AI can judge a person in minutes. Here’s where it should slow down.

AI can now judge a person in minutes. That’s exactly why it needs to slow down in the right places.

More and more AI hiring tools are being built fast with AI app builders. I get the appeal. Shipping fast is exciting, and a lot of the ideas are good. But if your product scores humans, your interface is part of the decision, and most of the ones I look at ask for blind faith.

The real clickable prototype, recorded on a phone. Talkround is a concept redesign with a fictional brand, and every name, answer and score is sample data. No sound.

The page I started from

I recently tore down a real, live AI interview tool, built with an AI app builder. Candidates talk to an AI, and recruiters get a ranked shortlist. It’s a good idea. But the page selling it shows an illustrated AI head, “fit” percentages nobody can explain, and face and emotion detection sold as a feature. It never says whether a person makes the final call.

I wrote up the full teardown with all eight fixes. This post is about the one that matters most for any AI that judges people. Trust has to be designed into the flow. You can’t bolt it on in the footer.

Same idea, a different promise

So I redesigned it as Talkround, a concept with a made-up name, and built both sides as a clickable prototype. Here’s what the video walks through.

The candidate knows the deal up front. The invite says it takes about 15 minutes, that everyone gets the same four questions, that a person at the company decides, and that they’ll hear back within five working days. It also says the AI never sees their face and doesn’t judge their accent or tone of voice.

They choose how to answer. They can speak or type, and neither option scores differently. There’s no countdown, and they can edit their answers before they submit.

The score shows its evidence. On the recruiter’s side, the answer comes back as a score report. Every rubric level quotes what the candidate actually said and links to the full transcript. There’s no mystery number.

A person makes the call. The AI suggests, and a recruiter decides. Borderline or flagged interviews go to a person with no suggestion until someone has looked. A decline can’t be sent until a person confirms they’ve read the transcript.

Why this matters if you’re a founder

If you can’t explain a score, you can’t defend it, whether it’s to a candidate, a hiring manager or a buyer doing due diligence. In Europe, the EU AI Act treats AI used to screen job candidates as high-risk, so “who decides?” is a question you will get. I’m not a lawyer, so I treat it as a design problem: make the answer visible.

The most important screen in an AI product is often the one where a human takes over. Design it on purpose, and show it on your landing page.

Here’s a quick test for your own product:

Can a user see why the AI reached its answer?
Is it clear who makes the final decision?
If someone needs a different way to take part, like typing instead of speaking, does it count against them?
Does your page show the product, or just say it’s smart?

Try it yourself

Take the interview in the clickable prototype, then switch to the recruiter’s side and find your own score report.
Read the design decisions in the Talkround case study.
Scroll the redesigned landing page. It’s best on a laptop.

AI will keep getting faster at judging people. The products that win will be the ones people are willing to be judged by.

Building an AI product people need to trust?

Book a free 20-minute call, show me what you’ve built, and I’ll tell you honestly what I’d fix first.