The question is no longer "what can AI do?" The answer is now: almost everything we thought was the core of managerial work. Synthesis, analysis, reporting, recommendations, draft pitches, first-pass strategy. The real question has become: "what's left that AI can't do?"

Harvard Business Review's April 2026 dossier answers it, under the title "What AI Can't Do: The New Job of Leadership."

Five territories AI doesn't cover

1. Choosing the question. AI brilliantly answers a well-framed question. It's useless at deciding which question to ask. Identifying the right problem remains the leader's number one task.

2. Allocating capital with conviction. AI can produce a business case. It can't say "We're going. We're committing 30% of the annual budget to this bet." That signature stays human, and binding.

3. Sustaining trust over time. No model rebuilds trust between a team and a manager after a crisis. Humans trust humans, not protocols.

4. Holding accountability. An algorithm can't be fired. A leader can. That asymmetry changes everything we expect of leadership.

5. Giving meaning. AI can produce a speech. It can't make people believe it. Meaning always comes from someone taking a personal risk to carry it.

What this changes for leaders

The consequence is uncomfortable: you can no longer hide behind production. If AI builds the slide deck, the briefing, the market analysis, the board memo, then a leader's value sits entirely in decision and relationship. That is, in two things which, in most management cultures, are under-invested.

PwC's 29th Global CEO Survey 2026 puts it differently: 68% of CEOs admit that the number of meetings they hold no longer matches the number of decisions they make. Producing the appearance of decisions has become the first symptom of organisations lagging behind on AI.

The new leadership skills hierarchy

The World Economic Forum, in Future of Jobs Report 2023 and its 2025 update, consistently puts the same skills on top: analytical thinking, creative thinking, resilience, motivation, curiosity and lifelong learning. These are precisely the skills AI does not absorb.

The OECD, in its Skills publications, adds two often-forgotten dimensions: the ability to hold ambiguity for longer, and the ability to make a diagnosis intelligible to different audiences. Again, structurally human capabilities.

The 8% trap

The Gartner study relayed by Global Leaders Today — "Only 8% of HR leaders say managers skilled in using AI" — is a quiet thunderclap. If only 8% of organisations believe their managers know how to use AI, then the remaining 92% have managers who know neither what AI can do nor what it can't.

These are the same managers expected to lead the transition, support teams and explain what's changing. It's a bridge with no pillars.

The new leader equation for 2026

Here's how HBR sums it up, and it cuts: produce less, decide more. Comment less, arbitrate more. Validate less, commit more. Justify less, own more.

For many seasoned executives, that's an existential threat. Their original value — producing a clean deliverable, a beautiful slide, a well-structured memo — has become commodity. Their future value — deciding without hiding, arbitrating without diluting, committing without ducking — is precisely what they were never trained for.

The MOOOVE link

That's exactly the transformation GMHR (Grow My HR / Manager Operating Upgrade) installs: out of the "training theatre," into short cohorts, ready-to-use tools (HR Decision Playbook, Manager System Kit), and immediate field application. The goal isn't to learn how to use ChatGPT. It's to learn how to be a manager who decides, in an era where production is free.

The new job of leadership isn't to beat AI. It's to do what AI can't — and stop doing what AI does better than us.

Sources

• HBR: What AI Can't Do: The New Job of Leadership (April 2026).
• PwC: 29th Global CEO Survey 2026.
• Gartner / Global Leaders Today: only 8% of HR leaders say managers skilled in using AI.
• WEF: Future of Jobs Report 2023 (and 2025 update).
• OECD: Skills.