An AI sourcing agent builds a candidate pipeline overnight by working the role in rounds: it turns your job description into criteria, reads profiles against every one of them, finds where its shortlist is thin, and searches again for exactly those gaps. By morning you have a ranked shortlist with the evidence for each person, and a few decisions from you steer the next round.
That is the short version. The longer version is worth knowing, because the difference between a sourcing agent that saves your week and one that hands you a confident list of the wrong people lives in the details: what it reads, how it scores, and what it does with your feedback.
This post walks through how the Sourcer in the 100Networks Workspace works a single role through the night, using one example throughout.
What does an AI sourcing agent actually do?
Strip away the interface and a sourcing agent does five things a good human sourcer does, at a volume no human can:
- Turns the brief into criteria. "Senior backend engineer, payments" becomes specific requirements, each marked as required or preferred.
- Searches several places at once. Professional profiles, a people database, the open web (where public work lives) and your own candidate pool.
- Skims widely, then reads closely. A quick scan across many profiles, then a full read of the most promising ones.
- Checks each person against each requirement. Not one blended score, but a judgement per criterion, with the evidence.
- Ranks, then looks again. It works out which requirements the shortlist covers poorly and runs another round aimed at them.
The last step is the one most tools skip, and it is the one that matters most for hard roles.
Why does sourcing in rounds beat one big search?
A single search answers the question you asked. Rounds answer the question you meant.
Here is the problem with one pass. Say your brief has four must-haves. The first search will over-supply whichever is most common in the market and under-supply the rarest. For a payments backend role in Bengaluru, Java and Go engineers with high-volume experience are plentiful. Engineers who have owned reconciliation, ledgers or settlement in production are not. A one-pass tool hands you forty strong Java engineers and two people who have actually touched money movement.
An agent working in rounds notices this. After round one it reviews its own shortlist criterion by criterion, sees that "production payments experience" is the thin one, and spends round two looking specifically for it: different titles, different companies, different phrasing in profiles. The gaps close because something is checking for them.
A worked example: Senior Backend Engineer, Payments
Here is how an illustrative night goes in the Workspace.
The brief, 9:40 pm. You attach the job description. The Sourcer drafts a brief and shows it to you as rows you can edit:
| Requirement | Type |
|---|---|
| Six or more years building backend services | Required |
| Production payments work: ledgers, reconciliation, settlement or payment gateways | Required |
| Go or Java in production | Required |
| Systems handling high transaction volume | Required |
| Event streaming experience | Preferred |
| Based in or willing to move to Bengaluru | Required |
You change one thing (event streaming is a nice-to-have, not a must-have) and let it start. If you have already told the Workspace something that applies to every search, such as "we never approach people at our own clients", it is applied at the start as a visible note you can undo.
Round one, through the late evening. The Sourcer searches, skims widely and reads the most promising profiles in full. Each person read in full gets a fit in words: Strong fit, Good fit, Partial fit or Poor fit. Anyone it has only skimmed is marked First look, so a quick scan never passes for a judgement.
Round two, in the early hours. The shortlist is heavy on generalist backend engineers and light on payments. The next round goes after that gap specifically.
Morning. A ranked shortlist is waiting, grouped by round, with outreach drafted for you to review. Nothing has been sent.
What does "evidence per criterion" look like?
This is where you can tell a serious sourcing agent from a keyword filter with a chat window.
For each person, the Workspace shows each requirement and what met it:
- In their words. Evidence quoted from the person's own role or project descriptions, for example "built the reconciliation service for card and UPI settlements". This is the strongest kind.
- Inferred. Something the agent read between the lines, labelled as such and never presented as proof. Working at a payments company for three years suggests payments experience; it does not prove it.
- Partly shown or not shown. The profile only partly speaks to the requirement, or not at all. A Partial fit tells you which requirement is missing or unproven, so you know what to ask before you call.
The Workspace deliberately says fit in words and never shows a match score. The score still exists (it orders the list), but a number invites the wrong argument. "Why is this person a 78?" goes nowhere. "The agent inferred payments experience from the employer; is that enough?" is a conversation your hiring manager can have in thirty seconds.
Why decide on a few people before the next round?
Because the brief is never the whole story, and your decisions are the clearest signal the agent will get.
In the Workspace you mark people Shortlist, Maybe or Not a fit. Those decisions are yours alone; the agent never makes them. Once you have decided on five people in a round, you can start the next one, and it looks for more people like the ones you shortlisted.
A few things make that feedback sharper:
- Give a reason for "Not a fit". "Payments, but only on the merchant integration side" teaches the search something no brief would have said.
- Comment on a person. "More like this one" on a single profile becomes guidance the next round reads.
- Tell the Sourcer what to change. In plain words, at any time. Each change is kept as a version you can undo, and the next round goes back to how it was if you do.
- Answer its questions. When the brief is ambiguous, the Sourcer asks rather than guesses, and your answer carries forward.
Five decisions early in the night beat a perfect job description. They also catch a bad brief fast: if you rule out the first five for five different reasons, the problem is the brief, not the market. We wrote about that pattern in why you can't hire good engineers.
What else should you expect from the morning view?
A few things we found teams use daily once the shortlist exists:
- Side by side. Your shortlist and the best people still to review, one row each, against every requirement, with the line each verdict came from.
- Share with the hiring manager. Send the person who decides a link to the shortlist, so their view reaches you first-hand rather than through a forwarded spreadsheet.
- Wrap up now. If round one is already enough, tell the Sourcer to finish; it ends after the step it is on, runs a final review and hands over.
- Exclusions. Hide people who already appeared in an earlier search or are already in a sequence, so nobody is approached twice.
- Run it again on a schedule. A role that stays open can be searched again weekly, monthly or quarterly, with each run arriving as a new round.
What an AI sourcing agent cannot do
It is worth being plain about the limits.
Found is not reachable. A strong profile with no way to contact them, or a person who never accepts connection requests, is still a strong profile you cannot use. Outreach is a separate job.
A brief cannot be better than the conversation behind it. If the hiring manager has not decided what the role is, the agent will find, accurately, people for a role nobody has decided on.
It does not convince anyone. 100Networks reports that 72% of the people it places were not looking for a job when they were found. An agent can find those people. Persuading a senior engineer to leave a job they like is human work, and for the hardest roles that is where a specialist recruiter earns their fee.
How this works in the 100Networks Workspace
The Sourcer is one agent on an AI hiring team (see what an AI hiring team is). You attach a job description, check the brief it drafts, and let it run; the shortlist arrives in rounds, each person with their fit in words and the evidence against your brief. You decide, and the next round learns from you. When you are ready, Outreach drafts messages to the people you shortlisted, held until your team switches them on.
The Workspace is for companies' in-house teams. Access is by request, and pricing is set with each team, with no per-hire fee.
See how the Sourcer and the other agents work on the 100Networks Workspace page.