Nearly half of workers now spend more time directing AI than doing the work itself. Pay systems were built for managing people, not agents, so that new responsibility mostly goes unrecognized.
For decades, the path to management was clear. You did the work well, you got promoted, and you started overseeing other people’s work. That came with a new title, more pay and the authority to go with it.
AI is breaking that sequence. A growing share of workers now spend their days assigning tasks to AI tools, checking their output, fixing their mistakes and keeping multistep workflows on track. That is the substance of supervision. But the workers doing it are mostly still individual contributors, and their job descriptions haven’t caught up.
The shift in the numbers
Boston Consulting Group‘s fourth annual AI at Work survey, released in June and based on responses from 11,749 workers across 14 markets, found that 47 percent now spend more time managing and directing AI than doing the work itself. BCG described the findings as evidence of “a true managerial revolution in the age of AI.”
Adoption is behind the change. Among frontline employees, meaning white-collar workers with no managerial responsibilities, 74 percent now use AI regularly, up 23 percentage points from 2025. Thirty percent of respondents said AI agents are already built into their workflows, more than double the 13 percent a year earlier.
The workers taking on this supervision say they aren’t being prepared for it. Eighty-eight percent believe they will need major upskilling in the next five years, but only 36 percent feel properly trained. Half say they lack clear governance for managing mixed teams of humans and AI. And while 67 percent say AI has improved their job satisfaction, 41 percent report that it has increased their cognitive load, a tension BCG calls the “joy paradox.”
Why the work gets harder
The heavier load follows from how AI divides tasks. When a tool handles the repetitive parts of a job, what remains for the human tends to be harder, more ambiguous and more judgment-heavy. The easy decisions get automated, and the difficult ones land on the person overseeing the tool.
That’s what management has always involved. Managers are accountable for results they didn’t personally produce. They need to know what good work looks like, catch problems before they spread and take responsibility when things go wrong. A worker running a set of AI agents is doing the same thing, often across more outputs than a first-line supervisor would oversee.
Why pay isn’t following
Compensation systems weren’t built to recognize this. Management pay has traditionally been tied to title and to the number of people a manager oversees. An AI agent doesn’t show up on an org chart and doesn’t count as a direct report. A worker who reviews the output of five agents all day looks, on paper, exactly like one who doesn’t use AI at all.
The money that is going to AI skills mostly goes to new hires. A KPMG survey found that 76 percent of leaders would pay up to 10 percent more for candidates who demonstrate strong AI skills, and 22 percent would pay 11 to 15 percent more. Meanwhile, Mercer’s 2026 compensation planning survey found that 83 percent of U.S. employers still spread salary increases evenly across the organization instead of targeting high-demand skills. Taken together, the market rewards AI capability at the point of hire, while current employees who develop the same capability on the job tend to get the same raise as everyone else.
Titles are lagging too. BCG’s research suggests companies are adopting AI faster than they are redesigning jobs around it, so formal job architecture still describes work the way it existed before these tools arrived. Until those descriptions change, supervising AI remains invisible labor, done every day but not formally recognized.
A role that may not last
There is another complication. The oversight role workers are growing into may itself be temporary. In KPMG’s surveys, the share of leaders expecting humans to primarily manage and direct AI agents was 76 percent in the third quarter of 2025. By the fourth quarter, 44 percent expected AI agents to take the lead in managing specific projects with human team members within two to three years. BCG found that 65 percent of managers and leaders believe agents will take over at least half of their own jobs within three years.
That creates an uncomfortable possibility for workers. They are being asked to take on supervisory responsibility without supervisory pay, in a role that their employers may already expect to automate. If the management layer being built around AI is only transitional, companies have even less incentive to create titles and pay bands for it.
What recognizing the work would take
The argument for changing this isn’t only about fairness. Oversight is where AI’s risks get caught. If companies want workers to review AI output carefully rather than approve it quickly, they need to treat that review as skilled work, with training, clear standards and pay that reflects the responsibility.
That would mean pricing jobs by accountability as well as headcount, counting AI oversight when evaluating roles and giving workers the upskilling most say they aren’t getting. Until then, many individual contributors will keep doing a manager’s job at an individual contributor’s pay, and the gap will only widen as more of the work moves to agents.
Sources: Business Insider, BCG press release, BCG, BCG AI at Work slideshow, Work-Self, Enterprise DNA, KPMG AI Quarterly Pulse, Christian & Timbers (Mercer data)