Two things happened this week that look unrelated. They aren't.
ServiceNow laid off hundreds of employees — solution consultants, sales staff, product marketers, L&D. Its explanation to the press: "real AI efficiencies inside our own business." The company that sells AI automation just automated a chunk of its own workforce.
Nearly three weeks earlier, KPMG and Microsoft announced they would roll out Microsoft's Agent 365 platform across KPMG's entire global workforce. That's 276,000 professionals across 138 countries. Not a pilot. Not a limited rollout. The whole firm.
Agent 365 isn't a chatbot. It's a control plane — a governance and security layer for observing, managing, and securing fleets of AI agents running across multiple systems and workflows simultaneously.
Put those two events together and you have the clearest possible picture of where enterprise AI is in June 2026: the agents are in production, the headcount reductions are following, and most organizations haven't answered the one question that determines whether this goes well or badly.
Who manages the fleet?
The Deployment Is Ahead of the Governance
Gartner's latest forecast says 40% of enterprise applications will embed task-specific AI agents by end of 2026, up from less than 5% in 2025. That's not a prediction about the future — that's a description of what's already being scoped in boardrooms right now.
The agents being deployed aren't doing trivial work. They're handling customer escalations. Running workflow automation across ERP systems. Doing intake and triage in recruiting. Generating code, drafting contracts, summarizing financials. These are consequential actions taken at machine speed, at machine volume.
And here's what most organizations haven't accounted for: a single AI agent making a mistake is an incident. A fleet of agents making the same mistake is a crisis.
The ServiceNow layoffs are instructive here. The company didn't just cut headcount. It restructured toward AI-focused skills. That means the roles that survived aren't the ones that execute workflow — they're the ones that define, monitor, and correct it. The people who can tell an agent what it got wrong, and why, and how to not get it wrong again.
That's a different job description than anything your HR system has on file.
What Fleet Management Actually Requires
KPMG's rollout of Agent 365 is worth studying because they're not just deploying agents — they're deploying a governance layer on top of the agents. The explicit capability: manage, monitor, audit, and secure AI agents across the organization from one place.
That governance layer needs humans behind it. Start with scope. Someone has to decide what each agent can do, what it can't touch, and what requires a human in the loop before an action gets taken. Those aren't technical questions. They're judgment questions, and the wrong answer doesn't surface in logs. It surfaces in a client relationship, a compliance audit, or a press story.
Then there's monitoring at scale. An agent running 10,000 actions a day produces a signal-to-noise problem no human can read without tooling. But the tooling doesn't know what to flag without human-defined thresholds. Someone has to set those thresholds, own the alerts, and close the loop when behavior drifts.
Debugging is its own discipline, and it's nothing like debugging code. An agent failure often looks like a correct action taken in the wrong context. Diagnosing that means understanding both the system and the judgment the agent was supposed to exercise. That intersection of systems thinking and domain expertise is rare. It's also expensive.
Then the part most folks miss entirely: redesigning the workflow once the agent changes it. Deploy an agent that handles first-level customer support and the second-level escalation queue changes. The hiring profile changes. The training program changes. The SLAs change. Someone has to map that dependency chain before it becomes a problem.
The Staffing Gap Nobody's Naming Yet
I talk to CTOs and VPs of Engineering every week. The conversation used to be about how to hire engineers who know how to use AI tools. That conversation has moved.
The new conversation is about something harder: how do you staff a team that governs AI agents?
It's not a pure engineering problem. The best agent governance people I've seen in the wild are former technical program managers with domain expertise — people who can read a workflow, understand what the agent is supposed to be optimizing for, and notice when the outputs don't match the intent. They're not writing the agents. They're the ones who can tell when the agents are wrong in ways the metrics don't capture.
Even KPMG, the firm rolling agents out across its entire workforce, frames the human side as a new skill set rather than a technical one:
"Managing AI isn't just about technology—it's about enhancing collaboration between humans and machines to drive strategic goals. New skills, abilities, and knowledge are required to work with and manage agents."
— John Doel, Partner, AI Workforce Transformation, KPMG US, "Agents of change: New organizational roles in the age of AI"
Those people don't apply to your job postings because your job postings don't know they exist yet.
Meanwhile, Gartner is forecasting more than $200 billion in AI agent software spend in 2026 — more than double last year. That money is going into building the fleet. Very little of it is going into building the team that manages it.
ServiceNow learned this internally. The people they kept aren't the ones who were doing the work the agents now do. They're the ones who know how to make sure the agents do it right.
What to Do Monday Morning
If you're a CTO, VP of Engineering, or head of TA, here's where to start. Audit your current agent deployments — not the tools, the workflows. What decisions are agents making? What actions are they taking? What's the human override path when they're wrong?
Then map the dependency chain. For every agent workflow you've deployed, list the downstream human processes that changed. If you can't name them, you have an unmanaged dependency.
Then write the job description that doesn't exist yet. What does an AI agent operations role look like at your company? What domain expertise does it require? What's the ceiling on that career path? If you can't answer those, you can't hire for it.
And don't wait for the incident. The enterprises ahead of this — and there aren't many — started thinking about governance before they had a fleet problem. The ones behind are about to find out what happens when a fleet problem finds them.
The agents are already deployed. The question is whether you're managing them or hoping they manage themselves.
VC5 Consulting specializes in placing technical leaders and engineering talent for companies navigating AI-driven transformation. If you're building the team to govern your agent fleet — or figuring out what that team even looks like — let's talk.