What happens when your team includes machines that sound like colleagues but have no stake in the outcome

Every leadership development program I have ever seen was designed for the same basic assumption: your team is made of people.

People who get tired. People who have egos. People who need motivation, clarity, recognition, and trust. People who push back, burn out, grow, and leave.

That assumption is breaking.

Today, a growing number of teams include AI agents. Not tools that sit in a toolbar waiting to be clicked. Agents that plan, reason, execute, and deliver, sometimes with no human in the loop at all. Seventy-six percent of executives now describe AI as more like a coworker than a tool. By 2027, half of the companies using generative AI will have launched agentic applications capable of complex work with limited oversight.

This is not a future scenario. It is a current operating condition. And leadership development has not caught up.

We have been training leaders to lead people. We have never trained them to lead a team where some of the members have no skin in the game.

The trust calibration problem

When a human colleague is uncertain, you can read it. The hesitation in their voice. The qualifying language. The slight pause before they commit to a recommendation. These are signals leaders have spent their careers learning to interpret.

AI does not produce those signals.

An AI system delivers a hallucination with the same fluency and confidence as a verified fact. There is no vocal tremor, no hedging, no body language to decode. The output sounds authoritative, whether it is brilliant or fabricated.

This creates a leadership challenge with no precedent. Leaders must now maintain critical judgment in the presence of articulate, authoritative-sounding output that may be completely wrong. No previous tool in the history of work has been this persuasive. A spreadsheet with a formula error looks like a spreadsheet with a formula error.

An AI-generated analysis with a foundational flaw looks like a polished, well-reasoned brief.

The leadership skill this demands is trust calibration: the ability to dynamically assess the reliability of AI output, to know when to lean on it and when to override it, and to teach your team to do the same.

It is not a technology skill. It is a judgment skill. And almost no one is teaching it.

The relationship that only goes one way

Here is something most organizations are not ready to discuss: people form emotional relationships with AI systems.

They thank them. They apologize to them. They feel guilt when they are curt. They develop preferences and even loyalty to specific AI assistants. Research shows this is not a fringe behaviour; it is a predictable human response to interacting with something that communicates in natural language, remembers context, and responds to your needs.

But the relationship is asymmetric. The AI has no stake. It does not care about the outcome. It does not feel invested in the team, the project, or the mission. It will give you the same quality of attention whether you are making a decision that affects ten thousand employees or choosing a font for a slide deck.

Leaders need to understand this dynamic because it shapes how their teams make decisions. A person who feels emotionally connected to an AI recommendation is less likely to challenge it. A team that has come to rely on AI as a pseudo-colleague will experience genuine grief and disorientation when the system is changed, upgraded, or removed. This is not weakness. It is human nature. And leaders who do not account for it will be blindsided by reactions that seem disproportionate to what they think happened.

Delegation is no longer developmental

Every leadership framework I know teaches delegation as a developmental tool. You give someone a stretch assignment. You calibrate the level of autonomy to their readiness. You coach them through it. You build trust incrementally.

None of that applies to AI.

Delegation to an AI agent is categorical, not developmental. You do not coach the AI into readiness. You define the boundaries of its authority correctly, or you do not. And the failure mode is fundamentally different. A person who is in over their depth will usually slow down, ask questions, or signal distress. An AI agent will confidently execute beyond its competence without a flicker of hesitation.

This means the leadership skill is no longer "how do I delegate effectively?" It is "how do I define the precise conditions under which delegation to this system is safe, and how do I verify that the boundaries held after the fact?"

That is a design skill, not a relationship skill. It requires leaders to think like systems architects, even if they have never written a line of code.

That is a design skill, not a relationship skill. It requires leaders to think like systems architects, even if they have never written a line of code.

The accountability void

This is the one that will define the next decade of leadership.

When a person makes a bad decision, accountability is clear. When an AI makes a recommendation, a person acts on it, and the outcome is harmful, the chain of accountability fractures. Who is responsible? The person who deployed the system? The person who accepted the recommendation without verifying it? The leader who approved the workflow that put AI in the decision chain? The vendor who built the model?

Right now, most organizations have no answer. And the absence of an answer does not mean the absence of consequences. It means that when something goes wrong, the response will be reactive, political, and destructive to trust.

Leaders need to build accountability structures for human-AI decision chains before the failures happen, not after. This is not a compliance activity. 

It is a leadership identity question: am I responsible for outcomes I did not directly produce, but that occurred through a system I authorized and a workflow I designed?

The answer has to be yes. And leaders need to be taught how to own that.

Protecting the ability to think

When people use AI heavily, they offload not just tasks but cognition. Research has already documented reduced brain activation during AI-assisted problem-solving. The practical consequence is straightforward: over time, teams lose not only the ability to do the work the AI does for them, but the ability to evaluate whether the AI’s work is any good.

This is the dependency trap. And it is subtle, because the early symptoms look like success. Productivity goes up. Output quality appears to improve. Timelines compress. It is only when something goes wrong, when the AI produces a flawed output and no one on the team catches it, that the dependency becomes visible.

Leaders need to maintain what I would call cognitive sovereignty: the team’s capacity to think independently, even when they do not have to. This means deliberately preserving spaces where AI is not used. Where people reason through problems on their own. Where the slow, effortful work of human judgment is valued, practiced, and protected.

No leadership model has ever needed to address this. No previous tool was seductive enough to make people stop thinking.

When speed becomes a weapon

AI changes the power dynamics of every room it enters.

The person with the best AI fluency can produce in minutes what takes others hours. In a meeting, this creates a new kind of inequity: the person who generated a polished, AI-assisted analysis can dominate a discussion with volume and apparent rigour, while the quiet expert with decades of contextual judgment gets drowned out because their contribution takes longer to articulate and looks less impressive on a screen.

Leaders need to see this happening and actively counterbalance it. The most valuable human contribution in a hybrid team is not information, because AI has more, and it is not speed, because AI is faster. It is contextual judgment. The ability to say "that analysis is technically correct, but it will destroy trust with this stakeholder" or "that recommendation ignores the political reality of our restructuring."

Protecting and elevating that kind of contribution, even when it arrives more slowly and less polished than the AI-generated alternative, is a leadership responsibility that has no precedent.

What are leaders for?

This is the question underneath all the other questions.

When AI can write, analyze, strategize, code, design, and create, what is a leader’s distinctive value? For most leaders, their professional identity is built around being the smartest person in the room, the one with the sharpest analysis, the most compelling presentation, the most thorough preparation. AI has matched or exceeded those capabilities in a remarkable number of domains.

What remains is harder to name and harder to measure, but it is more important than everything AI can do: meaning-making, relationship-holding, judgment under ambiguity, and moral accountability. These are not the consolation prizes of the AI era. They are the core of leadership. They always were. But leaders were never taught to see them that way, because the analytical and executional skills were easier to develop, reward, and promote.

The transition from "I am valuable because of what I produce" to "I am valuable because of the judgment, meaning, and accountability I bring" is the deepest identity shift most leaders will face. And it is not happening in a weekend retreat. It is happening in real time, in the daily experience of watching AI do things they used to be proud of doing themselves.

We trained leaders for teams made of people. The team has changed. The leadership must change with it.

This is not a theoretical challenge. It is showing up now, in every organization where AI is woven into daily work. In the meetings where no one questions the AI-generated brief. In the teams where junior people outpace senior ones because of AI fluency, not judgment. In the quiet erosion of thinking that happens when the machine does the hard part.

The leadership development industry has not caught up. The models we use were designed for a world where every team member was human. That world is ending.

What comes next requires something different. Not better technology training. Not another change management framework. A fundamentally new set of capabilities for leading in a world where your team includes systems that are confident, tireless, and persuasive, but have no stake in the outcome, no contextual judgment, and no moral accountability.

That is the work. And it is urgent.

Donna Tulloch

Donna Tulloch is a Change Management Consultant, Coach, and the creator of the People-Change Intelligence (PcQ™) framework and the A.H.E.A.D. 2.0 AI-Empowered Change Leadership Model. She works with organizations navigating the intersection of AI transformation and human capacity through Pulse by DNK and Change Connection Lab.

www.pulsebydnk.com  |  [email protected]