An AI recruiting agent does recruiting work — finding candidates, reaching out, screening, interviewing, writing up the notes — directly inside the recruiter's workflow, with a human approving every action. It's a different category from the AI most teams have used so far: not a chatbot that answers questions or an autocomplete that suggests text, but software that can actually act on your pipeline, under supervision.
Chatbot vs. autocomplete vs. agent
Three things often get lumped together as "AI for recruiting," and they're not the same:
- A chatbot answers questions ("what's our PTO policy?"). It informs; it doesn't act.
- An autocomplete / writing assistant drafts text — a job description, an outreach email. Useful, but a human still does all the work around it.
- An agent takes actions in the system — advancing a candidate, sending an interview, booking a call — and a human supervises.
The leap from the first two to the third is the leap from "AI that talks" to "AI that does the job." That's also where the trust question gets real, because something that can act on your pipeline can, in principle, get it wrong.
What makes it trustworthy: propose → confirm → undo
An agent you can hand real authority to needs a control model, and this is the heart of it. On 100Networks, every action that changes something is governed by propose → confirm → undo:
- Propose — the agent stages the action (e.g. "reject these 12 candidates," "send this take-home to these 8") as a proposed action, not something it executes silently.
- Confirm — a recruiter approves it. Or lets it run and…
- Undo — reverses it afterwards, within a 24-hour window on supported actions.
The result: AI cannot quietly change a hiring outcome. A human is accountable for every change, and the agent's reach is always reviewable. That is what separates an agent you can actually deploy from a demo that looks impressive but nobody trusts with their pipeline.
Around that sit the rest of the guardrails: approval rules you write, per-skill autonomy set to off / suggest / auto, a company-wide kill switch, and a full audit log you can export.
One agent, or several?
The useful distinction is between an agent that does everything vaguely and several that each do one thing precisely. 100Networks ships six: five specialists, and a conductor that runs them.
- Sourcing Agent — turns your job description into a structured spec, screens profiles against it with the biased fields stripped out, critiques its own coverage and re-sources the gaps. Returns at least 35 ranked profiles, each with per-criterion evidence.
- Outreach Agent — reaches candidates over email, LinkedIn and AI voice calls, picking the channel per person and escalating on its own, inside pacing caps that protect your domain.
- Screening Agent — writes a scoring blueprint from the JD rather than applying a canned rubric, sorts candidates into advance / review / reject, calibrates the thresholds against who you actually hired, and reports disparate impact.
- Interview Agent — runs five kinds of round: AI voice screening, live coding with real execution, system design, MCQ and take-home assessments, all proctored.
- Notetaker — joins your Zoom, Meet, Teams or Webex round, scores each competency on your own template's weighting, and leaves the scorecard already drafted.
- 100Networks Pilot — the conductor. It runs the other five from 116 specialized tools and one plain-English chat.
Each of those is an action, not a suggestion — and each is proposed for a human to confirm.
Why teams adopt them
The value compounds where recruiters feel the most pain: many open roles and more applicants than there are hours to evaluate them. Agents do the repetitive, high-volume work — sourcing, outreach, screening, first-round interviews, grading, scheduling, follow-ups — consistently and around the clock, and hand humans the decisions. You get the throughput of automation with the accountability of human approval.
The bottom line
An AI recruiting agent isn't a chatbot bolted onto your ATS. It's software that does the work, governed by a control model that keeps a human in charge. If you want to see what that looks like in practice, start with the six agents, or read how 100Networks compares to a traditional ATS like Greenhouse.