The AI skills gap starts at the top

The phrase "AI skills gap" almost always points downward, at the front line, the staff who need training, the people who have not caught up. It is a comfortable framing for leaders, because it makes the gap someone else's. The evidence points the other way. The most important AI skills gap in most organisations is at the top, and naming it honestly is the first step to closing it.

The numbers are striking. In the 2024 Work Trend Index, a survey from Microsoft and LinkedIn (and so a vendor source, worth reading with that in mind), 75% of knowledge workers said they already use AI at work, and 78% of those were bringing their own tools rather than waiting for the organisation. Yet 60% of leaders said their company had no vision or plan for AI, even while 79% agreed it was critical to stay competitive. The workforce is not the bottleneck. The direction is.

Willing staff, absent direction

This matters because employees cannot adopt AI intelligently in a vacuum. When no one has said what AI is for, what is permitted, or how success will be judged, people do the only thing they can: they improvise privately, which is how an organisation ends up with energy but no coherence. The MIT Media Lab's GenAI Divide report concluded that what separated the organisations getting value from those not getting it was not talent or infrastructure, but learning, integration and adaptation. All three are leadership responsibilities, not front-line ones.

There is even a regulatory echo. The EU AI Act places its AI literacy duty on the organisations deploying AI, not on individual employees. Literacy is treated as something a business owes its people, set from the top. The same logic runs through the NIST AI Risk Management Framework, whose "Govern" function is explicitly about leadership setting the culture and direction that everything else depends on.

The cost of leaving this gap unfilled shows up everywhere else, usually wearing a different name. It is the unofficial, unmanaged AI use that appears when there is no sanctioned route. It is the training that does not stick because nothing around the work actually changed. It is the promising pilot that never scales because no one with authority decided it should. Each of these tends to be treated as a separate problem with its own fix. More often they are the same problem seen from below: a direction that was never set, and permission that was never clearly given.

Four leadership obligations

If the gap is at the top, so is the work. We describe it as four obligations a leadership team owes the organisation. None of them require a leader to be a technical expert.

Establish direction. Say what AI is for here, and what it is not. A sentence that names the goal and a short list of where AI is welcome does more for adoption than any tool purchase. Without it, every team guesses, and the guesses do not add up.

Model responsible use. Leaders set the norm by what they do, not what they say. A leader who uses AI visibly, and visibly checks its output, teaches the organisation that this is serious work done with judgement. A leader who waves it away as something for "the young people" to handle has quietly abdicated the judgement the role exists to provide.

Create permission to experiment. People are already experimenting, just out of sight. A sanctioned, safe way to try things, with clear boundaries about data and approval, brings that energy into the open where it can be shared and made safe. Without permission, the experimenting does not stop; it just goes underground, taking the company's data with it.

Recognise and spread useful practices. A good use found by one person is worth little until it becomes something the organisation owns. Leaders close the loop by harvesting what works into shared playbooks and giving credit to the people who find it. This is how scattered cleverness becomes capability.

Teams cannot adopt what leaders cannot direct.

What this looks like in practice

Picture a professional services firm where the partners have decided AI is an associate matter and stayed clear of it themselves. The associates are capable and willing, so plenty happens, but it is fragmented, unshared, and quietly risky, and no partner can say whether any of it serves the firm's priorities. The gap here is not associate skill. It is partners who have not set a direction, modelled the behaviour, or built a way to share what works.

Contrast a firm where one partner owns AI as a real part of their role: sets a direction, uses the tools visibly on their own work, sanctions a few experiments with clear data rules, and runs a short monthly session on what worked. The associate talent is identical. The results are not, because the leadership obligations were met. (Both are illustrative examples, not specific firms.)

The honest limits

Two cautions. First, closing the gap at the top does not mean leaders becoming technical experts, and it certainly does not mean leadership doing the hands-on work. It means judgement and direction: knowing enough to set sensible boundaries and recognise good practice, which is a different and more durable skill.

Second, direction can curdle into control. A leadership team that responds to this by mandating one tool and one rigid process kills the very experimentation that surfaces the best uses. The aim is direction plus permission, not direction plus a clampdown. Set the destination and the guardrails; leave room for people to find the road.

What to do next

Turn the four obligations on yourselves before turning anything on the staff. As a leadership team, ask honestly: have we set a direction, do we model use, have we given permission, and do we share what works? The answers are usually uncomfortable, and that discomfort is the most useful thing a leadership team can sit with, because it points straight at the real gap.

The tool

To make that honest, we have built the AI Leadership Readiness Checklist: a short, scored checklist for a leadership team across direction, behaviour, permission, governance and learning, with a plain reading of where you stand and which obligation to address first.

Download the AI Leadership Readiness Checklist (PDF)

Building a leader's own confidence with AI is exactly what our AI lessons for leaders are for: practical, one-to-one, and shaped around your real work rather than a syllabus. It connects to setting an AI strategy worth the name and to the durable literacy that lets a leader direct without needing to become an engineer.

Sources and further reading

• Microsoft and LinkedIn, 2024 Work Trend Index. Vendor survey. Source for 75% of knowledge workers using AI, 78% bringing their own tools, 60% of leaders reporting no AI plan, and 79% calling AI critical to competitiveness. Read as vendor research, not independent measurement.
• MIT Media Lab (Project NANDA), The GenAI Divide: State of AI in Business 2025. Industry report, not peer-reviewed; the 95% figure has been contested. Source for the divide being about learning and integration, not talent.
• EU AI Act, Article 4: AI literacy. Applicable from 2 February 2025. Source for AI literacy being an organisational duty.
• NIST AI Risk Management Framework. Independent US standards body. Source for the leadership-level "Govern" function.