The hiring industry is now being super exposed to AI agents who now find candidates, write the outreach, follow up and initiate the first screening calls without a human touching it.
We looked at what AI and agentic AI in hiring has actually changed. Sadly and to no surprise to us, the answer is - not much. The funnel is the same. The only difference is who runs it.
What "agentic hiring" actually is
The newest tools are autonomous sourcing agents, and the money is flowing in. One well funded vendor has just raised US$60M to take its agent to general availability. The pitch is that recruiting capacity no longer needs to grow with recruiting headcount.
That is a real efficiency gain. But look at what the agent is asked to do. A person writes the criteria. The agent finds people who match them. Those criteria usually come from past hires, job ads and resumes.
So the whole system rests on a self-assessed input, the resume, and a human-written definition of "good." The AI makes it faster. It does not make it fairer.
The bias is measured, not assumed
Researchers at the University of Washington took more than 550 real resumes and ran them through three open-source language models. They changed only the first names. Experience, education and skills stayed identical. Across about 3 million comparisons:
White-associated names were preferred 85% of the time. Black-associated names were preferred 9% of the time.
Male-associated names were preferred 52% of the time. Female-associated names were preferred 11% of the time.
The models never preferred a Black male name over a white male name.
A 2025 follow-up found something worse. When people reviewed candidates alongside a biased AI recommendation, they copied the bias. The human in the loop did not fix the problem. They absorbed it.
These were name swaps on specific open-source models, not a test of every commercial product. But the pattern matches decades of human audit studies. The models learned from us.
The Australian evidence points the same way
This is not only a US problem. According to the Responsible AI Index, 62% of Australian organisations used AI in recruitment moderately or extensively in 2024.
Dr Natalie Sheard at the University of Melbourne ran the first study of AI hiring systems used by Australian employers. She interviewed recruiters, AI experts, developers and career coaches, and analysed materials from two major vendors operating here. Her finding: these systems can enable, reinforce and amplify discrimination against historically marginalised groups. The mechanisms are specific:
Accent and language. Systems judge communication skills by how well someone speaks standard English. Non-native speakers score lower, even when they can do the job.
Time limits. Timed video answers disadvantage people who speak English as an additional language.
Career gaps. Time out for caring acts as a proxy for gender.
Disability. Personality tests can screen out people with conditions like depression, even when they meet every essential requirement.
Dr Sheard calls it algorithm-facilitated discrimination. She says our discrimination laws need review, and that there is merit in calls to pause these systems until proper safeguards exist. The tools are already in use. The protections are not.
The noise problem is compounding
Candidates now use AI to apply. Employers use AI to filter. Each side is arming itself against the other.
LinkedIn reports around 11,000 applications per minute globally, up more than 45% year on year.
77% of hiring teams say they regularly see AI-generated or AI-assisted applications (Willo, 2026).
67% of US HR leaders say reviewing them has slowed hiring, and 65% of hiring managers say AI-polished resumes make skills harder to verify (Robert Half, 2026).
The CEO of a major hiring software company calls it an "AI doom loop." Candidates apply more, employers filter harder, rejected candidates apply even more.
By 2028, 1 in 4 candidate profiles is expected to be fake (Gartner).
Add agentic outreach on the employer side and the loop tightens. More automated messages reach more people, who respond with more automated applications. We are not fixing application noise. We are paying machines to make it at both ends.
Nobody wants to change the system
Changing it would mean admitting the input is broken. It would mean asking whether a resume tells you anything real, whether years of experience predict performance, and whether the criteria we copy from the last hire are just a record of who we hired before.
That is hard, slow and political. Wrapping an AI sticker around the old process is easy, and it sells well. So the process stays, the bias stays, and the rejections arrive faster.
The new hiring standard
You cannot debias a signal that was biased when it was created. If the input is a self-written resume, the best AI in the world is still ranking self-assessment.
The fix is a better input: verified evidence of what people can actually do, for every pathway into work. That is the new hiring standard, and Gradstack, Australia's Workforce Activation Layer, is building it.
Verification also reduces application volume. That is exactly why systems that profit from volume will resist it.
Before you hand your hiring to an agent
Here is a thought to sit with. If an AI agent hires for you, it also speaks for your brand. Every message it sends, every person it ignores and every candidate it never finds is now your reputation at work.
So before you buy, ask three things:
What is it actually reading? If the answer is resumes and a brief copied from your last hire, it is working from a record of the past.
Is it finding the most verified, whole-person fit? Or just the fastest match to a keyword?
Is the way we engage and hire talent inclusive? Who is it reaching, and who never gets a reply?
If you can't answer all three with confidence, pause. Take time to look deeper into agentic hiring first. It is not a quick fix, and it is not automatically a valuable one. The right tool, built on the right signal, can change how you hire. The wrong one will only make the same mistakes at scale.
We shared what we have seen because better hiring helps everyone: employers, candidates and the wider market. We hope it helps you ask better questions.