AI recruiting agent that does the work and asks before it decides
An AI recruiting agent is software that runs a hiring pipeline by itself. RecruitingBot takes a role, finds and scores candidates, writes to them from your mailbox, handles replies, books interviews and returns a shortlist that a human approves.
- GDPR built in
- Human approves every shortlist
- No hiring fees
-
Day 1
First messageSent
Personal note on their payments work, role facts, salary range, opt out link
-
Day 4
Follow upSent
Short nudge with one new detail about the team
-
Day 9
Last noteScheduled
Polite close, door left open
An agent acts, a tool waits to be operated
Most recruiting software is a better shovel. A sourcing database returns a list, a screening module ranks a folder of CVs, a sequencer sends whatever you load into it. Every one of those products sits idle until a recruiter clicks the next button, so the real bottleneck (a person moving work from stage to stage) never goes away. An agent is built the other way around. You describe the outcome you want, which is five qualified people on the hiring manager's calendar, and the software plans and executes the steps needed to get there.
RecruitingBot works through the same eleven stages a careful recruiter would: intake, job description and rubric, sourcing, profile analysis, outreach, replies, screening, skills test, scheduling, shortlist and follow ups. The difference is that it moves a candidate forward the moment the previous stage is done, at two in the morning or on a Sunday, without anyone opening a dashboard. You come back to a pipeline that has already advanced, with a clear record of every action the agent took and why.
- 01 Tool behavior. Returns results and waits for a recruiter to decide the next action.
- 02 Agent behavior. Takes the next action itself inside rules you set, then logs what it did.
- 03 Your role with an agent. Set the brief, review evidence, approve the people who move to interviews.
What the agent decides on its own
Autonomy is only useful if the boundaries are clear. The agent owns the repetitive, high volume decisions that a recruiter makes hundreds of times per role and rarely reconsiders. It decides which public professional profiles and imported records are worth analyzing, scores each candidate against your rubric criterion by criterion, picks who receives the first outreach message, chooses which follow up to send and when, classifies every reply, and sends screening questions to people who said they are interested.
It also handles the logistics that eat afternoons. Interested candidates get a scheduling link tied to the hiring manager's connected Google Calendar or Microsoft 365 calendar, reminders go out before interviews, and non responders who were sourced from public data drop off automatically once the retention clock runs out. None of these choices changes who gets hired. They decide where the agent spends its effort, which is exactly the part of recruiting that burns time without needing judgment about fit.
| Decision | Who makes it | Where you see it |
|---|---|---|
| Which sourced profiles get analyzed | Agent, within your search criteria | Role pipeline, sourcing log |
| Score per rubric criterion | Agent, with written evidence | Candidate drawer |
| Who receives outreach and when | Agent, inside daily caps and sending window | Outreach queue |
| How a reply is classified | Agent, editable by you | Inbox |
| Who appears on the shortlist | Agent proposes, you approve | Shortlist link |
| Who gets an offer | You, always | Outside the agent |
What always waits for a human
The agent never makes the final call on a person. Every shortlist needs an explicit approval from someone on your team before it reaches the hiring manager as a recommendation, and rejecting a candidate after an interview is a human action with a human reason. This is a design choice, and it also keeps you on the right side of Article 22 of the GDPR, which restricts decisions with significant effects that are based solely on automated processing.
You also keep control of anything that would be awkward to undo. The rubric is yours to edit before launch, the outreach sequence can be reviewed step by step, and you can pause a role, take over a conversation or remove a candidate at any moment. If a candidate asks about something outside the facts you gave the agent, such as equity details you never entered, the conversation is flagged for you instead of being answered with a guess. The rule of thumb is simple: the agent handles volume, you handle judgment and anything with consequences.
How the agent handles replies and candidate questions
Replies are where most automated outreach falls apart, because a sequence keeps firing while a real person is trying to talk. The agent reads every incoming message and sorts it into one of four groups: interested, not now, not interested, or question. Interested candidates move straight to screening. A not now reply stops the sequence and sets a polite reminder for later if you allow it. Not interested ends contact, and an opt out adds the address to your workspace suppression list so nobody on your team writes to that person again.
Questions get answered, but only from the facts of the role: salary range, location or remote policy, seniority, team, must haves and the interview process you described at intake. If someone asks whether the role allows four day weeks and you never said, the agent does not invent an answer. It tells the candidate a colleague will confirm and routes the thread to the inbox for you. You can read any conversation, edit the agent's classification, or take over the thread entirely, at which point the agent stops writing to that candidate.
- Interested. Moves to knockout and pre-screening questions, then the skills test if the role has one.
- Not now. Sequence stops, optional reminder at a date you choose.
- Not interested. Contact ends and the reason is kept for your records.
- Question. Answered from role facts, or handed to you when the facts do not cover it.
Guardrails that keep an autonomous recruiter safe
An agent that writes to people on your behalf needs hard limits, not good intentions. Outreach is sent from your own connected Gmail or Outlook mailbox with a daily sending cap per mailbox, and only between 8:00 and 20:00 in the recipient's local time, so a message never lands at midnight and your domain reputation is not spent in one afternoon. Every first message carries an Article 14 notice explaining who you are, why you are writing, where the profile was found, how long data is kept and how to object, plus a one click opt out.
Data handling follows the same discipline. The lawful basis is logged per candidate and per source, sourced people who never answer are deleted after 30 days by default (you can set 30 to 180), and any candidate can be exported or erased in one click. Plan limits also act as guardrails. Monthly allowances for analyses, contact reveals, emails and tests cap how far the agent can go, and consumption is counted honestly, so deleting a role never resets the meter. More detail on GDPR compliant candidate sourcing explains each control.
A week with the agent on one role
Picture a startup opening a senior backend engineer role on Monday morning. The founder spends ten minutes on intake, setting a remote within US time zones requirement, a salary band, three must haves and two deal breakers, then edits the generated rubric to weight production Go experience more heavily. By Monday evening the agent has analyzed several hundred profiles and imported records and selected the strongest matches with written reasons for each score.
Outreach starts Tuesday within the sending window. Over the next days replies come in, questions about the stack get answered from the role facts, and interested engineers complete a short screening and a role specific skills test. In a typical example run this looks like 500 found, 80 selected, 25 replies and 5 on the shortlist, though every role behaves differently. By Friday the founder opens one link, reads the evidence, approves four candidates and rejects one, and the approved people pick interview slots directly. The page on how AI recruiting works shows the full sequence.
When an agent is the right choice
The agent pays off when the work is repetitive and the volume is real: agencies juggling many client roles, startups hiring a few people every quarter without a recruiter, and hiring managers who want screened candidates instead of a pile of CVs. It works best for knowledge work roles such as engineering, data, product, design, sales, marketing, customer success, operations, finance, HR and support, where public professional profiles and written CVs carry enough signal to score against a rubric.
It is a weaker fit for roles hired almost entirely through referrals or for very senior searches where every conversation is a negotiation from the first message. If you want to compare the agent with assembling separate products, the AI recruiting platform page covers the single pipeline approach, and AI recruiter pricing lists the monthly limits per plan. Plans start at 299 USD per month, or 149 USD per month billed yearly, with no hiring fees.
Questions buyers ask
Anything else goes to [email protected] or the help panel inside your dashboard.
What is the difference between an AI recruiting agent and an AI recruiting tool?
A tool produces an output and waits, for example a ranked list or a drafted email, and a person has to take the next step. An agent carries the candidate from one stage to the next by itself, inside rules you set, and keeps a log of each action. You still approve the shortlist, but you no longer push every candidate through every stage by hand.
Can the agent reject candidates without me?
It can decide that a sourced profile does not meet a must have and therefore not contact that person, and it stops writing to anyone who declines or opts out. It does not make final hiring decisions. Every shortlist requires a human approval, and post interview rejections are made by your team. This keeps the process fair, reviewable and consistent with Article 22 of the GDPR.
Will candidates know they are talking to an agent?
Messages go out from your own mailbox under a signature you control, and the Article 14 notice in the first message explains who is processing the data and why. We recommend an honest signature that mentions your team uses an AI assistant for scheduling and questions. Any candidate can reach a person, because you can take over a thread at any time.
What happens if a candidate asks something the agent does not know?
The agent only answers from the role facts you entered at intake, such as salary range, location, seniority and interview process. When a question falls outside those facts, it replies that a colleague will confirm and moves the conversation to your inbox with a flag. It never guesses about compensation, benefits or visa support.
How do I stop the agent if something looks wrong?
Pause the role and every scheduled message for that role stops immediately. You can also pause a single candidate, edit a reply classification, rewrite a sequence step or take over a conversation. Changes apply to future actions right away, and the activity log shows what was already sent so you can follow up personally if needed.
Does the agent need a long setup before it starts?
No. You connect a mailbox and a calendar, fill in the role intake and review the generated job description and rubric. Most of that is a few minutes of reading and editing. Once you launch, sourcing and scoring start right away and outreach follows inside the sending window.
Keep reading
Give it a role and see the shortlist it builds
Type a job title on the homepage. The demo runs instantly without an account and shows an example pipeline, job post and shortlist for that role.