AI screening AI applications. The arms race nobody asked for.

The 2025 hiring market produced a strange feedback loop. Candidates discovered that AI could write their CVs and cover letters better than they could. Recruiters discovered that AI could screen the resulting flood of applications faster than they could. The two sides scaled their AI use in parallel, and the whole industry is now in an arms race that nobody set out to start.

The arms race has a problem. The signal in the middle is collapsing. When both sides are using machines to communicate, the machines start optimising for each other and the human content starts to disappear.

How we got here.

Applications used to be a slow trickle. A candidate would spend a half-hour writing a cover letter, a few minutes adjusting their CV for the role, maybe an hour total per application. The cost of applying capped the volume.

When that cost dropped to near zero (because an AI can produce a tailored cover letter in 30 seconds and a CV variant in 60), the volume exploded. Recruiters started seeing application counts triple or quintuple on the same job postings. The flow of high-quality, AI-polished applications became indistinguishable from spam, except that some of it was real.

Recruiters responded with their own AI. Screening tools that could process hundreds of CVs in minutes, surface the top matches, filter out the obvious noise. The match quality of these tools is genuinely impressive on a structured-data basis. They reliably surface candidates whose qualifications align with the role.

What they cannot do is distinguish a real candidate from an AI-amplified one. Because the inputs to a screening AI and an application AI both come from the same training distribution, they meet in the middle. The screener thinks the application looks good. The applicant's machine has produced something the screener's machine likes.

What gets lost in the middle.

What is missing is the person. The CV is polished, the cover letter is targeted, the keywords match. None of that tells the recruiter whether the candidate actually wants the role, whether they can communicate clearly, whether they will be the kind of colleague the team needs. The arms race optimises for fit on paper. Fit on paper turns out to be a worse predictor of hiring outcome than it ever was, because both sides are now actively engineering it.

The honest signal in modern hiring is the part that resists being AI-generated. A candidate's voice, their hesitation, their word choice when they are not reading from a script. A 60-second video moment carries all of that and is, currently, beyond the reach of the application-writing AIs. Not because the AIs cannot generate voice (they can, badly), but because the human watching the video can tell when something is not real.

Why a short video moment is the way out.

Three properties of a 60-second video moment make it AI-resistant in the way that text no longer is.

• It is too short to be useful for a generative model to mimic at scale. Generating a believable 60-second video of a specific person answering a specific question is technically possible and operationally too expensive to be worth doing across an entire application funnel.
• It carries non-textual signal (tone, hesitation, eye-contact) that is inaccessible to a screening AI scoring text. The signal goes straight from candidate to recruiter, with no machine layer to flatten it.
• It is too short to be over-rehearsed. A candidate cannot script and memorise a polished 60-second answer to an unknown question and still come across as natural. The short cap actively prevents the kind of over-preparation that creates the same arms-race problem text has.

The result is that a 60-second video moment gives the recruiter a piece of evidence that the AI arms race has not touched. The candidate is in front of them. The recruiter watches, the recruiter forms an impression, the recruiter decides. The decision is informed by the CV (which has been mediated) and by the video moment (which has not).

Where this is heading.

The arms race is not slowing down. Generative models keep getting better, screening AIs keep getting more sophisticated, the polishing on both sides keeps getting more aggressive. The asymmetric weight of a short, human, real moment is only going to grow.

This is what bring authenticity back to hiring actually means. Not the absence of technology in the funnel, but a deliberate place in the funnel where the human shows up directly. Everything else can stay automated. The 60-second moment is what is left when the automation has done its job.

Flipbase exists because the arms race is real and because there is one place in the hiring flow where a human moment beats a machine one. We built that place. The rest of the funnel can stay as automated as you want.