Why Your Next Top Performer Might Not Be Human

Who manages a worker that isn’t a person? It’s not a hypothetical anymore. Agentic AI can complete multistep tasks, make decisions and act on the company’s behalf with little human involvement – and it lands in the gap between two jobs. Legal has to answer for what it does; HR has to figure out how it’s managed. Neither role was built for a worker like this.

The shift is worth being precise about. For two years, AI meant tools that helped people work faster.

The human stayed in the driver’s seat; the AI rode along. Agentic systems break that arrangement. They don’t wait for the next prompt. They act inside defined boundaries and commit the company to the outcome. 

Google and Microsoft are already moving agents into live workflows rather than parking them at the prompt window.

That is not a faster tool. That is a new kind of worker, and someone has to manage it.

The legal line most companies haven’t drawn

There’s a distinction here with significant legal and human implications. When a person uses an AI assistant, the person is still the actor. A bad output is a typing mistake or a tool failure, in the same category as a botched spreadsheet formula.

An agent is different. Once a system has authority to act and decide on your behalf – to negotiate a contract, approve a claim, issue a refund – the company is generally bound by what it does, in the way it’s bound by an employee.

The law will not have much sympathy if the cause was a “glitch.” You don’t get to grant a system agency and then disown its decisions. That turns the manager’s job from a productivity question into a legal-exposure question.

What management actually has to answer when AI joins the team

Managers know how to lead people: coach, hold accountable, escalate. Agents don’t fit that playbook, because they can act before anyone weighs in. The open questions are practical, not theoretical:

• Should every agent have a named human manager?
• What level of autonomy is appropriate for a given task?
• At what point does human review become mandatory?
• What is the appropriate “discipline” for a wayward agent or its human manager?

Those read like technology questions, but they really are management questions.

In an agentic world, governance has to live where the work happens

The biggest mistake is treating agentic AI like any other software rollout – approve it, document the policy, move on.

Policy documents don’t supervise anything. Once a system is acting inside a workflow, governance has to reach into how it behaves minute to minute.

Oliver Patel, who runs enterprise AI governance at AstraZeneca, makes this point well: Governance cannot simply live in policy documents; it has to be built into how systems function.

It’s the logic behind the “digital leash” framework. A dog owner is responsible for the dog in the park, because the owner understands that a dog’s behaviour can be unpredictable and the owner has chosen to enjoy the benefits of having a four-legged best friend.

The party that profits from an agent owns what the agent does. That means real boundaries and real escalation paths at the point where decisions get executed, not in a paragraph in a handbook.

The performance review problem

Here’s where worlds collide. If a manager’s results increasingly depend on how well they deploy agents, then agent performance becomes part of how we evaluate the manager. We’ll be reviewing people on things that didn’t exist on a scorecard two years ago:

• Quality of oversight, including catching hallucinations before shipment
• Escalation judgment—knowing when to pull the leash
• Governance discipline and the instinct to spot risk early

Leading the agent-empowered team

Managers need guidance soon, and it starts with three questions for every workflow: Is the AI assisting or deciding? What is off limits without a human in the loop? And how do we keep people from automation bias, the quiet tendency to trust an AI output because checking it is more work than accepting it?

What does good management look like when part of your team isn’t human? That may be the defining leadership question of the next decade. Agentic AI joining the workforce isn’t a forecast anymore. It’s happening. The only open question is whether leaders are ready to manage it responsibly.

Rob Feldman, Chief Legal Officer, EDB 

At EDB, a leading sovereign data and AI company, Rob Feldman is responsible for the worldwide legal and compliance functions, including its Responsible AI initiatives.

An experienced executive and lawyer, he builds high-performing legal teams to support growing technology companies in dynamic business and regulatory environments.

Rob spent more than a decade in private practice as a technology company litigator, focused in securities fraud defence, intellectual property disputes, and government and internal investigations.

He also serves on the U.N. Global Compact Legal Council, providing strategic guidance on global regulatory environments to help businesses drive transformative, long-term impact.