What AI interviews are actually good at, and what they are not

We make an AI that runs interviews. So take this with whatever pinch of salt you like.

But I would rather tell you where it does not work than have you find out in month two.

What it is genuinely good at

Asking everyone the same thing. A human at 6pm is a different interviewer than a human at 9am. The AI is not. That consistency is the actual product, more than the AI part.

Volume. 200 conversations at once, at 11pm, in six languages. No calendar involved. This is the bit that feels like magic and is honestly just computers being computers.

Not being someone's boss. People say things to a neutral interviewer that they will never say to their manager. This is underrated and it is most of why exit interviews and churn calls work better this way.

Following up. Forms cannot. Humans forget. The AI always asks the second question.

Writing it all down. Every word, same format, comparable. Nobody is reconstructing a call from memory three days later.

What it is bad at

Here is the honest half.

Reading a room. It does not know that someone is upset, or joking, or being sarcastic about their old employer. It reads the words. That is all it reads.

Judging craft. It can tell you whether someone explained their work clearly. It cannot tell you if the work was good. Those are different things and confusing them is expensive.

Anything high stakes on its own. A score should decide who you talk to. A person should decide who you hire, fire, or refund. If you skip that, you deserve what happens next, honestly.

Being the only voice. If nobody ever reads a transcript, you are not doing research, you are collecting numbers with extra steps.

Where you should not use it at all

• Anything where the person is in distress
• Performance conversations with your own team, they will hate it and they will be right
• Anything you are not going to tell people is AI. Just tell them. Nobody minds, everyone minds finding out later

The thing that decides if it works

Not the model. The criteria.

If you write "communication skills" as your scoring criterion, you will get useless scores, from any AI, forever. If you write "explained a technical decision so a non technical person could follow it", you get something you can act on.

The AI is a very fast, very patient interviewer with no ego. It is not a judge. You are still the judge, you just got 200 conversations of your life back.

That is the trade. Seemed worth building.

Free with your own AI key if you want to try it.