Your Newest Hire Is an AI Agent. Manage It Like One.

I spent two decades building HR technology, and I now run a company where AI agents do a lot of the research, drafting and prep work. Here is the thing nobody warned me about: almost everything that goes wrong with an agent is something that goes wrong with a new employee.

Not because agents are people. They aren’t. But the failure modes rhyme, and the fixes are ones HR and people leaders already know.

They both need a proper brief

Give a new hire a vague task and you get a vague result. Give an agent a vague prompt and you get confident nonsense, faster.

The first skill to teach anyone who works with agents is writing the brief. What does done look like? What is out of bounds? What has to come back to a human before it goes anywhere? People who can delegate well to a junior colleague can delegate well to an agent. People who can’t will get the same disappointing output they always did, just at higher volume.

They both need feedback, not just approval

I have spent years telling leaders that feedback is a gift and that approval is not feedback. The same is true here.

When a person reviews an agent’s work and simply clicks “approve,” two things happen. The agent doesn’t get better, and the person slowly stops knowing whether the work is right. Approval without explanation is how expertise quietly dies.

Make people explain why an output is right, not just that it is. That habit keeps their judgment sharp and gives you a record of what good looks like.

They both need someone accountable

When a new employee makes a mistake, their manager owns it. That is the deal.

The same rule has to apply to agents. The person who assigned the work owns the result. “The agent did it” should land about as well as “the intern did it.” If nobody in your organization can say who is accountable for an agent’s mistake, you don’t have an agent problem. You have a management problem.

They both need to keep learning, and so do we

Here is the part that worries me most. As agents take over the doing, people stop practising the doing. Task expertise fades, and with it the ability to judge the agent’s work.

The fix is the same one we use for any skill that atrophies: deliberate practice. Have people do the task by hand on a regular rotation. Not because it is efficient, but because judgment is built by doing and by catching mistakes.

That is also the answer for entry-level staff. Start them as reviewers, with a senior person reviewing their reviews. Agents supply an endless stream of mistakes to catch. That is a training program if you treat it like one.

They both need to be measured on the right thing

If you want to know whether your agent training is working, don’t track what people complete. Track what they catch.

Gravitee’s 2026 research found that 81% of teams have agents past the planning stage, but only 14% ship with full security and IT approval. The agents are already deployed. The managers are not.

The takeaway

If your organization is good at onboarding, briefing, giving feedback and holding people accountable, you are already most of the way to being good at agents. If it isn’t, agents will make that visible fast.

Which is, oddly, good news for HR. The skills the agentic workplace needs are the ones people leaders have been championing all along. Now the whole company has a reason to listen.