Industry
AI employees for recruiting agencies
Tobiloba Odejinmi · 14 Apr 2026 · 6 min · 938 words
Direct answer
An AI employee for a recruiting agency reads the pile and returns a shortlist a person still owns. It does not quietly reject people. NYC Local Law 144 wants a bias audit and notice when you use an automated employment decision tool. The EU AI Act treats employment AI as high-risk. The EEOC does not give you an exemption because a model did the screen. I build Sourzer to cut tab-switching, not to hide a decision.
- Shortlist. Do not auto-reject without a named recruiter on the outcome.
- NYC LL144: bias audit, candidate notice, publish what the law requires.
- EU AI Act: employment systems are high-risk. Oversight is not optional.
- The EEOC still sees your name on the hire and the pass-over. The vendor does not absorb that.
The pile of CVs is the job
Agencies do not fail because they lack a philosophy of talent. They fail because someone is reading the same stack in five tabs and the good file is on page twelve. That is the work I will take off a desk.
Sourzer exists to find and rank without the tab bounce. A shortlist is a product. A silent reject is a legal event. I keep those apart on purpose.
NYC Local Law 144 is not a footnote
If you use an automated employment decision tool on candidates in New York City, LL144 expects an independent bias audit, publication of the summary, and notice to candidates. You do not get to call it 'just software' and skip that.
I will not build you a screen that cannot survive those questions. If you are in that market, budget for the audit and the notice before you budget for speed. Speed without the paperwork is not speed. It is a pause waiting for a letter.
High-risk means you already needed a human
The EU AI Act puts AI used in recruitment and worker evaluation in the high-risk class. That is not a vibe. It is a list: risk management, data you can defend, transparency, human oversight, record-keeping.
Those are the same controls I want even when the Act is not in the room. A recruiter who can override. A log of why someone was ranked. A way to turn the tool off and still hire.
- Named recruiter on every outcome the candidate will feel.
- Reasons a stranger can read, not a hidden score.
- Bias audit where the law requires one, and honesty where it does not yet.
- A switch. Hiring cannot depend on a vendor outage.
The EEOC does not care about your demo
US equal employment rules did not pause for language models. If your screen filters people in a way that has disparate impact, the employer is still the employer. There is no exemption for 'we used AI'.
So I do not sell auto-reject as a feature. I sell a first pass with reasons, and a person who can say they read the edge cases. If a shop wants a machine that says no in the dark, they can find a different builder.
Rank is not a verdict
Ranking helps a tired recruiter start in a better place. It becomes a verdict the moment nobody looks below the cut line and you cannot say why the cut existed.
Write the rule down. Review it. Sample the rejects. If you cannot defend the sample, you do not have a ranking tool. You have a preference you have not admitted.
One desk, one process
Start with inbound CV screening for one role family. Map how the agency already scores. Plug into the ATS they already pay for. Hand over a shortlist workflow with logs.
Measure time-to-shortlist and whether the recruiter still changes the order. If they never change it, you do not have oversight. You have a rubber stamp. Fix that before you roll it to the next role.
What I tell an agency that wants 'full auto'
I tell them no. Not because the model cannot rank. Because the law, and basic decency, still wants a person on the outcome a candidate will feel. NYC will ask for an audit. Europe will ask for oversight. The EEOC will still see your name on the pass-over.
Take the first-pass read. Keep the reject with a recruiter. Publish what you have to publish. Sample the files below the cut. That is a company you can still explain in a room. The other thing is a lawsuit with a product page.
Questions people ask
Does NYC LL144 apply if we only 'assist' recruiters?
If the tool substantially assists or replaces discretionary decision-making on candidates in New York City, treat it as an AEDT and get counsel. 'We just rank' is not a magic phrase. Notice and a bias audit are the starting point, not a blog post.
Is employment AI high-risk under the EU AI Act?
Yes. Recruitment and worker-evaluation systems sit in the high-risk bucket. That means risk management, data governance, transparency, human oversight, and logs. Build those in before you celebrate a faster screen.
Does the EEOC care that a vendor built the model?
No exemption. If the screen has a disparate impact, the employer still owns it. 'The AI did it' is not a defense I would take into a room.
What does Sourzer actually do?
It helps recruiters find and rank candidates without bouncing between five tabs. The recruiter still makes the call. The point is the pile, not a hidden veto.
What should an agency automate first?
The first-pass read of CVs you already know how to score by hand. Output a shortlist and the reasons. Keep a person on every reject that the candidate will feel.
Written by
Tobiloba Odejinmi
Head of Engineering at 10mg Health. I have run engineering at Zeeh Africa and sold Insurpass and Shopl. I still write the code. If you have one process that still runs on people copying things, we can look at it in thirty minutes.
