What is an agentic ATS?

An agentic ATS is an applicant tracking system where a built-in assistant can plan and carry out multi-step recruiting tasks, not just store candidate data. You hand it a goal, such as build a shortlist for this role and schedule the top two, and it works out the steps, runs them inside the platform, and asks for approval before anything actually changes. A regular ATS waits for your clicks; an agentic one does the clicking and shows its work first.

What changes compared with a normal ATS?

A classic ATS is a database with a workflow on top: jobs, candidates, stages, and the audit trail that connects them, the baseline we unpack in what an applicant tracking system is. Everything it does, a person triggers. You open the screen, click the button, type the message.

An agentic ATS keeps all of that and adds an actor. The assistant can use the platform’s own functions: search candidates, score them against a job, draft messages, stage a posting, book an interview. The difference is not smarter suggestions, it is execution. Instead of producing text for you to paste somewhere else, the assistant stages the real change and shows you exactly what will happen before it does.

That one shift changes how the work feels. You stop navigating screens to perform a process and start reviewing proposed work. The judgment stays yours; the typing, clicking and coordinating largely stop being your job.

How much autonomy do these systems actually have?

The word agentic gets stretched over four very different levels of help. Knowing which one a vendor actually means is most of the evaluation.

Level	What it does	Who decides	Typical example
Automation rules	Fires a fixed action when a trigger matches	You, once, when you write the rule	A reminder two days before each interview
Copilot	Suggests and drafts where you are working	You, at every step	A job description draft you edit and post yourself
Agent with approval	Plans multi-step work and stages real changes	You, per proposed change	Shortlist built, outreach drafted, interviews proposed, all awaiting your confirm
Full autonomy	Acts without sign-off	The software	Auto-rejecting or auto-messaging candidates with no review

Rules are old news; every serious ATS has them. Copilots are where most of the market sits today. Agents with approval are the new and genuinely useful tier. Full autonomy is technically possible and, in hiring, a liability rather than a feature, for reasons the next sections cover.

What does an agentic ATS do in practice?

Concrete beats abstract here. Tasks a recruiting agent handles well today:

• “Post this job to the three boards we used last time.” The agent prepares the posting for each board, you review it once, it publishes on confirm.
• “Build a shortlist for the Utrecht support role and schedule the top two.” It scores applicants against the requirements, presents a ranked list with its reasoning, drafts the invitations and proposes interview slots. You approve, it books.
• “Compare these three candidates side by side.” A structured comparison, not three open tabs.
• “What is stuck in my pipeline this week?” Status questions answered from live data, with the obvious follow-up actions offered next.

Notice the shape: high-volume coordination work with a clear definition of done. Which recruiting tasks automate well and which should not is a topic of its own; we keep an honest map in recruitment automation: what can actually be automated. The short version is that agents are excellent at the work around a decision and have no business making the decision.

Why does human approval stay the safety line?

Two reasons, one practical and one legal.

Practical first: agents make confident mistakes. A wrong candidate in an outreach draft is a non-event when a recruiter reviews it, and a reputation problem when it sends itself. Approval converts errors from incidents into edits.

Legal second: in the EU, the rules point firmly at human oversight. GDPR gives candidates the right not to be subject to purely automated decisions with significant effects, and a rejection is exactly such a decision. The EU AI Act goes further and classifies AI used in hiring as high-risk, with human oversight and documentation duties attached. An assistant that proposes while an accountable person approves fits this regime; a system that rejects or messages candidates unsupervised is the configuration these rules exist to catch.

This is why agent-with-approval is not a watered-down compromise on the road to full autonomy. For hiring, it is the destination. Recruitifly’s assistant Fly is built this way on purpose: it plans and prepares everything, then stops and waits for a yes on every write.

Which questions separate a real agent from a renamed chatbot?

Plenty of products renamed their chatbot an agent this year. Five demo questions sort them out:

1. Give it a goal, not a task. “Get the strongest three applicants to interview next week.” A plan should come back, not an error or an essay.
2. Ask where the work lands. Does it stage changes inside the system, or hand you text to paste somewhere else?
3. Look for the confirm step. You should see precisely what will change before it changes, and cancel should work.
4. Switch screens and ask again. One assistant across the whole platform is an agent; a separate widget per page with separate memory is a feature.
5. Ask what it just did. An agent that executes work should account for it in plain language.

Apply these to every vendor, including us. Established names such as Greenhouse, Workable, Lever and Teamtailor ship genuinely useful copilots that draft and summarise; most stop short of planning and executing multi-step work, and that line, not the label, is what you are buying.

Where does Recruitifly fit?

Recruitifly is built around one agent, Fly, rather than a copilot bolted onto each screen. Fly is present on every workspace, takes goals in plain language, and runs the post-shortlist-schedule examples above through a propose-and-confirm flow: every change arrives as a card you approve or cancel. It is included on every tier and metered in Actions, one completed task being one Action; the Freelancer tier at EUR 70 per month includes 150 of them. From the Pro tier Fly also answers on Slack and Telegram, with WhatsApp available as an add-on, so approving an interview from your phone between meetings works too. Current numbers live on the pricing page.

The honest footnote: Recruitifly is in private beta, and no vendor, us included, should be believed on the word agentic alone. Talk to us, bring a live role, and put the five questions above to Fly before you take our word for any of this.