Skip to content

The Agentic Stack: Orchestrating Recruitment AI Through Your ATS

Picture of Mark Gray

Overview

  • Recruitment AI is splitting into specialised agents, each owning one stage of the hiring funnel

  • Nine agent types now cover the lifecycle, from role intake through to onboarding

  • Agents do not replace your ATS; they sit on top of it, with the ATS as the system of record

  • Orchestration, policy and human escalation determine whether the stack actually delivers

txthr Demo

The idea of one AI model running recruitment end to end has quietly been retired. What has replaced it is a stack of focused agents, each owning a single stage of the funnel, all coordinated through the ATS. We map the full picture in our guide to specialised AI agents across the talent lifecycle, and this article walks through how the stack fits together.

The shift is well underway. In July 2026, Josh Bersin reported that multi-agent AI systems are moving through talent acquisition at pace, particularly in high-volume hiring, with his HR 2030 blueprint mapping 24 distinct agent workflows.

Why one model was never going to work

Recruitment is not one job. Intake, sourcing, screening, scheduling, interviewing, compliance and onboarding each have their own data, rules and failure points. A general-purpose model stretched across all of them does each one badly.

Specialised agents flip that. Each is built for a narrow task, measured on a narrow outcome, and hands off to the next agent through the ATS. Gartner names the AI revolution as one of two forces shaping talent acquisition in 2026, and predicts that by 2027 three in four hiring processes will test candidates for AI proficiency. Hiring is becoming an AI-native workflow, whether or not your stack is ready for it.

The nine agents, stage by stage

1. Open the role

  • Intake Agent: runs structured intake with the hiring manager, captures success criteria and surfaces internal candidates before the requisition is even posted.

2. Build the pipeline

  • Sourcing Agent: continuously scans internal and external talent pools, scores candidates on fit and triggers outreach. Sourcing becomes ambient rather than a sprint.

  • Personalisation Agent: adapts message tone, content and timing to each candidate's profile and engagement signals, lifting response rates without adding recruiter workload.

  • Voice Screening Agent: conducts natural-language phone screens after hours, adapts questions to answers and fairly assesses far more candidates than a human team could reach.

3. Run the interview

  • Self-Scheduling Agent: lets candidates book against live interviewer availability, manages panels, time zones and reschedules. It removes the back-and-forth that drops candidates.

  • Interview Agent: provides live guidance, transcripts, structured notes and scorecards, so interviews become consistent and reviewable across teams.

  • Fraud Detection Agent: verifies identity continuity across stages, flags scripted answers and behavioural inconsistencies, and produces audit-ready evidence.

4. Close the loop

  • Compliance Agent: embeds POPIA, Employment Equity and role-specific document collection into the flow, tracking completion and flagging exceptions before they delay a start date.

  • Onboarding Agent: shepherds new hires through their first 90 days across HRIS, IT and facilities, escalating blockers before they turn into early attrition.

Agents are only useful when they are orchestrated

None of these agents replaces your ATS. They sit on top of it. The ATS stays the system of record: every candidate, decision and document lives there, while an orchestration layer routes work between agents, applies your policies and escalates to a human whenever judgement is required.

That architecture matters for two reasons. The first is accountability. When an agent screens or rejects, the evidence trail has to live somewhere auditable. The second is adoption. Recruiters will not manage nine separate tools, but they will happily manage one pipeline in a platform like Neptune where agents do the legwork behind each stage. Conversational tools such as txthr already show the pattern at the engagement layer: the candidate experiences one conversation, and the ATS records everything behind it.

Where the recruiter fits

The stack does not shrink the recruiter's role. It moves it up a level. LinkedIn's Future of Recruiting research found that recruiters using AI-assisted messaging are 9% more likely to make a quality hire, and describes the winning recruiter as a strategic talent advisor rather than an administrator. Orchestrated agents free the hours. Judgement, relationships and final decisions stay human.

If you are evaluating agentic AI, start with the funnel rather than the tools. Which stage leaks the most candidates, and which agent would fix it first? Then make sure whatever you add reports back into the ATS. A stack without a system of record is just nine disconnected experiments.

Explore the full agent map and how orchestration works in practice in our AI recruitment guide.
, , ,