What happens to managers when agents do the work

An operations manager kept the same team, the same title, and half the caseload moved to an agent. Nobody shrank the team, nobody changed the title, and nobody sat down with him to decide what happens when the agent gets a case wrong that he never saw.

Carlos Andrés Ramírez · 2026-09-13

An operations manager kept the same team, the same title, and half the caseload moved to an agent. Nobody shrank the team, nobody changed the title, and nobody sat down with him to decide what happens when the agent gets a case wrong that he never saw.

Every time a company puts an agent in charge of part of a team's work, someone in the room applauds the productivity gain. Almost nobody sits down with that team's manager to redesign the job underneath it. The agent moves in, a chunk of the work disappears from the queue, and the job keeps the same title, the same reporting line and the same metrics it had when the whole team was human.

The conversation circulating right now does not help. It talks about a brand new role, the "agent manager," built from the ground up to run fleets of agents. That discipline will probably exist one day. It is no use at all to the manager who already holds the job today, running a mixed team of people and agents, and who needs to know what changes in their work on Monday, not once the new role has matured three years from now.

While that conversation stays theoretical, something very specific is already happening to the manager who already has the job, and nobody is naming it.

The symptom

What happens to a manager when an agent takes over part of their team's work?

They lose visibility without losing authority. They used to see their team's work because they assigned it themselves and checked every piece before it went out. With the agent, that window closes: the only signal that reaches them is whatever exception the agent itself decides to flag. Everything the agent resolves with confidence, even if it is wrong in a consistent pattern, stays out of sight by design.

And they still answer for the whole result. Not for the half their people still handle by hand, but for the entire process, including the part run by a system they did not configure and cannot audit case by case. When something breaks, the question of who answers for it lands on their desk before it reaches anyone else.

• They find out about an agent's mistake once it has already reached the client or the board, never before it went out.
• They still sign off on the team's report even though they no longer control how half of what it reports was produced.
• Nobody has said whether the manager, the team that configured the agent, or whoever owns the platform answers when something goes wrong.
• Their week fills up with cases the agent escalated as exceptions, and none of that load appears in any job description.
• The metric they are judged on is the same one from last year, from before the agent did any of this.
• They are asked for people leadership over a share of the work that no person executes anymore.

The problem underneath

The job did not disappear. It got emptied of what made it visible.

A manager's authority never came only from the org chart. It came from watching the work pass through their hands: assigning it, checking it, catching an error before it became an incident. That visibility was the real control mechanism, more than any title. The agent does not take the title away. It takes the mechanism away and leaves the title standing as if nothing had changed.

It is not that the manager now does less work. It is that they carry the same work with less information to do it well, plus a new source of risk nobody trained them to read: an agent that gets things wrong slowly, staying just under the threshold that would trigger a flag, never shows up in any report until the accumulated error is too big to explain in one sentence.

You cannot ask a manager to answer for a process they can no longer watch run in full.

BECOME

The transition

What needs redefining in the job, instead of inventing a new one?

There is no need to wait for the "agent manager" discipline to mature to fix this. What is needed is to write down four things that today live in habit rather than in any actual decision, for the manager who already holds the job.

Minimum visibility

What share of the cases the agent resolves the manager actually sees, even when they are not flagged as exceptions, and how often. Without that minimum, the only reality they know is whatever the agent chooses to surface.

Declared owner

Who signs off when a case handled partly by the agent, under the manager's nominal authority, goes wrong, written down before it fails rather than argued over once the incident is already on the table.

Recognised load

The time the manager spends resolving exceptions the agent escalates counts as part of their formal workload. If it gets stacked on top for free, the agent's productivity gain is being paid for out of the manager's own time.

Updated metric

The indicator used to judge the manager gets rewritten to separate what depends on their people management from what depends on the agent's behaviour. Judging it with the old metric evaluates a job that no longer exists.

None of the four requires hiring anyone or inventing a new box on the org chart. It requires whoever designed the process the agent joined to sit down with the manager already running it and answer these four questions before a second agent joins the same team.

Pick one process where an agent already handles part of the work and ask whoever runs it today who answers if it fails. If the answer is "I don't know" or takes longer than ten seconds, that is the gap to close before the next agent joins the same team.

Let's redesign the manager's job