When AI takes over part of someone's job, the benefit is usually easy to see. Reports take less time, information is easier to find, and a first draft appears in minutes instead of hours. What's harder to see is what that work used to do for the person doing it. Some of the tasks AI absorbs were also how employees learned to notice problems, test their assumptions, and build the judgment that more senior roles depend on.

Business leaders are already treating this as more than a question of headcount. In KPMG's 2026 U.S. CEO Outlook, a survey of 400 U.S. chief executives, 86% said they are redesigning roles, career paths, and team structures to reflect human-AI collaboration, and 75% said AI has made them rethink the skills required for entry-level roles. In the same survey, 89% expect overall headcount to grow over the next three years, while 53% are planning reductions in some areas as part of their longer-term AI strategies. The total number of jobs may say less than how the work inside them is changing.

For HR and talent leaders, that changes the planning question. Beyond how many people the organization needs, it has to understand what people will do, which capabilities still matter, and how employees will develop those capabilities when some of the work that used to teach them is changing.

Look at how roles are changing, not just how many people you need

AI's effects are often easiest to see at the task level, even when its longer-term impact on entire roles remains uncertain. A tool might draft a report or sort incoming requests while responsibility for getting the decision right stays with a person. That makes the responsibilities inside a role a more useful thing to examine than the job title.

For any role AI is starting to touch, it helps to ask which responsibilities are being reduced, changed, or introduced, and what judgment the role still requires. The answers show where skill gaps are likely to open, including new knowledge people may need, such as recognizing where a particular AI tool tends to be wrong. Changes to how work moves between people and systems are a workflow question, which we cover in AI workflow redesign. The workforce question is what those changes mean for the people doing the work, and especially for how they learn.

Rethink how employees develop skills and experience

Much of the expertise inside any organization was built through routine work. People learned to spot a problem by handling hundreds of ordinary cases, and they learned what good looked like by producing imperfect first drafts and having someone more experienced mark them up.

Consider a hypothetical junior business analyst. Until recently, they learned the job by collecting data, chasing down numbers that didn't reconcile, preparing a first draft of each report, and walking a senior colleague through their findings. Tracing an inconsistency showed them how the data was produced, and defending a draft showed them which assumptions would be challenged and why. Now an AI system handles much of that preparation, and the analyst produces a useful report in a fraction of the time. The gain is real, but if they never trace a problem back to its source, it's unclear how they'll learn to judge whether an analysis holds up, notice when a pattern is unusual, or defend a recommendation under questioning.

Leaders appear to see both sides of this. In the 2026 Wharton–GBK AI adoption study, 82% of managers and more senior leaders said AI helps newer employees contribute faster, while 24% expect their departments to make fewer entry-level hires over the next two to five years. Neither finding shows that AI weakens expertise, but together they suggest that faster early contribution and long-term development need to be managed as separate goals.

Keeping junior employees away from AI would give up a real benefit, so the better option is to rebuild the learning around AI-assisted work. For the analyst, that might start with checking each AI-generated finding against the original data, paying closest attention to results that look surprising, and flagging anything they can't verify. They then explain their conclusions to an experienced colleague, and the conversation focuses on how they reached them: which numbers they trusted, what they questioned, and what evidence would change their view. As their reasoning proves reliable, they take on more complex cases with less review, and the senior colleague steps in mainly on the decisions that carry the most risk.

Progress in that model shows up in the work itself. Can the analyst spot a questionable conclusion before someone else does? Can they explain why they trust a result rather than simply stating it? Are their judgments reliable enough to need less review as the cases get harder? When those answers improve, the organization is still building the expertise its more senior roles depend on, even though the routine work that once did the teaching has changed.

Good managers have always coached this way. The difference is that routine work used to provide much of the practice on its own. As AI takes on more of it, someone has to decide where that practice now happens and who is responsible for it. KPMG found that 81% of CEOs say AI has made their organization rethink how employees are trained and developed, and the experiences that used to build expertise are a practical place to start.

Help people prepare for the roles they're moving into

Decisions about how roles, structures, and career paths should change belong to an organization's leaders, usually working with HR and workforce planning teams. Once those decisions take shape, the next challenge is making them usable for the people affected. That's different from introducing a new tool, which we cover in AI change management. Here, the goal is helping employees understand how their own responsibilities are changing and how to build the skills those responsibilities now require.

Employees need guidance written for their role: where AI can assist, where their judgment is still required, how to check AI-assisted work, and what good work looks like under the new expectations. Managers need practical support too, because they assign the review work, hold the conversations about reasoning, and decide when someone is ready for more responsibility, as in the analyst example. Learning content helps most when it's tied to those responsibilities and easy to return to after a training session ends. That's the work our Workforce AI Readiness, Playbooks & Training offering supports, turning the expectations an organization has defined into role-based playbooks, manager resources, and learning materials people can apply in their daily work.

People also need an honest picture of what comes next. If an entry-level role no longer leads naturally to the one above it, employees should understand which experiences now build toward more senior work and how their progress will be judged. The same applies outside the organization. Many employers describe career growth as part of their employer promise, and as AI changes the roles through which people have built careers, that promise needs to match what employees actually experience. Keeping it credible is part of employer brand and talent experience work.

No forecast can tell an organization exactly which roles will grow or shrink. What leaders can understand now is how the work inside each role is changing, which capabilities still matter, and where people will get the experience to develop them. Those answers give an organization a much firmer basis for the commitments it makes to its people.