The Rise Of The Agent Manager In The Modern Enterprise
Lenin Gali is a former CIO and currently the Chief Digital Officer at Atomicwork.
Earlier in 2026, McKinsey’s Bob Sternfels told the audience at CES that his firm now counts 25,000 AI agents alongside its 40,000 human employees. A year and a half ago, they had just a few thousand. His goal was that every employee would be working alongside at least one agent by year-end.
That number made me pause. When more than half of your workforce is nonhuman and growing, somebody has to write their role specs, set their access boundaries, watch their output and know when to widen their scope or rein them in.
That job is starting to get a name. People are calling it the “agent manager" or the “supervisor.”
Mercer’s Global Talent Trends 2026 report found that 82% of C-suite leaders believe HR’s future lies in managing human talent and digital agents side by side. The agents are showing up, but how ready are we to manage them?
I’ve watched org charts get rewritten before: e-commerce and social media, the cloud and DevOps era, followed by big data and machine learning eras. The people who succeeded in these new roles were the ones who figured out a new way of working. This AI era is the next round.
The traditional individual contributor (IC), someone measured on personal output, is being reshaped. Vercel’s CEO, Guillermo Rauch, says in a podcast that his mental model now is that the IC is the agent, and the human becomes the manager. Everyone, in his framing, is becoming a “mini CEO.”
HBR profiled Salesforce’s use of autonomous agents and highlighted how teams are beginning to manage AI across support, sales and marketing as a coordinated workforce rather than a collection of tools. In one example, support agents handled roughly 74% of cases autonomously, while sales teams scaled outreach from 12 to 15 prospects a day to more than 350 meetings per week, generating a reported $60 million annualized pipeline in four months.
Those numbers came from better management and not simply deploying AI agents as co-workers without the necessary reins. As ICs begin to manage agents, they also need the right infrastructure to do it well.
Most of the conversations I’ve had and signals I've read on this topic say, “Domain expertise matters more than coding skills.” That’s correct, but it only gets you partway. The agent manager’s actual job starts with something that sounds simple and turns out not to be: defining a boundary for each AI co-worker, what it does and the role it plays.
An AI co-worker has a role designation, a reporting line, authorization scope, behavioral guidelines, goals, guardrails, specific skills, authorized tools and a budget tracked against value. Building that profile is just like writing a job description before hiring employees.
1. Role Design And Skill Definition: What can this AI co-worker do? What tools does it get? Where do its boundaries sit? You wouldn’t bring a human onto a team without a clear role spec, and the same is true here.
2. Task Orchestration Between Agents And Humans: Not every piece of work should go to an AI co-worker, and not everything needs a human. The agent manager makes that call based on context, capability and risk. This is where deep domain knowledge earns its keep.
3. Governance: What data can each co-worker access? Which actions need human approval? These are the same questions IT has always asked about employee access and change management.
4. Performance Management: AI co-workers don’t stay calibrated on their own. Output quality drifts and satisfaction scores shift. The agent manager tracks these signals and tunes instructions, skills and tools in response as continuous reviews rather than a one-and-done configuration.
5. Life Cycle Management: Like any employee, AI co-workers get deployed, take on new responsibilities, gain or lose system access, and eventually get retired. The agent manager owns that arc, from onboarding through scope expansion to knowing when to pull one offline.
All five pillars map to how organizations already manage people. Agent managers are expected to apply these principles to how they run the AI co-workers reporting to them.
Let’s break down those five pillars to what it realistically means to get started as an agent manager. If you’re stepping into this role or building it for someone on your team, think about how you’d accomplish the following:
• Naming The Manager: Every AI co-worker or a group of them needs a human who’s accountable for its output, its behavior and its boundaries. Give the role a title, a reporting line and a dashboard. Make it a real job.
• Auditing Your Agent Inventory: How many AI agents are running in your environment right now? Who deployed them? What data can they touch? Most organizations I talk to can’t answer these questions cleanly, but it's important to start here.
• Thinking In Roles, Not Tasks: The instinct is often to list tasks you want to automate. Fight it. Ask which roles could be staffed by an AI co-worker as an augment instead of a replacement. This might be an IT support specialist or an onboarding coordinator. Roles force you to think about governance: scope, access, reporting lines and success metrics.
• Picking One Role And Defining Evals Before You Deploy: Go with the highest-volume role with clear success criteria. IT L1 support is a natural fit. Set the OKRs like you would for a human hire, with measurable performance indicators such as resolution rate, time to resolve, satisfaction and cost per resolution.
• Building Governance Before You Scale: Think in terms of directory, access governance and change management for every AI co-worker before running an entire workforce.
It's likely we'll soon have the “hybrid workforce” running the modern enterprise, and the role of a human “agent manager” will become critical, whether managing one co-worker or a group of them as a team. The magic happens when this hybrid workforce gets to work.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?