The Agent Manager: The Recruiting Job Whose Direct Reports Are Not Human

Somewhere in your hiring process there is probably a tool that writes outreach, ranks applicants or books interviews, and nobody's position description mentions it. That gap is what the agent manager recruiting role exists to close. This article defines the role as an organisational position: where it sits, what it is accountable for, the one metric it should be judged on, and how to pick the first agent to put a manager on. It is written for a 200-plus employee Australian employer, not a research lab.

To be clear about scope, this is about the job rather than the craft. How to write a good sourcing brief or prompt an agent well is practitioner work and sits elsewhere. What follows is what you would put in front of an executive to get the role funded.

Why the role appears now

Three things have to be true at once for a new role to be justified, and in most mid-sized Australian employers they now are.

First, the tools make consequential decisions. Ranking an applicant is not the same as formatting a document. Second, the failure modes are invisible by default: an agent that quietly screens out a whole category of candidate produces no error message, just a worse shortlist that looks normal. Third, accountability currently falls between chairs. The recruiter did not write the logic, the vendor did not make the hiring decision, and IT does not own hiring outcomes. Work that no single person owns does not get calibrated, and uncalibrated systems drift.

What the role is, defined against its neighbours

The fastest way to define the position is by contrast with roles you already have.

• Recruitment operations, or RecOps - Owns: process, systems of record, reporting; Does not own: whether a specific agent's judgement is any good
• AI or platform operations - Owns: infrastructure, access, cost, uptime; Does not own: hiring quality or candidate experience
• Agent manager - Owns: the behaviour and output quality of hiring agents, end to end; Does not own: the hiring decision itself, which stays with the hiring manager

The distinction that matters is the last one. The agent manager is accountable for the quality of what the agents produce, not for who gets hired. Blur that and you have either an unaccountable technologist or a recruiter with a new title.

The four functions of the job

Calibration

Agents are set against a target, and targets drift as roles change. Calibration is the recurring work of checking that what an agent treats as a strong candidate still matches what the hiring manager means, then correcting it. The practical form is a scheduled sample review: pull a set of recent agent judgements, have the hiring manager rate them blind, and measure the disagreement.

Evaluation

Calibration asks whether an agent matches intent. Evaluation asks whether it is getting better or worse over time, against a fixed test set that does not change when the tool does. Without a frozen test set you cannot tell a vendor model update from a change in your applicant pool. Building the test set is the unglamorous half of this role and the part most often skipped.

Orchestration

Most hiring processes end up with several agents touching one candidate journey. Orchestration is deciding what runs where, what hands off to a human, and what stops the line when something looks wrong. This is design work, and our guide to scalable hiring process design covers the underlying principles the agent manager applies.

Assurance and compliance liaison

The agent manager is not the lawyer, but they are the person who can answer what a system did on a given application and when. That makes them the natural owner of evidence for the automated decision transparency obligations that commenced in December 2026, working to an AI hiring compliance framework rather than inventing one per audit. Independent bias review stays independent: the practice of auditing recruitment AI for bias should not be graded by the person whose agents are being audited.

One hard outcome metric

A role with five metrics has none. The agent manager should be judged on hiring manager acceptance of agent-produced shortlists: of the candidates an agent advanced, what share did the hiring manager agree to interview.

It works as a single metric because it is hard to game in isolation. Push it up by advancing only the obvious candidates and pipeline volume collapses visibly. Let quality slip and the number falls. It also speaks the language of the person whose budget funds the role, which is the practical test of a metric at this level.

Two guard rails sit beside it rather than competing with it: time from application to first decision, so acceptance is not bought with delay, and the share of candidates taking an alternative path when they decline an automated step. Report the guard rails, be measured on the metric.

Reporting line

The role reports to the head of talent acquisition, or to the HR director where talent acquisition does not have its own head. Reasoning: the accountability is for hiring quality, so it belongs in the function that owns hiring quality. A dotted line to IT or data covers tooling and access.

The common mistake is placing it under IT because the work looks technical. That produces a role measured on whether the tools run rather than whether the shortlists are any good, which is the failure this position was created to prevent.

A job description outline you can adapt

The following is a skeleton, not a finished advert. Our job descriptions library has the formatting conventions to drop it into.

• Purpose - Own the output quality of the automated systems that source, screen and schedule, so hiring managers can trust what reaches them
• Key accountabilities - Calibration cycles; a maintained evaluation set; orchestration design; decision records for assurance; vendor performance review
• Measured on - Hiring manager acceptance of agent-produced shortlists, with time-to-first-decision and alternative-path rate as guard rails
• Experience - Recruiting or RecOps background with real exposure to hiring tooling; evidence of running a measurement process; no machine learning engineering required
• Decision rights - Can pause an agent in production; cannot override a hiring decision

That last line is worth keeping. The authority to stop a misbehaving system is what makes the accountability real, and the inability to override a hiring decision is what keeps the role honest.

Picking the first agent to put a manager on

Do not start with the most advanced agent. Start with the one where a wrong judgement is both consequential and measurable, which in most Australian employers means the screening or shortlisting step on your highest-volume role family. Volume gives you enough cases to measure within a quarter, and the step is close enough to the hiring decision that acceptance data already exists in some form.

Start the role at a fraction of one person's time rather than a new headcount. A 0.2 full-time-equivalent allocation with a named owner, a calibration cycle and one metric will tell you within two quarters whether the full role is justified. That is a far easier case to fund than a speculative hire, and it produces the evidence the business case needs.

Frequently asked questions

Do we need a dedicated hire for this? Usually not at first. Most 200-plus employee employers should start by giving the accountability to an existing RecOps or senior recruiting person at around 20 per cent of their time, with the metric and the calibration cycle written down.

Does the agent manager need a technical background? No machine learning engineering is required. The job is measurement, judgement and process design. Comfort with data and vendor management matters more than the ability to build a model.

How is this different from RecOps? RecOps owns process and systems. The agent manager owns whether a specific automated judgement is any good and improving. In a smaller organisation the same person may do both, but the accountabilities should still be written separately.

Who is accountable when an agent rejects a good candidate? The agent manager is accountable for the system that produced the error and for detecting it. The hiring decision stays with the hiring manager. That split is the point of defining the role.

What if our vendor says calibration is handled for us? Ask for the evaluation set, the cadence and the results on your own data. A vendor can calibrate its model and still not match your hiring managers' intent, because your intent is not in their training data.

Fund The Agent Manager Role Deliberately

The agents are already in the process at most mid-sized Australian employers. The only open question is whether anyone is accountable for what they produce. Name the owner, give them one metric and the authority to pause a system, and put them where hiring quality is owned. A fifth of someone's week and a written accountability beats a new headcount nobody has justified yet.