AI Can Delete a Manager. It Can’t Delete the Decisions They Made.

Oracle cut around 30,000 jobs in March 2026 and told analysts the savings were funding a shift from “people-intensive consulting and legacy support” toward GPU-intensive AI infrastructure. Meta set a target ratio of 50 engineers per manager on its new applied AI engineering team, roughly double the 25-to-1 ratio organisational researchers usually treat as the practical ceiling. Amazon’s Andy Jassy told his leadership team that fewer managers would “remove layers and flatten organisations” and “increase our teammates’ ability to move fast.” Microsoft executives have told reporters that recent cuts were partly designed to widen managers’ “span of control.” The story attached to all of this is straightforward: AI agents can now chase status updates, schedule syncs and translate strategy into tasks, so the layer can come out. It is a tidy story, and the industry has cut hundreds of thousands of jobs on the strength of it this year.

The flattening wave already happened once

It also is not a new story. Gallup has measured how many people report to the average American manager every few years since 2013, well before generative AI existed, and the number has risen from 8.2 to 12.1, a 48% increase. Run the same calculation on the last academic study of this kind: Raghuram Rajan and Julie Wulf’s survey of large US firms found the number of managers reporting directly to a CEO rose from an average of four to seven between 1986 and 1999, a 75% increase in 13 years, while hierarchy levels between division heads and the CEO fell by 25%. Annualised, that computer-driven delayering ran at roughly 4.4% a year. Gallup’s 2013–2025 figures average out closer to 3.3% a year. On this evidence, AI has not yet produced a faster hollowing-out of management than enterprise computing did three decades ago. What it has produced is a sharp jump in the latest measurement: spans of control rose from 10.9 to 12.1 between 2024 and 2025 alone, an 11% move in a single year, against a decade that had been running at under a third of that pace.

What AI actually removes

The recent acceleration has a specific cause, not a general one. Gartner’s own analysis of the trend describes AI agents automating scheduling, reporting and performance monitoring, the administrative core of a large share of management work. Andy Williamson, chief executive of ONLC Training Centers, puts a figure on what is disappearing: managers have historically spent close to a third of their week in meetings built for synchronising other people’s work, precisely the task software can now do instead. That accounts for the half of the job that moved information between people. It says nothing about the half that decided what the information meant.

What doesn’t disappear

Meta’s 50-to-1 ratio is a useful stress test precisely because it is the most extreme version running today. André Spicer, professor of organisational behaviour at Bayes Business School, has already called the likely result: junior staff get overlooked, remaining managers burn out, and attention gets monopolised by “the loudest people or the problem cases,” because someone still has to decide whose priority wins when two projects compete for the same engineer this week. If that stops being a manager’s explicit job, it becomes an informal one. That is close to what researchers found when they revisited the 1990s delayering wave: firms had fewer formal layers on the chart, but the coordination those layers performed did not disappear, it moved somewhere less visible and less accountable. Deleting a box on an org chart is not the same as deleting the decisions that used to get made inside it.

Gartner expects one in five organisations to use AI to flatten their structure and remove over half of current middle management roles by the end of 2026. A separate Gartner forecast, issued a year later, expects the resulting atrophy of critical thinking to push half of global organisations toward “AI-free” skills assessments in hiring, because it is no longer obvious where judgement gets learned if nobody is doing the job that used to teach it. Gartner does not present these as connected, but they sit inside the same body of research: one prediction the industry is racing to fulfil, the other a description of the gap that racing leaves behind.

The verdict

None of this makes flattening a mistake. Gallup’s own data shows the narrower spans that already existed before generative AI, five or six direct reports, work fine when managers keep giving regular feedback; team size alone does not move engagement, the absence of feedback does. AI genuinely can take reporting and scheduling off a manager’s plate, freeing time for the judgement calls that used to get squeezed out by admin. GitLab proved that years before it was fashionable: it replaced manager-mediated coordination on purpose, not by deletion, routing decisions through a public handbook and documented issues so judgement stays visible instead of living in one person’s head. The mistake is treating the headcount cut as the operating model, rather than the precondition for building one. Oracle, Meta and Microsoft have all proven they can remove the layer. None has yet had to prove, in public, that they replaced what it decided. That is the part of this restructuring nobody has put a number on, and it is the only part that decides whether the savings are real or just deferred.

Sources

• Gallup: “Span of Control: What’s the Optimal Team Size for Managers?”
• NBER Digest: “The Flattening of Corporate Management” (Rajan & Wulf)
• NBER Working Paper 9633: Rajan & Wulf, “The Flattening Firm: Evidence from Panel Data on the Changing Nature of Corporate Hierarchies”
• Harvard Business School Working Paper: “The Flattened Firm: Not as Advertised”
• CNBC: “Oracle sheds 21,000 roles over the past year amid wave of AI layoffs from tech giants”
• Forbes: “Oracle’s Massive 30,000 Layoff As AI Spending Surges”
• Fortune: “Meta’s AI team has 50-to-1 flat management structure”
• CIO Dive: “5 Gartner predictions about IT’s future” (AI-free skills assessments)
• Inc.: “Gartner Predicts AI Will Eliminate 50 Percent of These Management Roles by 2026”
• The GitLab Handbook
• Allwork.Space: “Managers Are Being Stretched Thin As Average Team Size Jumps To 12 Workers”