Go from a job description to a screened shortlist

Define criteria, post the role, let an agent score applications, then you decide who advances.

For anyone hiring without a recruiter · 8 steps · 7 min

How it works today.

You write a job description in Google Docs, paste it into your ATS or LinkedIn, and wait. Applications arrive in a steady drip for two weeks. You open each PDF, skim the career history, try to remember what mattered most for the role, and sort people into maybe, no, and 'come back to this one'. By day ten you have forty unread applications and you are second-guessing whether the person you liked on day two would still make the cut.

The last five applications get less attention than the first five. You miss someone good because their CV opened slowly. You advance someone because their cover letter was memorable, even though two of your must-haves are missing. Then you schedule six calls and realize halfway through the third that you are asking different questions each time.

Before you start.

All of it has to be true, or step one fails in a way that is annoying to debug.

A job description drafted in a doc or ATS, even if it is still rough.
An ATS or inbox where applications will land, with API or export access if you want the agent to read directly.
A list of the actual skills or experience that matter for this role, even if informal.

The steps.

01
Write down your screening criteria before any CV arrives
Open a new document and list three to five specific, job-related criteria that will determine who moves forward. These might be years of experience in a technology, evidence of a particular outcome, or a required certification. Write each as a yes/no question or a scale. Do not use vague language like 'culture fit' or 'passion'. Save this document; you will give it to the agent and refer to it when you make final decisions.
Paste this
```
Review this job description: [paste job description]. Suggest five specific, measurable screening criteria I can use to evaluate applications fairly. Frame each as a question I can answer from a CV or cover letter.
```
Finalize and post the job description
Publish the role to your ATS, LinkedIn, job boards, or wherever you source candidates. Make sure the application instructions are clear and that CVs will arrive in a single place. If you are using an ATS, confirm that it can export applications as structured data or that an agent connector exists for it. If applications come by email, set up a dedicated folder or label.
Connect the agent to your application source
In your agent platform, configure access to the ATS, inbox, or folder where applications live. If your ATS has no connector, export applications as a spreadsheet with columns for name, email, CV text or attachment link, and submission date. Upload this export to the agent. Confirm the agent can read each application; if CVs are PDFs behind a login wall, you may need to download and re-upload them.
04
Give the agent your scoring rubric and a sample application
Paste your screening criteria into the agent as instructions. Include the job description for context. Then feed the agent one real application and ask it to score that candidate against each criterion, showing its reasoning. Read the output carefully. If the agent is too lenient, too harsh, or citing irrelevant details, revise your criteria or add clarifying examples. Do not proceed until the agent's judgment on this sample matches yours.
Paste this
```
You are screening applications for this role: [paste job description]. For each applicant, score them against these criteria: [paste your screening criteria]. For each criterion, give a score of 0 (no evidence), 1 (partial evidence), or 2 (clear evidence), and cite the specific sentence or bullet from their CV that supports your score. Return a summary score and flag any applications that meet fewer than [X] criteria as 'do not advance'. Here is the first application: [paste CV text].
```
Run the agent on all applications and export the scored list
Point the agent at the full set of applications and let it score each one. Depending on volume, this may take minutes or hours. Once complete, export the results as a spreadsheet with columns for name, email, summary score, per-criterion scores, and the agent's cited evidence. Sort by summary score descending. Do not delete or filter anyone yet.
Review the top-scored candidates and the bottom-scored candidates
Open the spreadsheet and read the agent's reasoning for the top ten candidates and the bottom ten. Check whether the agent is citing the right details and applying your criteria consistently. If you see a candidate scored low but you know they should advance, note what the agent missed. If someone scored high but feels wrong, check whether your criteria need refinement. This is your calibration step.
Decide who moves forward and who is rejected
Go through the scored list and mark your decision for each candidate: advance, reject, or unsure. The agent has done the reading and provided evidence, but you make the call. For anyone you reject, you are accountable for that decision being fair and job-related. For anyone marked unsure, read their CV yourself or ask the agent a follow-up question about a specific criterion. Do not let the agent auto-reject anyone.
Send advance and rejection emails, and schedule interviews
Export the 'advance' list and send calendar invites or scheduling links. Use your ATS or a separate agent to send rejection emails to those you have decided not to progress, thanking them for applying. Keep the scored spreadsheet as a record of how you made decisions. If you are audited or questioned later, this document shows your process was consistent and criteria-based.

What you keep.

Automating the typing does not move the accountability. These stay with a person.

Writing the screening criteria before any CV is read, so the process is fair and defensible.
The final decision on who is rejected. The agent scores and surfaces evidence; you own the outcome.
Calibration: reading the agent's reasoning on a sample of candidates to confirm it is applying your criteria correctly.
Any decision to override the agent's score, with a note explaining why.

Once it works.

The first run is the demo. These are where the time actually comes back.

Fire it from an event

Set the agent to score each new application within an hour of submission, so your shortlist is always current and you can move fast on strong candidates.

Run it on a loop

After each hiring round, compare the agent's scores to your final hires and rejections, then update your criteria or prompt to reflect what actually predicted success.

Hand off to another agent

Once you have your shortlist, hand the scored spreadsheet to a scheduling agent to send calendar invites in batch, rather than doing it manually.

Scale it wider

If you hire for the same role repeatedly, save the scoring rubric and prompt as a template so the next search starts with a calibrated agent on day one.