What is the world's first fully agentic ATS?

By the strict definition of the term, the first fully agentic ATS is Recruitifly: one assistant, Fly, that can plan and execute multi-step recruiting work on every screen of the platform, with every action proposed first and confirmed by a recruiter. The phrase matters because most systems now advertise some form of AI, and almost all of it is a chatbot attached to a single feature. Fully agentic means something stricter: one assistant, the whole platform, real execution, human sign-off. This article pins the definition down so you can test any vendor’s claim, including ours, in a live demo.

What makes an ATS fully agentic?

Three tests separate an agentic ATS from an ATS with AI features.

1. One assistant, present everywhere. The same agent on the pipeline board, the candidate list, the job editor and the analytics view, not a different widget per screen with a different memory.
2. It plans. Give it a goal that takes several steps and it works out the sequence itself, using the platform’s own functions to search, rank, schedule, draft and post.
3. It executes inside the system. It does not hand you text to paste somewhere else; it stages the actual change, shows you exactly what will happen, and applies it once you confirm.

A chatbot that answers questions about your pipeline fails the second test. A job description generator fails the first and the third. Both are useful features; neither makes a platform agentic. If you are still mapping the basics of the category, start with what an applicant tracking system is and come back.

The distinction that matters: proposed, then confirmed

Agentic is often heard as autonomous. It should not be. An agentic system plans and performs work; an autonomous one decides without you. In hiring that difference is everything: candidates are people, stage moves and rejections have consequences, and European rules on automated decision-making expect a human accountable for hiring outcomes.

The honest design, and the one Recruitifly ships, is propose and confirm. Fly will find the candidates, draft the message, line up the interview and prepare the job post, and then it stops. Every write arrives as a proposal you approve or cancel. Nothing changes a record, messages a candidate or publishes a vacancy without a person saying yes. A vendor promising hands-off hiring is not offering a better agent; it is offering a liability.

What does a fully agentic ATS handle end to end?

The practical test is whether one conversation can carry a role from intake to report. A working example, six asks typed into one thread:

1. Match. Score the new applicants against the Rotterdam sales role. The agent ranks them against the job’s requirements.
2. Shortlist. Keep the strongest five and compare the top three side by side.
3. Schedule. Set up first interviews with the top three for next week. Slots are proposed, you confirm.
4. Draft. Write the outreach and follow-up messages, in Dutch or English. Drafts appear for approval.
5. Post. Publish the follow-up vacancy to the boards used last time. The posting is staged, then published on confirm.
6. Report. Build a dashboard of pipeline and source performance for this role.

In a conventional ATS each of those is a separate feature, often a separate screen, sometimes a separate vendor. The agentic version is the same work driven from one place. For a sober, task-by-task view of which of these steps automate well and which should stay human, see what recruiting work can actually be automated.

Chatbot, copilot, or agent: what do most ATSes ship?

Most established vendors, Greenhouse, Workable, Lever and Teamtailor among them, have added genuinely useful AI features, things like drafting job descriptions, summarising CVs or suggesting next steps. Almost all of it sits in the first two columns of this table.

	Chatbot feature	Copilot	Fully agentic
Where it lives	One screen, one task	A sidebar on some screens	Every screen, plus chat apps
What it produces	Answers and drafts	Suggestions in context	Executed, confirmed work
Multi-step goals	No	Rarely; you drive each step	Yes; it plans the sequence
Writes to the system	No	Occasionally, single actions	Yes, each one proposed first
Demo question to ask	Draft a job ad	What should I do next here?	Shortlist, schedule and report on this role

The gap between copilot and agent is not intelligence, it is integration. A copilot that cannot post the job, move the candidate or book the interview leaves you as the integration layer, clicking through the same screens as before with better suggestions. That is why partial copilots do not qualify as agentic, whatever the label says.

How does Fly work across the whole platform?

Recruitifly was built around one assistant rather than retrofitting one. Fly docks as a panel on every workspace, from the pipeline board to analytics, and a floating version follows you everywhere else, so the agent is never more than a click away. You ask in plain language; results come back as clickable cards, shortlists, side-by-side comparisons and dashboards rather than walls of text, and every proposed change arrives as a confirm-or-cancel card. From the Pro tier upward Fly also answers on Slack and Telegram, with WhatsApp available as an add-on, so you can approve an interview from your phone between meetings.

Commercially it is metered in Actions: one completed task, such as parsing a CV or drafting outreach, is one Action. The Freelancer tier at EUR 70 per month includes 150 of them, and Fly itself is included on every tier rather than sold as a premium module. The Fly page shows the full range of what it can do, alongside the rest of the platform’s features, and the pricing page keeps the numbers current.

How to verify an agentic claim in a demo

Do not ask a vendor whether their ATS is agentic; everyone says yes. Ask the assistant to do a real multi-step task and watch.

1. Give it a goal, not a command. Get the three best applicants for this role to interview next week. That should produce a plan, not an error.
2. Watch where the work happens. Text you must copy elsewhere is a writing aid. Changes staged inside the product are agency.
3. Check the confirmation step. You want to see exactly what will change before it does, and a cancel button that works.
4. Change screens and ask again. If the assistant only exists on one page, it is a feature, not an agent.
5. Ask it to report on what it did. An agent that executes work should be able to account for it.

Recruitifly is in private beta, and this is exactly the test we invite. Talk to us, bring a real role, and put the same five questions to Fly.