Artificial intelligence (AI) is most useful in hiring when it prepares the work and a person makes the call. If you’re working out how to use AI in recruiting, hand an AI agent the preparation at each stage of hiring, and keep every decision about a candidate with a person. An AI agent is software that carries out part of a hiring stage, such as building a list of candidates or reading applications, and hands the result to a person.

Many teams start with the tools and decide later what each one should do. Recruiters pick up tools one by one and hand each one a little more of the work over time. A few months later, no one has written down what each tool may do without a person checking it.

I recommend deciding where the agent’s work ends and a person’s decision starts for every stage before any agent runs. Then bring agents in gradually, starting on one live role, and read what each one produces before you trust it.

This guide is for recruiters and talent leaders working out where AI agents fit in their hiring process. It sets out what to hand an agent at each stage and what to keep. Then it covers the hiring brief every agent works from, the order to add agents in, how to run the first one on a live role, and how to tell whether your use of AI is working.

What to hand an AI agent at each stage of hiring.

Break hiring into seven stages. At each one, an agent can take on part of the preparation, and a person keeps the decision. Here’s a default split to start from and change for your team:

Stage What an agent can prepare What a person keeps What to check before you trust it
The brief A first draft of the requirements, from the job description and the intake conversation The requirements, what rules a candidate out, and sign-off on the brief The draft says only what the hiring manager said
Sourcing A list of candidates who match the brief, with the reasoning for each, and a first draft of each outreach message Who gets contacted, and what the message says You can see why each candidate is on the list
Reviewing applications A first read of every application against the brief, sorted into groups with a reason for each Every decision to advance or reject a candidate The reasons point back to the brief, and a recruiter would sort the same way
Scheduling Proposed interview times, from the calendars of the people involved Who interviews, and any exception a candidate asks for Times fit each person’s time zone and working hours, and the right interviewers are on each invite
Interview notes and scorecards Notes from the interview and a draft of the scorecard The rating and the recommendation The notes say what the candidate said, without adding to it
The debrief and decision A side-by-side summary of each interviewer’s feedback The hiring decision and the reason for it The summary keeps disagreements between interviewers in view, and each interviewer submits their own feedback before reading it
Reporting A weekly summary of where each candidate stands and what’s waiting on whom What to change in the process The summary agrees with the record in your applicant tracking system (ATS)

Read across any row and the split is the same: the agent prepares, and a person decides. The right-hand column lists what to check in the first weeks, before anyone on the team relies on what the agent produces.

For a longer look at how recruiters and agents share the work from the intake call to the offer, watch this talk, recorded live at a recruiting conference in Amsterdam.

Sign off the brief before other agents run.

The agents at every later stage in the table work from the hiring brief, so a vague brief turns into vague output at every stage. An agent can draft the brief, but a person signs it off before any other agent runs on the role. Make sure it covers these points:

  • What the new hire must deliver in their first months, in the hiring manager’s words.
  • The skills and experience the role requires, and the ones that are only nice to have.
  • What rules a candidate out.
  • The work the agent may finish alone on this role, such as the weekly report, and the work it hands to a person to read before anything moves forward.
  • Who reviews the agent’s output, and how soon after it arrives.

The last two points are the ones teams tend to leave out. Without them, each recruiter decides alone how far to trust the agent, and two recruiters on the same team end up working to different rules.

A separate guide covers how to write a hiring brief you can delegate to a person or an agent, point by point.

Add agents one stage at a time.

When agents arrive at every stage at once, it’s hard to tell which one a problem came from. Add them in an order that lets you check each stage before the next one starts:

  1. Start with the stages where a person reads everything the agent prepares and can check it easily, adding them one at a time: the first draft of the brief, interview notes, a first read of applications, the debrief summary, and the weekly report.
  2. Add sourcing once the brief has held up on a few roles, since a candidate list is only as good as the brief behind it.
  3. Add scheduling, the stage that reaches candidates directly, last.

Each stage hands its output to the next one, so the handoff between two agents needs a reader too. Name who reads each handoff before you add the stage after it.

There’s also a guide to running several AI recruiting tools as one coordinated stack, including how to check each handoff between them.

Run the first agent on one live role.

A pilot on one live role shows how an agent handles real candidates before you change how the whole team works. Work through these steps with the recruiter and hiring manager on that role:

  1. Decide what candidates are told about AI, and get sign-off from whoever owns legal and data privacy at your company, before the pilot starts.
  2. Pick a role that’s open now and has a written brief.
  3. Check that the brief names which tasks the agent finishes by itself on this role, and who reviews its output.
  4. Agree how often that person reads the output.
  5. Read every output for the first two weeks, and mark each one used, corrected, or discarded.
  6. Fix the brief wherever the same correction comes up twice.
  7. Add the next role, or the next stage, once most outputs are used without correction.

For the first step, write what candidates are told into a short AI policy that sets out what the team uses AI for and how candidates hear about AI. Another guide covers how to write an AI policy for recruitment.

Where Metaview fits.

Metaview is the Agentic Recruiting Platform, and its agents do the preparation at three of the stages in the table: sourcing, reviewing applications, and interview notes.

From one plain-language prompt, its sourcing agent searches the web, your connected ATS, and the conversations your team has recorded with Metaview, and returns candidates with the reasons it chose each one. It can also start from an intake call recorded with Metaview instead of a prompt. The recruiter decides who to contact.

Metaview’s Application Review agent drafts an ideal candidate profile (ICP) from the job description and the context you add. The ICP does the job of the brief at this stage, and a recruiter reviews and approves it before any application is read against it. Application Review then reads every application and sorts each one into a fit group, from Great down to Poor. It shows the reasoning for every application. The recruiter makes every call to advance or reject.

Applications in Metaview Application Review grouped by ICP fit, with the written reasoning for one applicant beside the list
Applications grouped by fit to the ICP, with the reasoning for one applicant open beside the list. The applicants shown are sample data.

Where consent has been given, the Metaview Notetaker records and transcribes an interview, then turns the conversation into structured notes and a draft scorecard. The recording captures every spoken word, so the draft starts from what the candidate said rather than from memory. The interviewer reviews the draft, rates the candidate, and submits it.

An agent’s preparation only saves work if the recruiter can trust what it returns. Shahriar Tajbakhsh, Metaview’s co-founder and chief technology officer, has described the cost of output a recruiter can’t rely on:

If results aren’t reliable, AI doesn’t reduce work. It just moves it, and recruiters end up validating, correcting, and second-guessing the output.”
Shahriar Tajbakhsh Shahriar Tajbakhsh Co-founder and Chief Technology Officer · Metaview

That’s why, during the pilot, a person reads every output: it shows whether the agent’s work holds up before the whole team depends on it.

See what an AI agent prepares for each application.
Book a demo to see Application Review sort applications by fit to the ICP, with the reasoning for each.
See it live

How to tell your use of AI in recruiting is working.

A month into the pilot, use these five checks to see whether it’s working before you add the next stage:

  1. Can each reviewer say which tasks the agent finishes by itself on their roles? If not, write them into the brief.
  2. Does the agent’s output match the brief? If the same correction keeps coming up, fix the brief before you add anything.
  3. Was every candidate decision made by a named person, with a reason recorded? If not, name who records the reason on each role.
  4. Does the next person in the process use the agent’s output? If they redo the work, ask them what the output left out.
  5. Do candidates know where AI is used in your process? If not, add it to your AI policy and your messages to candidates.

Using AI in recruiting works when every stage has a written answer to two questions: what the agent prepares, and what a person decides. Write both down for one role, run it, and add the next stage once the first one holds.

See it in action.

Hand agents the preparation, and keep the decisions.

See how Metaview’s agents prepare the work for sourcing, application review, and interview notes, with every call on a candidate left to your team.

Frequently asked.

What is agentic AI in recruiting?

Agentic AI in recruiting means AI agents that carry out part of a hiring stage, such as building a list of candidates or reading applications against a brief, and hand the result to a person. The person reviews it and makes the decision at that stage.

Which recruiting stage should get an AI agent first?

When you’re working out how to use AI in hiring, start where a person can check the agent’s output easily, such as interview notes or a first read of applications. Scheduling, the stage that reaches candidates directly, comes last.

Can an AI agent reject candidates?

Keep every rejection with a person. An agent can read applications and draft the reasoning for each one. The recruiter or hiring manager then reads that reasoning and makes the call, with their own reason recorded.

How do you tell whether an AI agent is helping?

Track how often the person who reads its output uses it as it is, corrects it, or discards it. More output used without correction shows the brief fits the role, and a correction that keeps coming up shows what the brief is missing.

What should stay human in recruiting?

The calls on candidates stay human: who to contact, who advances, who’s rejected, and who gets an offer. People also keep the brief, the rating on each scorecard, and what candidates are told about how AI is used.