Middle management is shrinking. Gallup finds the average number of direct reports per manager grew from 10.9 in 2024 to 12.1 in 2025, and Gartner projects that by the end of 2026, one in five organizations will use AI to flatten their structure, cutting more than half of current middle management roles.
The easy read on this is that AI is doing manager work, so companies need fewer managers. That's the wrong conclusion.
Bigger teams and thinner management layers aren't a sign that management matters less. They're a sign that the job itself has stopped being about personally overseeing every task. Something has to fill that gap: judgment, prioritization, and decisions about where AI helps and where it doesn't. That's still a human job. It's just not the same job.
The old job was built for a different kind of work
For most of the last century, "manager" meant someone who supervised people doing the work. Assign tasks, check the output, keep the team on track. The job was designed around a world where execution was the scarce resource and someone needed to make sure it happened.
That assumption doesn't hold anymore. AI has made a lot of execution fast and cheap: drafting, summarizing, analyzing, coding, scheduling. What's left over, and what's getting harder to find, is the judgment to decide what should get built, who or what should build it, and whether the result is actually good.
PwC's 2026 Global AI Jobs Barometer, which analyzed more than a billion job postings globally, found that roles built around routine execution are shrinking while roles centered on judgment and expertise are growing in both headcount and pay. Separate reporting on the shift describes it plainly: when the cost of doing falls, the value of deciding rises.
Managers didn't sign up to be judgment specialists. Most were trained to supervise people, not to orchestrate a mix of human and AI work toward a business outcome. That's the actual disruption, and it has almost nothing to do with prompt engineering.
From supervising people to orchestrating performance
The shift isn't manager to AI supervisor. It's people manager to performance orchestrator.
A performance orchestrator is still responsible for outcomes, but the day-to-day job looks different:
Deciding which parts of a project need a person and which can go to AI
Designing how work gets distributed across a team that includes both
Validating outputs, since faster work isn't automatically better work
Coaching judgment, not just checking task completion
Connecting the work in front of a team to the priorities the business actually cares about
None of this is about managers producing more. It's about managers directing the right work to the right resource, whether that resource is a person or a model, and being accountable for what comes out the other end.
The systems underneath managers haven't caught up
This shift is landing on managers who are already stretched. Gallup's 2026 State of the Global Workplace report found that manager engagement fell from 27% to 22% between 2024 and 2025, the sharpest year-over-year decline Gallup has recorded for any employee group. Managers used to be more engaged than the people they led. That gap has closed.
Part of the problem is structural. Most performance systems still run on annual or twice-a-year cycles, built for a slower version of work than the one managers are actually operating in now. When priorities shift monthly and team composition includes AI, a review written from six-month-old memory tells a manager, and the business, very little.
When judgment gets removed from the loop, the risk shows up fast
The stakes of getting this wrong are already visible. In July 2026, 26 current and former Meta employees sued the company, alleging it used a combination of internal AI systems, including performance scores, productivity metrics, and AI-usage data, to help select employees for layoffs. The lawsuit claims those inputs couldn't account for employees on protected medical or parental leave, and that the termination list wasn't built through the judgment of managers who knew the work. A federal judge has since allowed the layoffs to proceed while flagging that the underlying claims raise serious questions.
Whatever the legal outcome, the case makes a broader point: AI-generated scores and rankings aren't a substitute for a human who understands the context behind the numbers. Data without a validation layer isn't more objective. It's just less accountable.
What performance orchestration actually requires
If a manager's job is to allocate and validate work across humans and AI, then the tools supporting that job need to change too. A performance orchestrator needs:
Real-time visibility, not an end-of-cycle summary, into what's actually happening on their team
Evidence from real work, not self-reported activity, to ground coaching and calibration conversations
A human validation layer on top of any AI-generated recommendation, score, or summary
A clear line from an individual's work back to the priorities it's supposed to support
This is a big part of why Betterworks acquired Rypple, an AI-native platform built specifically to help managers act with more clarity and consistency in the flow of work, rather than relying on memory or a once-a-year form. Performance shouldn't be something a manager checks on. It should be something they can see, coach, and act on continuously. That's the argument behind Betterworks' broader case for treating performance as a business execution engine, not an HR process.
The definition of a great manager is changing
The old measure of a good manager was how well they supervised the people on their team. The new measure is how well they orchestrate the highest-performing combination of people and AI they have access to, and how good their judgment is when something needs a human call.
That's a harder skill to build than learning a new tool. It also means HR and talent leaders have a real role to play: giving managers the visibility, evidence, and support they need to make that call well, instead of asking them to do it on instinct with systems designed for a slower kind of work.
The organizations that get there first won't be the ones that cut the most management layers. They'll be the ones whose managers are actually equipped to orchestrate what's left.
See how Betterworks supports manager effectiveness →
Equip managers to orchestrate high-performing teams of people and AI—with the clarity, context, and continuous insights to make better decisions.
Book a Demo
