developing better conversations is a plausible management priority. The research does not establish that every interaction should become coaching, or that adopting an AI coach will produce the same experience as working with a person.
A plan worth discussing, rather than merely polishing
Consider a hypothetical cross-functional team preparing a product launch. This example is illustrative, not a Learning Coach client case.
The project lead uses an approved AI assistant to draft a milestone plan. It looks coherent, but one proposed step assumes that a specialist team will be available earlier than anyone has confirmed.
The manager knows two constraints: the launch date is fixed, and additional specialist capacity has not been authorized. Stating those constraints gives the employee a problem they can genuinely work on.
The next conversation might begin with a request to explain the most consequential sequencing choice. Which dependency makes that choice reasonable? What information would change it? Where does the plan rely on an assumption rather than an agreement?
The employee may identify the unconfirmed availability. That creates several possible next steps: confirm the dependency, change the scope or bring a trade-off back to the manager.
The manager's contribution is not simply to ask more questions. It is to help distinguish what the employee can investigate from what requires a management decision.
If the employee does not know how to map dependencies, the next move may be a demonstration. If two departments are competing for the same resource, the manager may need to resolve the conflict. Continuing to ask the employee for a solution would not turn an authority problem into a learning opportunity.
Choose between asking, teaching and deciding
Before beginning a developmental conversation, identify the obstacle.
| What is holding the work back? | A useful managerial response | What to make explicit |
|---|---|---|
| The employee has options but has not examined their implications | Explore the reasoning and trade-offs together | Which choice belongs to the employee |
| The employee lacks relevant knowledge or a method | Explain, demonstrate or arrange appropriate help | What they need to understand before proceeding |
| Priorities or resources conflict | Make or escalate the management decision | Who has authority to resolve the conflict |
| The work does not meet an agreed standard | Give specific feedback and discuss the next attempt | The standard and the evidence of the gap |
This table is an editorial discussion aid, not a validated diagnostic instrument. Real conversations can move between these responses.
In the launch example, a manager might teach a way to identify dependencies, ask the employee to consider the available options and then decide which scope trade-off the business will accept. That combination can be more honest than presenting the whole exchange as coaching.
The employee should not have to guess whether a question invites exploration or conceals an answer the manager has already chosen.
Ask about the choice, then listen to the answer
An AI-assisted document offers many possible points of discussion. Select one that matters to the work.
A question about the plan's main dependency is likely to be more useful than a general demand to explain every sentence. A question about the rejected alternative can reveal more than asking whether the employee is confident.
Then leave room for an answer. If the manager immediately supplies the preferred interpretation, the exchange can become a disguised instruction.
Follow what the employee says. If they identify uncertainty, help define how to investigate it. If they explain a sound reason, acknowledge it and move on. Coaching need not manufacture a problem in competent work.
For the project lead, a useful endpoint might be a specific action: contact the specialist team, check availability and return with the implications for the proposed sequence. The manager can state which decisions may proceed and which should wait.
That endpoint gives the conversation a purpose beyond sounding reflective.
Keep development and evaluation understandable
An employee may approach a conversation differently when the person asking questions also evaluates their performance.
Be clear about the purpose. Is this practice on an early draft, a review of a deliverable or a formal performance discussion? What decision will follow? What information needs to be recorded or shared?
These questions are especially relevant when AI is involved. An employee should know whether the discussion concerns the quality of the work, the use of an approved tool, a capability they are developing or several of these.
Avoid turning an exploratory conversation into an undisclosed test. Equally, avoid calling a conversation purely developmental when it will determine a consequential assessment.
In the fictional launch team, the manager might explain that the draft is being used to explore a planning choice before the team commits to it. If the work later needs formal approval, that is a separate decision with stated criteria.
Clarity does not erase the power difference. It makes the interaction easier to understand.
AI can help a manager prepare
A manager can use an approved assistant to generate alternative questions, rehearse a conversation with a fictional scenario or identify where a draft brief is unclear.
The preparation still requires judgment. A list of generic questions may ignore the employee's experience, the project's constraints or the manager's own responsibility for a delay.
For the launch conversation, an assistant might suggest asking the employee to compare several plans. The manager should first consider whether those alternatives are actually available. Asking someone to choose between options that the organization will not permit wastes their effort.
Use information suitable for the tool and the purpose. A fictional or appropriately prepared practice case may be sufficient; a rehearsal does not automatically require uploading a real employee's private history.
The useful test of the preparation is whether it helps the manager understand the next conversation, not whether it produces an elegant script. The employee's answer should still be able to change what happens next.
Practice the transition between managerial responses
Leadership development can make this practical.
Give managers a fictional AI-assisted project plan and a short description of the employee's experience. Ask them to conduct a conversation, then introduce new information: a missing skill, a fixed constraint or a resource conflict.
Observers can note whether the manager changes response appropriately. Did they provide information when it was needed? Did they leave a genuine choice with the employee? Did they take responsibility for a decision within their authority?
These are proposed practice criteria, not proven predictors of business results. They offer more concrete feedback than counting how many open questions a manager asked.
An event can introduce the dilemma and give managers a shared vocabulary. Continued practice and feedback belong in the workplace. The guide to designing a corporate AI learning summit explores how to connect an event with that subsequent work.
The Learning Coach perspective: make room for a better next decision
Learning Coach's perspective is that employee development becomes visible when a person can participate more thoughtfully in the next decision. A conversation can contribute by making assumptions discussable, providing missing support and clarifying what the employee can own.
This is one practical expression of the human learning advantage alongside AI skills. It does not require treating every tool-assisted answer as suspicious. It requires keeping attention on the work and the person developing through it.
The evidence discussed here was checked on 28 October 2026. It is a selected research base, not a systematic review or a guarantee of coaching effectiveness. The scenario, questions and practice activity are original editorial proposals.
For the manager looking at an impressive AI-assisted plan, the next question is therefore specific: does this person need help examining a choice, learning a method or obtaining a decision from me?
Bring that discussion to your next management or workplace learning event. Explore Learning Coach's international keynotes and request a tailored proposal.
References
Peláez Zuberbuhler, Salanova and Martínez (2020), coaching-based leadership controlled study. A combined development program in one company, with nonrandom assignment and sample limitations.
Forshaw, Hale and Longmire (15 October 2026), The Reflection Penalty. Exploratory human and AI coaching study; small sample and differing modalities.
OECD (June 2026), AI and skills. Policy synthesis on changing workplace skill needs.
