AI Agent for Structured Recruitment
Most screening is unstructured comparison against an unwritten standard, which is where bias enters and stays. This agent turns the role into explicit, testable requirements, assesses each candidate against each one with the evidence quoted, and never produces a ranking or a decision.
What is an AI recruiting agent?
An AI recruiting agent is a governed software worker that supports structured recruitment. It converts a role into explicit essential and desirable requirements that a candidate could evidence, assesses each application against each requirement independently with the supporting passage quoted, prepares interview questions targeting unevidenced requirements, and leaves every progression decision to a named human reviewer.
What it does
What it is not
Six seconds a CV, against a standard nobody wrote down
Screening at volume is done fast and against a standard that exists mostly in the reviewer’s head. Two reviewers reject different candidates for the same reason, the criteria shift as the pile is worked through, and none of it is recorded well enough for anyone to establish afterwards what actually happened.
The standard is not written down
A job description lists desirable attributes in prose and each reviewer converts it into their own working checklist.
Comparison is against other candidates
Assessment drifts into ranking against whoever else applied rather than against what the role requires.
Proxies substitute for requirements
A named employer or university stands in for a capability nobody stated, which is precisely how indirect bias operates.
Decisions are unexplainable later
Asked why a candidate was rejected, the answer is a recollection rather than a record against a criterion.
An explicit standard, applied the same way every time
Requirements
Write The Standard Down First
Before a single CV is opened.
The role is converted into explicit requirements separated into essential and desirable, each expressed as something a candidate could evidence, with requirements that are really proxies for something else flagged for the hiring manager to restate.
- Essential and desirable separated explicitly
- Each requirement stated as evidenceable
- Proxy requirements flagged for rewording
- Standard fixed before screening begins
Fixed before screening
Assessment
Against The Requirement, Not The Pile
With the passage quoted.
Each candidate is assessed against each requirement independently, recorded as evidenced, partially evidenced or not evidenced with the passage that supports it — and never compared against other applicants, which is what turns assessment into ranking.
Never against others
Controls
The Things It Deliberately Ignores
And the reasoning it has to show.
Attributes with no bearing on capability are excluded from assessment, proxies are surfaced rather than used, and every judgement carries its evidence — so a decision can be explained to the candidate, the regulator or the tribunal that asks about it.
Evidence attached
How the AI Recruiting Agent runs a task
- STEP 01
Turn the role into a standard
The job description is converted into requirements split between essential and desirable, each phrased as something a candidate could demonstrate, and anything that is a proxy for an unstated capability is returned to the hiring manager to restate.
Requirement extractionProxy detection - STEP 02
Fix it before screening
The requirement set is agreed before any application is read, because a standard that is still forming while the pile is worked through will differ between the first candidate assessed and the fortieth.
Standard approvalVersion locking - STEP 03
Assess independently
Each application is evaluated against each requirement on its own terms, with no reference to other candidates, and graded as evidenced, partially evidenced or not evidenced with the passage that supports the grade.
Per-criterion assessmentEvidence citation - STEP 04
Target the gaps
Requirements a candidate has only partially evidenced become interview questions designed to test exactly that, which makes the interview an extension of the same standard rather than a separate unstructured conversation.
Gap analysisQuestion generation - STEP 05
Hand every decision over
Assessments go to a named reviewer who decides progression, and the agent produces no ranking, no overall score and no recommendation, because those are the forms in which a decision gets made by software while appearing to be made by a person.
Reviewer handoverDecision recording
Systems the AI Recruiting Agent connects to
Application intake
Assessment
Inputs, outputs and runtime
- Ingests
- Job descriptionHiring policy and frameworkApplications and CVsApproved candidate sourcesInterview structure
- Produces
- Explicit requirement setPer-requirement assessment with evidenceUnevidenced gap listInterview questionsConsistency report
- Triggered by
- Role openedApplication receivedInterview stage reached
- Human oversight
- A named reviewer decides every outcome
- Models
- Open-weight LLMs you host — Llama, Qwen or Mistral class
- Typical latency
- Under a minute per application
- Deployment
- On-premise or sovereign cloud with egress control
- Data residency
- Candidate data never leaves your network
Where the Recruiting Agent pays back
Requirement Definition
Turn a job description into explicit essential and desirable requirements a candidate could actually evidence.
Structured Screening
Assess each application against each requirement with the supporting passage quoted for the reviewer.
Candidate Summaries
Produce a profile organised by requirement rather than by the order the candidate chose to write things.
Interview Question Preparation
Generate questions that probe the requirements a candidate has only partially evidenced on paper.
Consistency Review
Report where the same evidence led to different outcomes across reviewers or across the screening period.
Process Coordination
Track where each candidate is, what is outstanding and which reviewers have not yet completed their assessment.
AI Recruiting Agent vs chatbots and SaaS copilots
Any tool that returns a ranked shortlist has made the hiring decision and left a person to ratify it, which is both the appeal and the reason this area attracts specific regulation.
| Generic chatbot | SaaS copilot | VDF AI | |
|---|---|---|---|
| Standard applied | Implicit | Keyword match | Explicit written requirements |
| Comparison basis | Against other CVs | Against other CVs | Against the requirement only |
| Output shape | A ranking | A score | Per-criterion with evidence |
| Proxies | Used freely | Used freely | Flagged and not applied |
| Explainability | None | A score | Quoted passage per judgement |
| Makes the decision | Effectively yes | Effectively yes | Never — a named reviewer |
| Where applications sit | Vendor service | Vendor cloud | Inside your own network |
Governance and controls
Recruitment is among the few areas where automated assessment is specifically regulated, and the reason is straightforward: the decisions materially affect people outside the organisation who have no visibility of how they were made.
No ranking or overall score
Candidates are never ordered
No automated rejection
A named reviewer decides each outcome
Evidence quoted per judgement
Every grade cites its passage
Protected attributes excluded
Not read into any assessment
Proxy requirements surfaced
Indirect criteria sent back for rewording
Full decision trail retained
Outcomes explainable to a candidate
Evidence it leaves behind
What changes after rollout
Who runs the AI Recruiting Agent
Talent acquisition lead
Can show that every candidate for a role was assessed against the same written standard, and can produce the specific criterion and evidence behind any outcome months after the process closed.
Hiring manager
Is made to state what the role actually requires before seeing anyone, which is uncomfortable once and then repeatedly useful, and gets interview questions aimed at what the paper did not establish.
Employment counsel
Finds a process where no decision was made by software, no candidate was scored against another, and the reasoning behind every outcome is recorded against a stated criterion.
Questions about the AI Recruiting Agent
What is an AI recruiting agent?
It is an agent for structured recruitment: converting a role into explicit evidenceable requirements, assessing each candidate against each requirement with the supporting passage quoted, preparing interview questions for the gaps, and coordinating the process.
How is an AI recruiting agent different from a generic chatbot?
A chatbot will rank CVs and tell you the best one. This agent assesses each candidate against a written standard, never compares applicants with each other, and never produces a ranking.
Can an AI recruiting agent run on-premise on recruitment data?
Yes, and it is required rather than preferred. Applications are personal data from people outside your organisation, frequently including special category information, and they stay inside your perimeter.
What does an AI recruiting agent produce, and in what format?
An explicit requirement set with proxies flagged, per-candidate assessment against each requirement with quoted evidence, gap-targeted interview questions, and a consistency report.
Where does an AI recruiting agent fit in a governed AI programme?
It prepares assessment; people decide. Progression, rejection and hiring are human decisions with a named reviewer, and candidate sourcing belongs to the talent sourcing agent.
Why does it refuse to rank candidates?
Because a ranking is a decision wearing the clothes of a recommendation. Once a shortlist is ordered, the human review that follows overwhelmingly ratifies the order, and the organisation has automated a hiring decision while believing it has not. Assessing each candidate against the written standard independently keeps the comparison and the judgement where they belong, with the reviewer.
Is automated candidate assessment lawful?
It depends on the jurisdiction and on how the system is used, which is why this agent is built the way it is. Under the EU AI Act, systems used for recruitment and candidate evaluation fall into the high-risk category, and data protection law restricts decisions based solely on automated processing. Producing evidence rather than decisions, keeping a named human reviewer and retaining an explainable trail is what keeps the deployment on the right side of both. Your own legal advice still governs.
How is this different from the AI Talent Sourcing Agent?
Sequence and subject. The talent sourcing agent finds candidates: searching approved sources against a role and building a pipeline of people who might be interested. This agent works on people who have applied, assessing them against an explicit standard and running the process from requirements through interview preparation. Sourcing fills the funnel; this one structures the assessment inside it.
What does flagging a proxy requirement mean?
A proxy is a requirement that stands in for a capability nobody stated — a named employer standing in for scale experience, a degree standing in for analytical ability, a minimum number of years standing in for depth. They filter on something correlated with the real requirement and also correlated with things that must not be filtered on. The agent identifies them and asks the hiring manager to state the underlying capability instead.
Can it read candidate profiles from external sites?
Only sources your organisation has approved, and only material the candidate has made available for that purpose. General searching of a named individual pulls in personal information that has nothing to do with the role and everything to do with how bias enters a process, and it is frequently unlawful to base an assessment on. Assessment stays on what was submitted for the application.
Assess against a standard you wrote down
See the AI Recruiting Agent turn a role into requirements and evidence each assessment.