A conversation we keep having with CEOs goes something like this. They ask, “Is AI going to cost us headcount?” What they are often hoping for is a clean answer — yes or no, this percentage, over this many years. The honest answer is messier: AI reshapes roles, but it reshapes them unevenly. Some functions compress dramatically. Some functions grow. Some stay almost unchanged. The enterprise-wide “AI displaces N percent of jobs” number is the wrong frame for almost every practical decision a leader has to make.

The more useful question: where is the change largest in our specific business, where is it slower, and what do we need to be investing in now?

This article lays out where we see the impact concentrating across enterprises in 2026, where it is muted, and what the organizations that handle this transition well are doing that the others are not.

Where AI is compressing roles

Four categories of work have seen the largest AI-driven productivity gains, and in each case the shape of the change is similar: one senior person plus AI tooling now handles what used to take a team of junior contributors.

Junior analysis work

The first-year analyst at a consulting firm, the associate at an investment bank, the market-research assistant — these roles were defined by producing the research that senior professionals used to make decisions. Slide compilations, market sizings, competitive scans, synthesis of public data into a coherent briefing. AI does the first draft of this work in minutes, leaving the senior professional to judge, extend, and present.

The compression is real and already visible in hiring patterns. Firms that historically hired large junior classes have quietly narrowed them. The work is not gone; the people who did it are fewer.

Boilerplate coding

Code that follows a pattern — CRUD endpoints, UI components from a design system, glue code between well-documented APIs, unit tests for standard logic — is now produced in a fraction of the time it used to take. The work that remains for the engineer is design, architecture, debugging, and the judgment calls that AI still struggles with.

Senior engineers have become significantly more productive. Junior engineers have become harder to justify, because the gap between what a junior produces and what a senior with AI produces has widened. The implication: the traditional training path from junior to senior is under pressure, and the organizations that handle this well are the ones investing in new ways to develop mid-level engineers who did not come up through the old junior ladder.

Level-one customer support

The first-line support agent who answered password resets, account-status questions, and standard troubleshooting from a knowledge base — this role has compressed sharply. Modern support platforms combine retrieval-augmented AI with human handoff for complex cases, and the ratio of AI resolution to human escalation has shifted dramatically.

The remaining human support roles are different. They are higher-tier specialists who handle the cases AI cannot, plus a smaller number of workflow designers who build and maintain the AI-plus-human systems. Total headcount in customer support has dropped; the distribution of seniority has shifted upward.

Structured writing

Marketing copy variants, standard contract drafts, routine communications, SEO-optimized content, email templates, product descriptions — structured writing at scale is a task AI is genuinely good at. The writer’s role has shifted from production to editing, voicing, and judgment about what to publish.

Teams that used to have five junior copywriters now often have one or two senior ones. Agencies that sold volume content production have had to pivot or shrink.

Where AI augments but does not replace

Four other categories show a different pattern. AI makes the human more productive, but does not remove the human from the workflow.

Senior judgment roles. The partner at the law firm, the senior engineering architect, the chief medical officer, the executive decision-maker. AI accelerates their preparation and expands their research reach, but the decision remains theirs. Customers, boards, and regulators still expect a human to be accountable for the call. This is unlikely to change on a time horizon relevant to this decade’s workforce planning.

Client trust work. Complex sales cycles, strategic account management, executive relationships — these are fundamentally human-to-human activities. AI helps the account manager prepare for a meeting; it does not replace the meeting. The same dynamic applies to therapy, healthcare, education, and any function where the relationship itself is part of the value.

Novel problem-solving. Research, R&D, novel product development, the kind of engineering that solves problems that have not been solved before. AI is a powerful tool in these domains but not a replacement. The humans working here are often more productive, not fewer.

Physical and situational work. Skilled trades, field service, healthcare delivery, logistics, manufacturing. Robotics is making some progress, but the combination of fine motor control, situational judgment, and customer presence is still firmly human work. AI-driven planning and scheduling optimizes these workers’ time; it does not displace them.

What stays human

A smaller category remains almost untouched: roles where the entire point is that a person is accountable. Legal liability, moral responsibility, ambiguous judgment calls, relationship-building at the executive level, decisions that affect people’s lives.

This category is smaller than most organizations assume. It is also resilient in a way that other categories are not. A CEO’s calendar is not about to be run by AI, but a CEO’s briefing materials certainly are.

What leaders should be building now

The organizations handling this transition well share four investments.

Retraining at scale

Instead of treating AI-displaced roles as a headcount problem to manage down, they treat them as a retraining opportunity. A former L1 support agent becomes a workflow designer for the AI system that replaced L1. A former junior analyst is retrained as a more senior analyst, faster than the old ladder permitted. The organization preserves institutional knowledge and avoids the reputational and cultural cost of large layoffs.

This only works if it is budgeted seriously. Retraining is an investment, not a rounding error. The companies treating it as a check-the-box exercise are getting check-the-box results.

Role redesign, not just role cutting

When a role is compressed by AI, the temptation is to eliminate it. The more productive approach is to redesign it. What can the freed-up human time do, now that the repetitive portion is handled? Often the answer is higher-leverage work that was always there but never had time allocated — deeper customer conversations, more experimental projects, quality investments, strategic planning.

Organizations that ask “what is now possible?” after AI adoption find opportunities that were invisible before. Organizations that ask “how many people can we cut?” extract the short-term efficiency and miss the larger value.

Human-in-the-loop processes

For most high-stakes workflows, the right shape is not AI alone or human alone but AI plus human. The human handles judgment, edge cases, and accountability; AI handles volume, speed, and consistency. Designing these combined workflows is itself a new discipline, and the companies investing in that discipline — explicitly, with dedicated roles and tools — are getting compound returns.

A new path for early-career talent

The traditional entry point to many professions was doing the grunt work that AI now does. That means the old training ladder is broken. Organizations serious about talent pipelines are building new ones: apprenticeship programs, rotational roles, structured exposure to senior work earlier in careers.

This is not charity. It is talent strategy. A firm that figures out how to grow mid-career professionals in an AI-accelerated environment will have a durable advantage over firms that continue to hire the same way and discover in three years that they have a senior cohort aging out with nobody behind them.

The organizations winning this transition

A pattern is emerging across industries among the organizations handling workforce transformation well. They share a few characteristics.

Leadership treats AI as a tool for reshaping work, not primarily as a cost-reduction lever. They communicate this explicitly, so employees are not quietly optimizing to protect their own jobs against an unstated threat.

HR and technology functions collaborate on role design, not just tool procurement. The workflow redesign and the software deployment are the same project, not two separate ones.

The metric that matters is output per employee plus employee capability growth, not just output per employee. Short-term efficiency is easy; long-term capability is harder and more valuable.

Most importantly, leaders are honest with their workforce about what is happening. Employees can tell when a strategy is being obscured. The organizations that hide the transition behind euphemism are losing trust; the organizations that name it, explain the logic, and invest in the people in front of them are building the cultural substrate that transitions of this size require.

The AI workforce transition is not a future event. It is happening now, unevenly, across every enterprise. The question for leaders is not whether to prepare but what to prepare. Investing in retraining, role redesign, human-in-the-loop processes, and new talent pipelines is not a hedge. It is how the organizations that emerge stronger are getting there.