Setting targets, allocating tasks, monitoring the quality of work, correcting errors…
This sounds like a manager’s day-to-day routine. Yet, in the future, these tasks will no longer apply solely to human staff. They will also apply to AI agents capable of analysing information, carrying out multiple tasks in sequence and acting with a degree of autonomy.
So, can you really manage an AI? Yes, provided you do not treat it as a traditional employee. The manager’s role is instead to organise, supervise and ensure the smooth collaboration between humans and intelligent systems.
📌 Key takeaways
- Managing an AI involves overseeing a task, rules and outcomes, not managing a personality.
- The manager must decide which tasks to delegate to the AI and which to keep under human responsibility.
- AI agents require specific objectives, quality criteria and checkpoints.
- The management of teams comprising humans and AI agents is based on training, ethics and human supervision.
What is an AI agent? ✨
An AI agent is a system capable of pursuing an objective, handling several stages of a process and, depending on its level of autonomy, using various tools to carry out an action.
Unlike traditional generative AI, which primarily responds to a request, the agent can analyse a situation, propose a plan, carry out certain tasks and adjust its actions based on the results.
These capabilities are advancing rapidly. In 2026, AI agents achieved a success rate of around 66 per cent on OSWorld, a benchmark that assesses their ability to carry out practical tasks on a computer. This also means that they still fail in nearly one in three cases: autonomy therefore does not eliminate the need for human oversight.
Managing AI: what does that actually involve? ☀️
Managing an AI does not mean organising its annual appraisal or boosting its motivation! It is about providing it with a working framework that is clear enough to produce a reliable result.
In particular, the manager must:
Define a specific mission
AI works best when it knows the objective to be achieved, the context, the resources available, the constraints to be observed and the format of the expected output.
A vague brief often produces an approximate result. As with a human team, the quality of the brief directly influences the quality of the work.
Allocating roles between humans and AI
Not everything should be delegated to an AI agent. Repetitive, structured, high-volume or easily verifiable tasks are generally the most suitable.
Conversely, sensitive decisions, emotionally charged situations, ethical judgements and actions that directly engage the company’s liability must remain under human control.
Set performance criteria
The manager must define how the agent’s work will be assessed: accuracy, time saved, compliance, user satisfaction, the number of errors detected or the human rework rate.
The aim is not merely to check that the AI has produced an output, but to ensure that this output actually creates value.
Organising human supervision
The more actions agents carry out, the faster errors can accumulate. Microsoft therefore emphasises the need to appoint staff to assess agents’ performance, adjust their processes and capitalise on the lessons learnt. Indeed, the number of active agents in Microsoft 365 increased fifteen-fold between March 2025 and March 2026.
The manager becomes a human–AI conductor 🎻
The arrival of AI agents does not diminish the manager’s importance: it transforms their role.
They must understand the capabilities and limitations of each tool, support staff, prevent technological dependency and establish common rules. The most advanced teams are already formalising the handover between humans and AI, sharing their mistakes and collectively defining their quality standards.
The benefits can be significant: by 2026, 42 per cent of frontline staff who regularly use AI report saving around eight hours a week. However, many organisations are still struggling to turn this time saved into value, due to a lack of strategy and reorganisation of work.
Managing AI also means managing the risks involved 📈
Hallucinations, biases, the use of confidential data, opaque decisions or excessive automation: an AI agent must never become an uncontrolled black box.
This vigilance is also becoming a regulatory requirement. The European AI Act strengthens the requirements for risk management, traceability, transparency and human oversight. It also requires organisations to develop the AI skills of those who use or supervise these systems.
How can we prepare managers? ✨
The first instinct should not be to deploy agents everywhere, but to identify the uses that are genuinely relevant.
The Managing Human–AI Agent Teams training course helps managers define the roles of humans and AI, adapt their practices to their teams’ digital maturity, and ensure ethical and responsible use.
Because the real question is no longer simply: ‘What can AI do?’
It is now: how can we organise work so that human intelligence and artificial intelligence, together, deliver the best results?
