AI Hiring Isn't Colder Than the Process It Replaced

The reflex when you hear that Orbio just raised $21M to run hiring and onboarding for frontline workers is to wince on the candidate’s behalf. A bot screens you, a bot onboards you, a bot probably runs your exit interview on the way out. The instinct is correct about the discomfort and wrong about the comparison. The question worth asking isn’t whether an AI hiring process feels cold. It’s what a fair one owes the person on the other side of it — because the human process it replaces stopped feeling warm a long time ago.

What Orbio actually built

Orbio’s Series A, led by Dawn Capital and announced June 14, brings the company to roughly $26–28M raised including a seed round last September. The founders are not first-timers: Sergi Bastardas spent a decade at Amazon and co-founded Colvin, Nacho Travesí co-founded the benefits platform Cobee, Antonio Melé co-founded Nucoro. Each has at least one exit behind them. That matters, because the thing they built is more opinionated than a typical hiring chatbot.

The product is three agents that run over WhatsApp, voice, and SMS — no app, no corporate email. María handles recruiting: she writes the job description, posts it, screens an applicant by automated phone or WhatsApp interview within about two minutes of them applying, ranks the qualified ones, and books them with a hiring manager. She works in 60+ languages. The human in the loop is the manager making the final call on a ranked shortlist. Daniel runs onboarding. Claire handles retention and exit interviews, and the part worth noting is that the exit signal feeds back into María’s screening criteria. The system is designed to learn what kind of hire stays.

The market is the reason this exists. Orbio is aiming at the 2.7 billion deskless workers in sectors — fast food, healthcare staffing, logistics, retail, security — where annual turnover runs past 70% and roughly 80% of the workforce has no corporate email address. This is hiring at a volume and churn rate that breaks normal recruiting infrastructure.

A note on the numbers: Orbio’s own figures (up to 80% faster time-to-hire, 65% lower cost-per-hire, candidate-satisfaction scores in the high 90s) are marketing, not audited results. The platform reports 2M+ interviews and 100+ clients across 15 markets, which is real scale. Treat the outcome percentages as claims until someone independent checks them.

Two true things that look contradictory

Here’s the tension the whole debate hangs on.

Ask people in the abstract and they hate it. Pew found 66% of US adults would not even apply to an employer that used AI in hiring, and 71% oppose letting AI make the final call. Gartner put applicant trust in AI fairness at 26%. Greenhouse’s late-2025 survey found just 8% of job seekers consider AI hiring “fair” — against 70% of hiring managers who trust it to make faster, better decisions. That is the widest trust gap anyone has measured, and it’s getting more concrete: by May 2026, 63% of job seekers had faced an AI interview, 70% weren’t told AI was evaluating them, and 38% withdrew from a process specifically because it used one.

Now put a real job in front of them. A field experiment out of Chicago Booth and Erasmus University ran 70,884 actual entry-level customer-service applicants in the Philippines. Given the choice between an AI interviewer and a human one, 80% picked the AI. The AI-screened candidates ended up with 12% higher offer rates, were 18% more likely to start, and 17% more likely to still be employed at 30 days.

Both results are real. The gap between them is context. The Pew question is hypothetical — would you, in principle, accept being judged by a machine. The Booth experiment is a person who needs a job this week, weighing an AI that responds now against a human screen that might come in days, if it comes at all. When that’s the real alternative, the machine wins. Not because people love it. Because the human option is worse than we pretend.

Stop romanticizing the process being replaced

The argument against AI hiring quietly assumes the status quo is humane. For high-volume frontline roles, it isn’t. About 75% of applications get no response at all. In Greenhouse’s data, 61% of candidates were ghosted after an interview, and 80% of hiring managers admit to ghosting candidates themselves. Median time from application to offer has stretched to 68.5 days. At peak intake, recruiter-to-applicant ratios hit 1:300 and beyond.

You cannot give 300 people a warm, attentive, individually-considered interview. Nobody is. The “human touch” being defended is, for most frontline applicants, a form that vanishes into a black hole. An AI screener has no feelings. A recruiter holding 300 open reqs has no bandwidth, which from the applicant’s side of the table is the same thing, except the AI at least replies.

So the choice in front of these employers was never warm-human versus cold-machine. It was no-response versus a response. For the candidate, fast and consistent and multilingual can be a genuine improvement: the Booth applicants reached an interview in a third of a day instead of half a day, could screen at 9pm after a shift, and could do it in their own language rather than getting filtered out by an English-only phone screen. For the 80% with no corporate email, WhatsApp-first hiring is access, not just efficiency.

The part that should actually scare you

None of that means the technology is safe. The failure history is specific and ugly, and it’s not about hurt feelings — it’s about discrimination at scale.

Amazon built a resume screener and scrapped it in 2018 after it taught itself to penalize the word “women’s.” HireVue dropped its facial-analysis scoring in 2021 after a federal complaint. iTutorGroup paid $365,000 in the EEOC’s first AI hiring settlement because its software auto-rejected women over 55 and men over 60 — not a subtle statistical drift, a hard-coded age cutoff. Mobley v. Workday is now a conditionally-certified nationwide collective action covering potentially millions of applicants over 40; in June 2026 a court let Workday keep its internal bias-testing data privileged. A University of Washington audit fed 554 resumes through production language models and found they preferred white-associated names 85.1% of the time, and disadvantaged Black male candidates in 100% of direct comparisons against white male ones.

And the human-review safety valve everyone reaches for is weaker than it sounds. A separate UW study found that people working alongside a moderately biased hiring AI tend to mirror its bias rather than correct it. A human rubber-stamping ranked output isn’t oversight. It’s laundering.

This is the real reason candidate distrust is rational. Not because AI lacks empathy — because it can encode discrimination behind a clean interface and an audit trail that says everything was consistent. Consistent and fair are not the same word.

What to demand on Monday morning

If you’re an engineering or ops leader buying one of these systems — and given the funding flowing in, a lot of you will be — the useful move is to stop arguing about whether AI hiring is good and start writing requirements. Regulators and researchers have already converged on four, and the Greenhouse data shows how rarely they’re met:

1. Disclosure. Tell candidates an AI is evaluating them, before it does. Only about 18% of employers currently do. This is the cheapest thing on the list and the most commonly skipped.
2. Explained criteria. Say what the system measures. 39% of candidates want this before they’ll proceed, and “we can’t tell you” is a defensible answer only if you don’t actually know — which is its own problem.
3. A human escape hatch. A real path to a human interviewer or human review of the AI’s recommendation, not a checkbox. 46% of candidates want it; the EU AI Act will require human oversight on final hiring decisions once high-risk obligations bite in August 2026. Build it now.
4. Published bias audits. Independent adverse-impact testing, results made public. NYC’s Local Law 144 already mandates this for tools used there, though a December 2025 state audit found enforcement near-toothless — which means the audit is on you, not the regulator.

Treat the vendor itself with the same scrutiny you’d apply to any system that touches sensitive data: ask for its testing methodology and its last audit before you connect it, the same way a vendor onboarding process gates every other dependency.

Notice none of these is “keep a human in the loop for warmth.” They’re controls. The thing worth protecting about frontline hiring was never the feeling. It was the fairness — and that was already failing under the human process. AI can make it worse, quietly and at scale, or it can make it faster and more auditable than a recruiter’s gut ever was. Which one you get is decided by the requirements you write, not by the technology you buy. Write them down before you sign.