When AI adoption becomes a management problem

BetterUp's recent AI management work uses a simple distinction that is useful because it is immediately graspable. Some employees are "Pilots": they approach AI with agency, optimism and a sense that the technology can extend what they are able to do. Others are "Passengers": they feel carried along by the change, less confident, and less able to turn AI into practical advantage.^{1, 2}

It is a neat analogy. It is also more than a neat analogy.

For decision-makers, the important signal is not the label itself. It is what the label reveals about the next phase of AI adoption. The first phase was largely about access: who had the tools, who had permission, who had a login, and who was experimenting. The next phase is about behaviour: who can turn access into useful work, repeatable practice and better judgement.

That is a management problem as much as a technology problem.

Generative AI adoption has moved fast enough that many organisations no longer have the luxury of treating it as an abstract future issue. Bick, Blandin and Deming's NBER work reported that, by late 2024, nearly 40 per cent of the US population aged 18 to 64 used generative AI, with 23 per cent of employed respondents using it for work at least once in the previous week. McKinsey's 2025 workplace AI report presents a related tension: almost all companies are investing in AI, but only 1 per cent of leaders describe their companies as mature in deployment.^{3, 4}

That gap — between widespread experimentation and mature operational use — is where the management question sits.

The easy answer is to buy more tools, run more training, or encourage more experimentation. Those may all be necessary. But they do not, by themselves, explain why one employee uses AI to redesign a recurring task while another avoids the same tool, quietly waits for instruction, or uses it only to polish an email.

The difference is not just capability. It is confidence, permission, judgement, and managerial context.

Access is not adoption

One danger in AI strategy is mistaking tool availability for organisational change.

A business can provide a sanctioned AI assistant, issue guidance, publish a policy, and run introductory training. That may produce visible adoption statistics. It does not necessarily produce changed work.

The stronger question is not "Do our people have AI?" It is: "Can our people identify where AI belongs in their work, use it responsibly, improve the workflow, and know when not to use it?"

BetterUp's Pilots-and-Passengers frame is useful because it gives senior leaders a vocabulary for that distinction. A Pilot is not merely an employee who has tried an AI tool. A Pilot is someone with enough agency to test, adapt, question and steer. A Passenger may have the same access, but less confidence or less belief that the tool can be used constructively.^{1, 2}

That matters because the next commercial advantage is unlikely to come from first-contact novelty. Many organisations have already passed the stage where ChatGPT, Copilot, Claude or Gemini feel like exotic novelties. The emerging question is whether employees can find the awkward, repetitive or knowledge-heavy edge cases where AI can become part of a real workflow.

Tool access gets people into the cabin. Management determines whether more of them learn to fly.

A small management scene

In this illustrative scenario, consider a regional professional services firm with 120 employees. The firm has paid for AI licences, issued a sensible policy, and invited staff to use AI for research, drafting, summarisation and meeting preparation.

Six months later, senior management sees uneven results.

One team has quietly built a repeatable process for preparing client briefing notes. A senior associate uses AI to assemble first-pass research questions, a junior team member checks sources, and the partner uses the output to prepare sharper client calls. The process saves time, but more importantly it improves consistency.

Another team, with similar work and the same tools, has barely changed. Staff use AI occasionally, mostly for cosmetic rewrites. The manager is nervous about errors, gives vague warnings, and has not defined which parts of the work can be safely assisted. Nobody wants to be the person who gets AI use wrong.

The technology is identical. The managerial environment is not.

That is the issue BetterUp's analogy helps expose. AI adoption does not spread evenly through a company because software has been purchased. It spreads through teams in which people feel authorised, supported and sufficiently skilled to turn experimentation into judgement.

The manager becomes the adoption layer

BetterUp argues that managers are a major lever in shifting employee mindsets, and its deck claims that Pilot-minded managers are associated with more Pilot-minded teams. As a vendor of management development and coaching, BetterUp has a commercial interest in that conclusion, so it should not be swallowed whole without scrutiny. But the underlying direction is plausible and commercially important.²

Managers are where AI strategy becomes daily work.

They decide whether AI use is treated as cheating, risk, experimentation, efficiency, professional development, or simply another tool. They decide whether staff are punished for small mistakes or helped to build better review habits. They decide whether AI is discussed openly in team meetings or hidden as private productivity theatre. They decide whether employees are expected to follow a procedure or improve one.

This is why the management layer matters. AI adoption is not only a question of individual enthusiasm. It is shaped by the signals employees receive from the people who allocate work, judge performance, reward initiative and define acceptable risk.

Microsoft's 2025 Work Trend Index points in the same broad direction from a different angle. Its "Frontier Firm" framing describes organisations with broad AI deployment, advanced maturity, current and planned agent use, and a belief that agents are central to ROI. It also argues that organisations must rethink knowledge work rather than merely add AI to existing workflows.⁵

That is a more structural view than BetterUp's mindset framing, but the two ideas meet in the same place. If AI changes work, someone has to redesign the work. In most businesses, that "someone" is not an abstract transformation office. It is the manager who decides how tomorrow morning's work is actually done.

The paradox: AI may make management more human

One of the more interesting claims in BetterUp's deck is that employees may prefer AI for some routine supervisory tasks, while still preferring human managers for motivation, feedback, recognition and coaching. In the deck, BetterUp presents employee preferences suggesting that performance tracking and deadline management may be more acceptable to automate than the relational parts of management.²

This is not proof that AI should become the office supervisor. It is a signal that the managerial job may be splitting.

Routine coordination, monitoring, summarising and reminder work may become increasingly machine-assisted. The human value of the manager then shifts towards judgement, trust, motivation, feedback, interpretation and exception-handling.

This is where a shallow automation story becomes misleading. If AI removes some managerial administration, it does not automatically create better management. It creates the opportunity for better management. The opportunity is only realised if managers know what to do with the time, visibility and leverage that AI gives them.

Microsoft's discussion of human-agent teams makes this point indirectly. It suggests that some work will involve humans setting direction while agents perform tasks or workflows, with humans checking in as needed. It also notes the need to get the "human-agent ratio" right, because too much or too little automation can create different risks.⁵

That is a management question, not a procurement question. It requires someone to know which work can be delegated, which work needs human judgement, which errors matter, and where the employee remains accountable.

What the BetterUp frame gets right — and where to be careful

BetterUp's Pilots-and-Passengers analogy is strong because it converts a diffuse adoption issue into a memorable management distinction. It helps leaders ask a better question: not simply "Who is using AI?" but "Who is using AI with agency?"

That is useful.

The caution is that the analogy can become too tidy. Employees are not fixed types. A person may be a Pilot in one workflow and a Passenger in another. A confident marketer may use AI fluently for campaign variants but avoid it entirely for data analysis. A finance manager may be cautious not because they lack agency, but because the error cost is real. A junior employee may appear passive because the organisation has not made it safe to experiment.

The better use of the analogy is therefore diagnostic, not judgemental.

It should not be used to divide the workforce into heroes and laggards. It should be used to ask what conditions make intelligent AI use more likely. Where are employees confident? Where are they anxious? Where is the workflow clear? Where is review built in? Where does the manager model good AI use? Where is the employee left to guess?

The original BetterUp framing is a useful lens. The Andrew's Intel interpretation is that "Pilot behaviour" is not merely a personality trait. It is an organisational outcome.

The edge case that becomes the workflow

This is where the issue connects to the larger AI market.

Many of the most commercially significant AI use cases will not arrive as dramatic executive announcements. They will begin as small edge cases. One employee works out how to prepare a better first draft of a board note. Another creates a safer process for comparing supplier proposals. A manager uses AI to turn call notes into structured follow-up actions. A marketing team moves from occasional copy generation to a repeatable content-testing workflow.

At first, these look like isolated experiments. Then they become habits. Then they become expected ways of working. Eventually they become part of the operating model.

The difference between a casual AI user and a useful AI adopter is often the ability to recognise that transition. Managers sit close enough to the work to see it early. Senior leaders often do not.

That makes middle management newly important. In some companies, managers may become the weak link: too busy, too anxious, too underskilled or too unclear about AI to help teams change. In others, managers may become the adoption accelerant: the people who translate broad AI ambition into practical, safe and repeatable work.

Meeting-ready artefact: the AI adoption management test

A senior team considering AI adoption should ask these questions before assuming the problem is solved by more software or more training.

Questions to ask vendors, consultants or internal AI teams

If a vendor, consultant or internal programme promises stronger AI adoption, decision-makers should ask:

• Are you helping us increase tool usage, or helping us change work?
• How do you distinguish confident adoption from careless adoption?
• What role do frontline managers play in your rollout model?
• How will we identify employees who are anxious, passive or unclear rather than merely "resistant"?
• What evidence will show that AI is becoming part of a repeatable workflow?
• Which tasks should remain human-led because they require trust, judgement, accountability or emotional intelligence?
• How do we prevent AI adoption from becoming hidden individual productivity theatre rather than shared organisational capability?
• What will managers need to stop doing if AI takes over some routine coordination work?
• How will we know whether our people are becoming better Pilots, not merely more frequent users?

What to watch next

The signal to watch is not simply whether AI tools become more powerful. They will.

The more interesting signal is whether organisations build the managerial capability to absorb that power. This includes AI literacy, workflow redesign, review discipline, employee confidence, role clarity and the ability to decide where human judgement remains essential.

McKinsey's workplace AI report argues that the barrier to scaling is less about employee readiness than leadership's ability to steer. Microsoft's Frontier Firm work points towards organisations that move beyond individual experimentation and begin rebuilding work around AI and agents. BetterUp's Pilots-and-Passengers analogy adds the human adoption layer: employees need enough agency and optimism to use AI as something they can steer, not something being done to them.^{4, 5, 1}

Taken together, the larger point is clear enough for decision-makers.

The next AI bottleneck may not be model capability. It may not even be tool access. It may be the ability of managers to turn uncertainty into agency, and scattered experiments into repeatable work.

That is why the Pilots-and-Passengers analogy is useful. It is not because every employee can be neatly classified. It is because the analogy asks the right management question.

In the age of AI, who is steering?