The Reasonable Fear
The anxiety about AI and leadership comes in two versions, and they need different responses.
The first is personal: what were you actually good at, and is that still valuable now that a model can do the analytical work faster than you can?
The second is a leadership question: what do you tell the people you're responsible for about where human value sits now, and how do you answer them honestly when you're not certain yourself?
These legitimate questions are being answered, in most organisations, with platitudes. "AI augments, it doesn't replace" is technically accurate in some contexts, yet it's mostly unhelpful, because it says nothing about which specific parts of a role are being augmented, which are being replaced, and which aren't at risk. A leader who can't answer those questions can't help a team navigate them either.
The displacement isn't theoretical. Goldman Sachs research puts current AI-driven job losses in the US at around 16,000 a month, concentrated in entry-level, rule-bound roles such as data entry, admin, and customer service1. Understanding where that displacement is concentrated matters more than knowing whether it's happening at all, because it makes it possible to give a specific, honest account of what's changing and what isn't, both to yourself and to the people who are relying on you for one.
Complicated Problems, Complex Problems
Arthur Brooks, writing in Harvard Business Review in April 2026, states the central distinction clearly: "AI can solve complicated problems, but leadership is about complex ones."2
A complicated problem has many variables, many steps, and a discoverable correct answer. Given sufficient data and processing capacity, there's a right path through it. Demand forecasting based on historical data is complicated. Financial modelling with dozens of input variables is complicated. Identifying operational inefficiencies across a business unit is complicated. These are demanding problems; they require expertise and careful work. But they reward speed, pattern recognition, and consistency, which is precisely where AI now has an advantage.
A complex problem has no discoverable correct answer. It's ambiguous. It involves people and values. Deciding whether to restructure a team to protect financial stability is complex. Choosing between two strategic directions when the evidence is mixed is complex. Navigating the realisation that the commercially correct option conflicts with what the organisation claims to stand for is complex.
The distinction isn't about how difficult a problem feels. Some complicated problems take enormous effort to work through. Some complex ones resolve quickly when the right person with the right context applies their judgement. The difference is structural: a complicated problem has a best answer waiting to be found; a complex problem requires a judgement call, made by someone prepared to be accountable for it.
This matters for leadership because AI's advance is concentrated on complicated work specifically. Leadership isn't safe from disruption, but the disruption is targeted. It's taking the work that rewards processing speed and analytical thoroughness. It's leaving the work that requires tolerance of ambiguity, contextual understanding, and accountability for decisions no dataset can fully evidence.
The Shifting Foundation
Much of what has counted as strong leadership in analytical and strategic roles, in practice, has been sophisticated complicated-problem-solving. Thorough research. Comprehensive plans. Being first to the right answer. These have been signals of competence for a long time. They were also useful: complicated work matters, and doing it well creates value. The current shift is that the premium on having a senior human do it is eroding, because it can now be done faster and more comprehensively by a well-configured tool.
Leaders who built their credibility there are watching the ground shift. But was that ground ever where leadership's actual value was located, or was it a proxy?
What doesn't get displaced is a different kind of work. Making a judgement call when multiple defensible options exist with little-to-nothing to separate them. Making a result legible to the people who have to act on it, in a way that generates conviction rather than compliance. Holding an organisation to what it says it values when the efficient path would cost it something important. Deciding which problems deserve to be solved before any tool gets pointed at them.
For leaders whose credibility already sat in this territory, the shift improves their position. The analytical layer beneath them is becoming cheaper and faster; their work becomes more consequential by comparison. For leaders whose credibility sat primarily in the analytical layer, the reframe is less comfortable: the thing being automated was never the deepest source of their value, even if it was the most visible one.
What the New Work Looks Like
The leadership work that AI isn't competing for shows up in moments that are identifiable in any organisation.
There's the moment when an analysis is complete and several options exist, with nothing to separate them. Someone has to choose, and carry accountability for the decision. That accountability doesn't transfer to AI.
There's the moment when a result needs explaining, not just reporting. A model's output is information; what a team needs is meaning. The work of making a forecast or a risk assessment legible in a way that produces understanding and commitment requires knowing the people, the history, and the context that the model doesn't hold.
There's the moment when the efficient option conflicts with what the organisation claims to value. Faster, cheaper, or more comfortable in the short term, but in tension with what leadership said the organisation stood for. Choosing the harder, values-consistent path, and doing it visibly enough that it holds, is human work.
There's the moment when nobody has yet identified which problem to solve. Problem-framing at strategic scale requires contextual perspective that precedes analysis. AI answers the questions it's given well; the question of which questions deserve to be asked doesn't come from the model.
These four moments sketch Mantage's working synthesis: arbitration between defensible options the data can't separate; sense-making for people who need meaning, not just output; stewardship of what the organisation claims to value when upholding it is inconvenient; and framing of which problem to solve before anything else begins. The territory isn't vague or soft. It's specific, it's identifiable, and it's what leadership was supposed to be about before it found that it could also be good at analysis.
Recognising the Shift in Your Own Role
One thing that's useful for any leader to know is how their current week is really being spent. The ratio of complicated to complex work is a better predictor of how exposed a role is to AI-driven change than any title or seniority level.
A short diagnostic:
- How much of what you did this week had a findable right answer, versus a judgement call another person could have made differently?
- If a tool did the complicated part of your job perfectly tomorrow, what would you spend the freed time doing?
- When did you last make a call because it was right, rather than because it was defensible?
- Do the people on your team know why a decision was made, not just what it was?
- If your confidence in your own value dropped when you imagined a tool doing your analysis faster than you, what does that say about where your value was really sitting?
Most leaders who work through these questions will find their week is more complicated-work-heavy than they'd expected. The change that leaders should be making is to redirect attention toward the work that was always theirs, and was always what determined whether the team's output was sound rather than just technically correct.
The same questions are useful to share with a team. The goal isn't to reassure people that nothing will change; it's to help them locate, specifically, which parts of their role sit in complicated territory and which don't, and to be honest with them about what that means.
The Honest Limits
There are a few things this reframing doesn't do.
It doesn't justify disengaging from technical fluency. Leaders still need enough understanding of what AI tools are doing to evaluate their outputs, recognise when something is wrong, and know when to override; evaluate rather than generate.
It doesn't lower the bar on the work that remains. Complex work done badly is as costly as complicated work done badly used to be. Poorly grounded, indecisive judgement calls that can't withstand reasonable scrutiny are a different category of leadership failure, not a lesser one. The change of focus raises what's expected of judgement; it doesn't make the job easier.
The shift is also gradual. Most leaders will continue doing both kinds of work for years. The point isn't to stop doing complicated work immediately, but to be deliberate about where effort goes and which work is generating the value.
The Stronger Ground
Leadership's value was never supposed to be located in complicated work. That work mattered, and doing it well was useful. It was just the easiest part of the job to point to, not the reason the job mattered.
What AI is doing, at scale and with increasing speed, is making that difference visible. The complicated work it's absorbing was always going to be done better by something faster and more consistent. That doesn't mean the complex work AI can't reach has grown scarcer in itself. What's changed is the layer beneath it: complicated work is now cheap and abundant, so complex work is what's left carrying the premium.
Leaders who were already doing that work find themselves in a stronger position than they may realise; leaders who were doing complicated work and calling it leadership are facing a tougher reckoning. They aren't redundant: the part of what they do that's genuinely irreplaceable was always there. It just needs finding.
References
- David Curry, "AI Is Slashing 16,000 Jobs a Month in the US (Gen Z Hit the Hardest)", TechRepublic, 8 April 2026.
- Arthur C. Brooks, "What AI Can't Do: The New Job of Leadership", Harvard Business Review, 8 April 2026.
Ady Coles works as a thinking partner and mentor to leaders and teams navigating complexity. His work centres on judgement, perspective, and the often-invisible work of translation - helping people understand their role in the system, make better decisions, and operate with confidence in uncertain environments.