This content may be of interest to you
Resources
Your managers are already using AI. Meeting summaries, feedback preparation, and briefing notes: what used to take half a day can now be done in a few minutes. But this speed creates an illusion. Just because a deliverable is well-written doesn’t mean it’s accurate or appropriate for the situation. The real risk isn’t that managers aren’t using AI. It’s that they’re using it without adjusting their approach or questioning what it produces.
This article is intended for HR directors, training managers, and L&D teams who want to support their managers in usingAI and management in their day-to-day work. By the end of this article, you’ll know which skills to focus your training efforts on and which practices to prioritize.
Why starting with tools is a false start
Most companies approach AI through tools and prompts. That’s useful, but it’s a blind spot. The real issue is managerial. When AI speeds up production, what used to define a manager’s value—producing quickly, synthesizing effectively, and writing clearly—is now handled by machines. What remains irreplaceable is the ability to ask the right questions, make decisions, and stay the course.
In the age of AI, a manager who doesn’t adapt will not stay at the same level. They will fall behind. Their team and leadership are getting used to faster, more structured deliverables. Expectations are automatically rising. The top managerial skills for 2026 confirm this shift: what defines a manager’s value is shifting toward judgment, decision-making, and human support. The bar for what is considered “good management” is rising without warning.
Real-world example. A marketing manager sends an AI-generated briefing memo to the executive committee. During the meeting, a question throws him off balance: “What is your pricing assumption based on?” He doesn’t know how to answer because he hasn’t reconstructed the logic behind the document. The deliverable was sound. The preparation was not. This is exactly the type of situation HR must anticipate in their support systems.
A Practical Tool: The 5-Question Mini-Checklist to Incorporate into Your Management Training. Before using AI on any high-stakes issue, ask yourself: What is the objective? What decision should this output help inform? What is the level of risk? Who will be affected? What quality criteria are non-negotiable?
AI generates content that is fluid, well-structured, and seemingly convincing. That is precisely what makes it risky: a well-written output lulls people into a false sense of security. Managers who do not systematically challenge what AI produces end up validating assumptions they have not verified and recommendations they could not defend. This is likely the most underestimated skill today in managerial frameworks. This was also the observation made by Mirakl and Servier at our annual FORWARD 2026 conference: train in judgment, not in the tool.
Critical thinking applied to AI is not just an intellectual exercise. It is a daily practice built around three concrete habits.
Rule 1: Question the sources. AI doesn’t always cite its sources. Asking “Where does this figure come from?” or “What data does this conclusion rely on?” should be second nature before sharing any information. A manager who shares a recommendation without knowing its basis risks losing credibility with their team and senior leadership.
Reflex 2: Look for what’s missing. AI responds to what you ask it. It doesn’t point out the questions you forgot to ask. What blind spots does this analysis leave? What perspective isn’t represented? What objection might an interlocutor raise?
Tip 3: Test your understanding. Try to explain the deliverable to a colleague in two minutes. If the manager can’t do this, it means they haven’t fully grasped the content well enough to convey it effectively.
Real-world example. An HR manager uses AI to draft a skills development plan. The document is well-structured, and the key areas are coherent. But upon reviewing it with a critical eye, she realizes that the AI suggested standard training programs without taking into account an ongoing restructuring plan that completely changes the priorities. Without this critical thinking, she would have presented a plan that was out of touch with reality.
Actionable lever: the 4-question "challenge" checklist after each AI deliverable. Can I explain the logic behind each recommendation? Is the data verifiable? What did the AI not take into account? Would I feel comfortable defending this document in a meeting without any supporting materials?
From ad hoc assistance to a structured delegation
The real paradigm shift comes with AI agents, which are capable of performing complete tasks semi-autonomously. Managers no longer ask a question; instead, they delegate a scope of work. This requires knowing how to brief an agent just as one would brief a junior employee: a clear objective, scope of work, quality criteria, and procedures for review and approval before any content is published.
Without this framework, the agent works quickly but without a safety net. And mistakes aren't immediately apparent, because the form is always flawless.
At the same time, AI is creating divisions within the team. Some embrace it quickly, while others hold back out of fear or a lack of concrete use cases. The manager’s role is to harmonize its use without stigmatizing anyone, by identifying the real barriers and creating the conditions for everyone to progress at their own pace.
Real-world example. In a sales team of eight people, the manager notices that two salespeople are producing significantly more effective proposals. He asks them to share their method with the rest of the team. After a 30-minute session, the overall quality of the proposals improves. This wasn’t a matter of sales skills. It was a matter of clarity in the request made to the AI.
A Practical Tool: Three Rituals to Recommend to Your Managers.
These rituals are part of a broader challenge: that of organizational structure. How can we establish a common framework? What competencies should be included in managerial standards? NUMA has developed a three-tiered model (manager, team, organization) accompanied by checklists of specific questions to help position your company.
True competence isn’t about knowing what can be automated. It’s about knowing what it’s strategically important to keep in human hands: sensitive decisions, development discussions, delicate feedback, and decisions that shape relationships with clients or colleagues. Anything involving nuance, emotional context, or direct responsibility must remain within the manager’s purview.
Theimpact of technological transformation on the role of managers is already evident at companies like Doctolib and L'Oréal: managers are becoming key agents of change, focused on people and skills development.
Managers must also be mindful of a side effect of this acceleration: the cognitive load increases. The faster deliverables come in, the more there is to review, verify, evaluate, and decide. The pace intensifies without any change in staffing levels. If this overload is not managed, the team becomes exhausted and the quality of decisions declines.
Finally, with AI agents, work is being redistributed. Some tasks are being automated, while others are shifting toward supervision and decision-making. Managers no longer simply oversee operations. They now steer their team’s skill development: identifying which skills are becoming less useful, which are becoming critical, and how to help each member grow so they remain relevant.
Real-world example. A manager notices that his team is processing twice as many cases since the introduction of AI, but that the error rate has increased. His team members are submitting work without proofreading it. He implements a cross-checking process before submission. The volume drops slightly, but quality improves and customer feedback gets better.
Actionable lever: the 4-quadrant "human/AI" matrix to be developed with each team. What AI does autonomously (automation). What AI prepares and humans validate (assistance). What remains entirely human (non-negotiable). What was automated but is being returned to human handling (de-automation). This matrix should be reviewed every quarter.
Identify what is permitted, sensitive, and prohibited
Without an explicit framework, practices develop informally. Managers are left to decide on their own about issues that fall under company policy: Can AI be used to draft a customer email? To prepare an annual performance review? To draft a contract? In the absence of clear rules, each team makes up its own, practices diverge, levels of vigilance vary from one manager to another, and HR deals with risky situations after the fact. Setting these limits upfront protects managers as much as it does employees.
An often-overlooked prerequisite: the organization must be well-documented for AI to function effectively. If processes aren’t written down, if decisions aren’t tracked, and if the “why” behind those choices isn’t documented anywhere, AI operates blindly and produces outputs that appear rigorous but lack a solid foundation.
Measure the impact, not the volume
Adoption metrics indicate whether AI is being used. They do not indicate whether it is useful. The relevant indicators focus on the concrete impact AI has on the quality of work and decision-making.
Real-world example. After six months, an operations manager notices that the team most active in AI is also generating the most customer back-and-forths. The heavy use of AI was masking a problem with the clarity of the briefs. He replaces these metrics with the first-submission approval rate, average decision time, and internal customer satisfaction.
Actionable steps: three questions to help you define your framework. What is non-negotiable in your company (customer data, intellectual property, compliance)? Who approves sensitive deliverables before they are released? How do you measure whether AI improves the quality of work, not just its speed?
AI does not replace managers. It amplifies what they already are. If the request is vague, AI produces vagueness more quickly. If the thinking is sound, it accelerates it.
The challenge for HR is not to deploy AI, but to develop the skills that enable managers to steer it: critical thinking, agent management, human/AI regulation, anticipating roles, and organizational framework. These skills are not found in most managerial frameworks today. Companies that integrate them now are building a sustainable advantage. To go further, the NUMA AI and Critical Thinking Workshop offered by NUMA helps embed these reflexes directly in real-world situations.
Your managers are already using AI. Meeting summaries, feedback preparation, and briefing notes: what used to take half a day can now be done in a few minutes. But this speed creates an illusion. Just because a deliverable is well-written doesn’t mean it’s accurate or appropriate for the situation. The real risk isn’t that managers aren’t using AI. It’s that they’re using it without adjusting their approach or questioning what it produces.
This article is intended for HR directors, training managers, and L&D teams who want to support their managers in usingAI and management in their day-to-day work. By the end of this article, you’ll know which skills to focus your training efforts on and which practices to prioritize.
Why starting with tools is a false start
Most companies approach AI through tools and prompts. That’s useful, but it’s a blind spot. The real issue is managerial. When AI speeds up production, what used to define a manager’s value—producing quickly, synthesizing effectively, and writing clearly—is now handled by machines. What remains irreplaceable is the ability to ask the right questions, make decisions, and stay the course.
In the age of AI, a manager who doesn’t adapt will not stay at the same level. They will fall behind. Their team and leadership are getting used to faster, more structured deliverables. Expectations are automatically rising. The top managerial skills for 2026 confirm this shift: what defines a manager’s value is shifting toward judgment, decision-making, and human support. The bar for what is considered “good management” is rising without warning.
Real-world example. A marketing manager sends an AI-generated briefing memo to the executive committee. During the meeting, a question throws him off balance: “What is your pricing assumption based on?” He doesn’t know how to answer because he hasn’t reconstructed the logic behind the document. The deliverable was sound. The preparation was not. This is exactly the type of situation HR must anticipate in their support systems.
A Practical Tool: The 5-Question Mini-Checklist to Incorporate into Your Management Training. Before using AI on any high-stakes issue, ask yourself: What is the objective? What decision should this output help inform? What is the level of risk? Who will be affected? What quality criteria are non-negotiable?
AI generates content that is fluid, well-structured, and seemingly convincing. That is precisely what makes it risky: a well-written output lulls people into a false sense of security. Managers who do not systematically challenge what AI produces end up validating assumptions they have not verified and recommendations they could not defend. This is likely the most underestimated skill today in managerial frameworks. This was also the observation made by Mirakl and Servier at our annual FORWARD 2026 conference: train in judgment, not in the tool.
Critical thinking applied to AI is not just an intellectual exercise. It is a daily practice built around three concrete habits.
Rule 1: Question the sources. AI doesn’t always cite its sources. Asking “Where does this figure come from?” or “What data does this conclusion rely on?” should be second nature before sharing any information. A manager who shares a recommendation without knowing its basis risks losing credibility with their team and senior leadership.
Reflex 2: Look for what’s missing. AI responds to what you ask it. It doesn’t point out the questions you forgot to ask. What blind spots does this analysis leave? What perspective isn’t represented? What objection might an interlocutor raise?
Tip 3: Test your understanding. Try to explain the deliverable to a colleague in two minutes. If the manager can’t do this, it means they haven’t fully grasped the content well enough to convey it effectively.
Real-world example. An HR manager uses AI to draft a skills development plan. The document is well-structured, and the key areas are coherent. But upon reviewing it with a critical eye, she realizes that the AI suggested standard training programs without taking into account an ongoing restructuring plan that completely changes the priorities. Without this critical thinking, she would have presented a plan that was out of touch with reality.
Actionable lever: the 4-question "challenge" checklist after each AI deliverable. Can I explain the logic behind each recommendation? Is the data verifiable? What did the AI not take into account? Would I feel comfortable defending this document in a meeting without any supporting materials?
From ad hoc assistance to a structured delegation
The real paradigm shift comes with AI agents, which are capable of performing complete tasks semi-autonomously. Managers no longer ask a question; instead, they delegate a scope of work. This requires knowing how to brief an agent just as one would brief a junior employee: a clear objective, scope of work, quality criteria, and procedures for review and approval before any content is published.
Without this framework, the agent works quickly but without a safety net. And mistakes aren't immediately apparent, because the form is always flawless.
At the same time, AI is creating divisions within the team. Some embrace it quickly, while others hold back out of fear or a lack of concrete use cases. The manager’s role is to harmonize its use without stigmatizing anyone, by identifying the real barriers and creating the conditions for everyone to progress at their own pace.
Real-world example. In a sales team of eight people, the manager notices that two salespeople are producing significantly more effective proposals. He asks them to share their method with the rest of the team. After a 30-minute session, the overall quality of the proposals improves. This wasn’t a matter of sales skills. It was a matter of clarity in the request made to the AI.
A Practical Tool: Three Rituals to Recommend to Your Managers.
These rituals are part of a broader challenge: that of organizational structure. How can we establish a common framework? What competencies should be included in managerial standards? NUMA has developed a three-tiered model (manager, team, organization) accompanied by checklists of specific questions to help position your company.
True competence isn’t about knowing what can be automated. It’s about knowing what it’s strategically important to keep in human hands: sensitive decisions, development discussions, delicate feedback, and decisions that shape relationships with clients or colleagues. Anything involving nuance, emotional context, or direct responsibility must remain within the manager’s purview.
Theimpact of technological transformation on the role of managers is already evident at companies like Doctolib and L'Oréal: managers are becoming key agents of change, focused on people and skills development.
Managers must also be mindful of a side effect of this acceleration: the cognitive load increases. The faster deliverables come in, the more there is to review, verify, evaluate, and decide. The pace intensifies without any change in staffing levels. If this overload is not managed, the team becomes exhausted and the quality of decisions declines.
Finally, with AI agents, work is being redistributed. Some tasks are being automated, while others are shifting toward supervision and decision-making. Managers no longer simply oversee operations. They now steer their team’s skill development: identifying which skills are becoming less useful, which are becoming critical, and how to help each member grow so they remain relevant.
Real-world example. A manager notices that his team is processing twice as many cases since the introduction of AI, but that the error rate has increased. His team members are submitting work without proofreading it. He implements a cross-checking process before submission. The volume drops slightly, but quality improves and customer feedback gets better.
Actionable lever: the 4-quadrant "human/AI" matrix to be developed with each team. What AI does autonomously (automation). What AI prepares and humans validate (assistance). What remains entirely human (non-negotiable). What was automated but is being returned to human handling (de-automation). This matrix should be reviewed every quarter.
Identify what is permitted, sensitive, and prohibited
Without an explicit framework, practices develop informally. Managers are left to decide on their own about issues that fall under company policy: Can AI be used to draft a customer email? To prepare an annual performance review? To draft a contract? In the absence of clear rules, each team makes up its own, practices diverge, levels of vigilance vary from one manager to another, and HR deals with risky situations after the fact. Setting these limits upfront protects managers as much as it does employees.
An often-overlooked prerequisite: the organization must be well-documented for AI to function effectively. If processes aren’t written down, if decisions aren’t tracked, and if the “why” behind those choices isn’t documented anywhere, AI operates blindly and produces outputs that appear rigorous but lack a solid foundation.
Measure the impact, not the volume
Adoption metrics indicate whether AI is being used. They do not indicate whether it is useful. The relevant indicators focus on the concrete impact AI has on the quality of work and decision-making.
Real-world example. After six months, an operations manager notices that the team most active in AI is also generating the most customer back-and-forths. The heavy use of AI was masking a problem with the clarity of the briefs. He replaces these metrics with the first-submission approval rate, average decision time, and internal customer satisfaction.
Actionable steps: three questions to help you define your framework. What is non-negotiable in your company (customer data, intellectual property, compliance)? Who approves sensitive deliverables before they are released? How do you measure whether AI improves the quality of work, not just its speed?
AI does not replace managers. It amplifies what they already are. If the request is vague, AI produces vagueness more quickly. If the thinking is sound, it accelerates it.
The challenge for HR is not to deploy AI, but to develop the skills that enable managers to steer it: critical thinking, agent management, human/AI regulation, anticipating roles, and organizational framework. These skills are not found in most managerial frameworks today. Companies that integrate them now are building a sustainable advantage. To go further, the NUMA AI and Critical Thinking Workshop offered by NUMA helps embed these reflexes directly in real-world situations.
Discover all our courses and workshops to address the most critical management and leadership challenges.
Training organisation registered under the number 11755339875. This registration does not imply state approval.
.svg)
.png){kind=small}
.png){kind=small}
