Management layers, AI and the strain on the managers who remain

In 2026 Korn Ferry surveyed over 16,000 professionals in eleven countries, from frontline staff to CEOs. Two in five organisations cut management roles in the past year. Of the managers who remained, 55 per cent say they are exhausted, and 61 per cent of employees now do the work of more than one role. In the same survey, 79 per cent of CEOs report that AI makes work more efficient; among frontline staff the figure is 51 per cent, and half of those suffering from AI fatigue say the technology has actually increased their workload.

What is new is the reason given. In June 2026 Fortune described restructurings at Cloudflare, Meta and GitLab, among others, against a backdrop of AI agents taking over the coordinating work of middle managers; Cloudflare cut around twenty per cent of its staff in a year of record revenue. One CEO quoted estimates that a middle manager spends a third of the week in meetings keeping people aligned, which is precisely the work software is expected to take over. The World Economic Forum hears the same from more than twenty large technology companies: the pressure is falling on mid-level roles sooner than on entry-level jobs, although the Forum still calls the evidence for this anecdotal.

McKinsey adds a caveat. In its State of Organizations 2026, a survey of more than ten thousand senior executives in fifteen countries, almost everyone is experimenting with AI and the majority do not yet see any effect on results. According to the report, the biggest productivity gains come from simplifying processes across the whole organisation, and fewer than a quarter of organisations sustain an improvement in performance. The report deliberately shifts the focus to how the work flows rather than to the org chart.

In October 2026 Ethan Mollick, a professor at Wharton, supplied the line of thinking. Part of management exists, in Mollick’s view, to compensate for human limitations: conflicting goals, information that gets stuck, the cost of coordination. Agents largely lack those limitations, although they bring risks of their own. Mollick does expect cheaper organising to generate more work. Organisations will try more things, and people will still be needed to decide what the agents are put to work on.

So the argument is running ahead of the evidence. The layers are gone, the agents are present in only part of the organisation, and the coordination work has meanwhile been spread across employees who already hold more than one role. See also the briefing Span of control and how much hands-on work managers still do, which does the sums on what lands on the remaining manager once the cuts are made.

In European grocery retail, the same reasoning has been put on paper. In April 2026 McKinsey and EuroCommerce calculated that AI agents could take over 40 to 50 per cent of routine tasks at a supermarket head office. The two organisations expect team leaders in some cases to manage people and agents together, and category teams to look after more categories. Yet in their survey of over 35 executives at European chains, 8 per cent have established agentic AI applications and not a single chain uses agents widely across departments. Seventy per cent see no measurable effect on results yet or think it too early to say.

How solid is this? Four independent organisations measure or describe the same movement within a single year, and Korn Ferry has recorded a figure of around forty per cent two years running. The series is short and the link with AI comes mainly from American technology companies, so the picture is not yet settled.