Enterprises are onboarding AI agents faster than they can govern them. The shortage isn't compute. It's a manager with the agent's name on their org chart.
Walk into most enterprises today and you'll find a workforce that no longer fits on a single org chart. Humans are on one chart. The AI agents are on a spreadsheet somewhere, owned loosely by IT, sponsored vaguely by a function, and supervised by no one in particular. That gap is not cosmetic. It is the largest unmanaged labor pool your company has ever run.
McKinsey now operates roughly 20,000 AI agents alongside about 40,000 human employees, and it has said openly that it expects parity, 40,000 agents to 40,000 people, before the end of this year. By 2026, an estimated 80% of enterprise applications will ship with embedded agents marketed as digital coworkers. In blended teams the working ratio is already one human to three, five, sometimes eight agents. We have spent two years asking whether AI will replace managers. The sharper question is who manages the AI, and in most companies the honest answer is nobody.
This is how you end up with thousands of digital workers that nobody set goals for, nobody reviews, and nobody can switch off when they drift.
A workforce with no line of reporting
Consider what it normally takes to add a single human to a team. There is a hiring manager. There is a job description. There are objectives, a probation period, a review cycle, an escalation path, and eventually an offboarding process when the role ends or the fit fails. None of that is bureaucratic ornament. It is the connective tissue that keeps a person accountable to the business.
Now look at how an agent gets added. A vendor enables a feature. A team flips it on. The agent starts touching production systems, customer records, financial workflows. And then the structure that governs a human employee simply isn't there. No goal was set. No one owns its output. No one is accountable when it acts wrongly at scale, which an agent can do in seconds and across thousands of transactions before a human notices.
The technology works. That was never really in doubt. What's missing is the management layer, and that absence is the actual exposure.
The market sees the gap before the org chart does
The labor market is, as usual, ahead of the policy. Roughly 28% of managers are now considering hiring dedicated AI-workforce managers to run hybrid human-plus-agent teams. Around 32% plan to bring on AI-agent specialists within the next 12 to 18 months. And 82% of C-suite leaders say they believe HR's future is managing human talent and digital agents side by side.
Read those numbers together and a pattern appears. Senior leaders broadly agree this is coming. A meaningful minority are starting to staff for it. But the gap between the 82% who see the future and the 28% who are acting on it is precisely the window in which unsupervised agents accumulate.
An agent without an owner is not an efficiency. It is a liability that hasn't been invoiced yet.
The management functions that are emerging have names now: task orchestration, agent governance, performance optimization, cross-system coordination. Strip the jargon and they describe an ordinary job. Someone decides what the agent should do, checks whether it did it well, intervenes when it doesn't, and decommissions it when it's no longer needed. The work is familiar. We just haven't assigned it.
Figure 1 · The management functions agents skip
Management function | For a human report | For an AI agent (today) |
|---|---|---|
Goal-setting | Objectives set, agreed, revisited each cycle | Usually none; behavior defined by config |
Performance review | Periodic, documented, tied to feedback | Rarely formalized; drift goes unmeasured |
Accountability | Named owner answers for outcomes | Diffuse; IT, function, and vendor each defer |
Offboarding | Structured exit, access revoked, role closed | Often left running long after its purpose ends |
How to read it: Every governance step a company applies to a human employee tends to vanish for an agent. The right-hand column is the ungoverned surface area scaling inside most enterprises right now.
Why the vacuum is the risk, not the model
It is tempting to frame agent risk as a technology problem, something to be solved with a better model, a tighter guardrail, a new evaluation suite. Those help. They do not address the structural issue. A perfectly capable agent with no owner is still an orphan process running against your systems with real permissions.
The failure mode here is mundane and therefore easy to miss. No single agent causes a crisis on day one. Instead, hundreds of them accumulate quietly, each doing something useful, none of them reviewed, until the day one of them does something expensive and the post-mortem reveals that no human was assigned to notice. The cost of governance always looks higher than the cost of doing nothing, right up until it doesn't.
What this means for leaders
Put a name next to every agent. The single most effective control is also the least technical: each agent gets a human owner who is accountable for its goals, its behavior, and its eventual shutdown. If you cannot name that person, you do not have an agent, you have an unsupervised process, and you should treat it as a finding.
Extend your HR operating model, don't bolt on a new one. The functions agents need are the ones you already run for people: objectives, review, escalation, offboarding. Adapt those cycles to include digital workers rather than inventing a parallel governance regime that no one maintains. The 82% of leaders who see this future should be funding it inside HR and operations now, not chartering a committee.
Hire the manager before you need the headline. The 28% considering AI-workforce managers are not early. They are on time. The cost of one such role is trivial against the cost of a fleet of agents discovered to be acting without oversight. Staff the supervision layer while it's a planning decision, not after it becomes an incident review.
The companies that win the agent era won't be the ones with the most agents. They'll be the ones that knew, on any given day, exactly who was responsible for each one. Capability was never the scarce resource. Accountability is.
A BusinessInfomatics original. Drawn from 2026 reporting on digital-coworker adoption, including McKinsey's published agent figures, Okta workforce data, and contemporary management surveys.
