Greenhouse opens its ATS to AI agents while 30% of candidates already use AI agents to apply: what agent-to-agent recruitment means for Sydney agencies

On 7 May 2026, Greenhouse confirmed the launch of Greenhouse MCP, a governed Model Context Protocol layer that lets any approved AI agent act directly inside the Greenhouse ATS. Design partners include StubHub and Komodo Health. Customer rollout begins June.

The same announcement included new data from a Greenhouse survey of 2,950 active job seekers. Thirty per cent are already using AI agents to search for jobs. AI agents are auto-submitting applications. AI agents are scheduling interview screens. This is not a future-state survey. This is the current state.

The implication is straightforward and uncomfortable. The candidate filling out the application is increasingly an AI. The reviewer screening the application is increasingly an AI. The recruitment agency in the middle is still pricing the work as if both sides are human.

The agent-to-agent workflow is here

The MCP launch is not the news. The 30% number is the news. The infrastructure for agent-to-agent recruitment is being built on both sides simultaneously:

• Candidate side: AI agents search job boards, parse JDs, draft tailored applications, submit them, schedule screens, and prepare interview answers. Used by 30% of active job seekers today.
• Employer side: AI agents act inside the ATS via MCP, doing sourcing, screening, scheduling, and increasingly first-round interviewing. LinkedIn Hiring Assistant and now Greenhouse MCP are the visible examples.

Within nine to twelve months, the dominant traffic on most enterprise ATS systems will be agent-to-agent. The human recruiter sits in the middle of a conversation between two AIs.

What survives the agent-to-agent workflow

The work that gets commoditised first is the work both agents can do equally well at low cost:

• Sourcing: both AIs can search
• Application processing: both AIs can read JDs and parse CVs
• Basic screening: both AIs can match keywords and qualifications
• Interview scheduling: both AIs can coordinate calendars

These four lines used to make up the bulk of a contingent recruitment fee. They no longer carry significant price. The cost per transaction is now measured in cents, not dollars. Any agency still pricing these lines at 2024 rates is going to lose them.

The work that survives is the work no agent can do reliably:

• Qualifying conversation with the hiring manager. The JD is wrong. The agency knows it is wrong. The conversation that uncovers what the role actually needs to look like is judgement work. No agent can hold this conversation in 2026.
• Counter-offer management. The candidate accepts the offer. Two days later the current employer counters. The candidate goes silent. The conversation that brings the candidate back is human work: pressure, silence, motivation, reading what is not said.
• Hiring manager pushback. The shortlist is wrong. The hiring manager wants more candidates. The CFO has frozen hiring. The agency's job is to broker the conversation between these three people. Agent work cannot do it.
• Candidate rebuilding after a failed final. The candidate fell over at the last stage. They need to be back in the pipeline for the next role. Agents cannot rebuild confidence. Humans can.

This is the work the surviving agencies will reprice up. The work the platforms have commoditised gets repriced down, or removed from the offer entirely.

The repricing every Sydney recruitment agency needs to do

Three steps.

Step one: Audit the current retainer line by line. Mark every line as "agent-replaceable" or "agent-resistant." Be honest. Sourcing, application processing, basic screening, scheduling: these are agent-replaceable. Diagnostic conversations, counter-offer management, hiring manager negotiation, candidate rebuilding: these are agent-resistant.

Step two: Reprice each line to match the new reality. Agent-replaceable lines need to be repriced at the new market rate, which is near zero for the in-house team. Either fold them into a fixed monthly subscription that bundles them at low cost, or remove them from the offer entirely. Agent-resistant lines should be repriced up: these are now the entire defensible value of the agency.

Step three: Restructure the agency's pitch around the work no agent can do. The old pitch led with "we will source and screen faster than your in-house team." That pitch is dead. The new pitch leads with "we will hold the four conversations your AI cannot hold": diagnostic, counter-offer, hiring manager, candidate rebuild. The fee structure follows.

The agencies that will not survive the agent-to-agent ATS

The agencies that will not be around at the end of 2026 are the ones that respond to the agent-to-agent workflow by trying to compete on speed. "We will source faster than your AI." This argument has already lost. The AI sources at zero marginal cost and operates 24/7. No human recruitment team competes on speed against that.

The agencies that survive are the ones that stop competing on the commoditised work entirely. They reprice it down or remove it. They reprice the judgement work up. They restructure the conversation with the client around the part no agent can replace.

Greenhouse just made the agent-to-agent workflow real. Every Sydney recruitment agency owner has 90 days to decide whether they reprice now, or get repriced by the market in Q3.