AI is applying. AI is screening. Now what?

Notes from our first Talent Crunch all hands.

Our first all hands was meant to be a friendly catch-up, and it turned into a long conversation about AI in hiring. One of our members got us going: he's built an agent that applies to jobs for him every morning, sent out hundreds of applications, and landed an offer. So our discussion question went: “if one person can game the system this easily, how is TA meant to hold the other end?”

Thirty-odd of us spent half an hour on it. Company and tool names are left out so nobody gets thrown under a bus (nor boosted for no reason)

Start with the real problem, not the tool

A few people slowed us down before anyone reached for a solution, and they were right to. Candidates will use AI at work every day once you hire them, so rejecting them for using it to apply, then handing them the same tools on day one, doesn't hold up. It's like banning a calculator in the interview for a job that runs on calculators.

Their point was to get clear on what's bothering you before you fix anything. Is it wasted time, budget, hiring the wrong person, or speed? 

What people are trying - #communitywisdom 🧠

Most of the ideas shared one instinct: move the filter earlier, or make the human bit count for more.

Push the filter into the application form:

• Case-study questions that need a real, specific example, like, I dunno, “how you handled a particular piece of employment law in a live situation”. AI can pad an answer, but it can't invent the details of a case someone never worked on.

• A short voice memo instead of a cover letter, two or three minutes on one question. Fewer people applied, and the ones who did were stronger.

• A quick skills test before the first interview, including how the candidate uses AI. One member cut their pool by around 40% before a human read a word.

Make the human screen stronger:

• A simple sanity-check script for the first call, built with senior engineers (or any professionals - dependent on what you hire for). Low tech, but sharp enough to tell a real practitioner from a tidy CV. More people made it through to later rounds after this was implemented.

• A quick call (or even an email) to check one or two claims when something feels off.

Use the ATS without hiding behind it:

• Flag CVs that read as machine-written, then send the flag to a recruiter rather than to an automatic rejection.

• If someone used AI to clean up their English and the experience underneath is real, let them through. Rejecting on an AI flag alone is unfair, and in plenty of places it's legally risky.

Set the rules up front:

• Tell candidates in writing how they're allowed to use AI in your process before they apply. It helps the honest ones who want to use it and are scared it'll cost them, and it gives you a clear line when someone ignores it.

For the job seekers reading this

If you're applying right now, one thing came up that's easy to misread.

Most automatic rejections aren't an AI reading your CV and deciding you're “not good enough”. Often it's a mandatory question / filter you didn't match. If a role needs five days in the office in Berlin and you said you can't, the system matches that and rejects you before anything else happens. It stings, but it's usually just logistics, so don't read a verdict on your worth, nor assume “it’s been AI rejecting me”.

What recruiters need to own

The AI does exactly what we tell it, including the daft parts. ‘
Someone shared a story: the business asked for candidates from a handful of elite universities, the AI obliged, and when someone finally checked the rejection pile it was full of brilliant people whose only “crime” was the wrong logo on their degree. The tool did its job, and we're the ones who gave it a bad brief.

Treat your ATS AI like a keen new starter in their first week. 

It'll follow your instructions to the letter and never question them, so if you tell it to only trust a 95% skills match or a top-tier university, it'll bin your best candidate and feel great about it. That's automation bias, and it's the thing that catches you out.

If you get to train these tools, take the prompting seriously. For example, fluency in a language shows up in a dozen ways depending on where someone grew up, and it's your job to teach the tool what to look for rather than let it guess and trust its own guess.

And every recruiter who works with an ATS with AI screening capabilities (for CVs), has the responsibility to write proper instructions (prompts) to teach the AI how to exactly check the CVs.

Don't over-filter

Every filter you add has a cost. 

Stack five AI checks and a strong candidate with a slightly messy application will wander off to a competitor who kept two sensible filters and a human at the end.

One member keeps the volume sane by working in batches: take 30 applications, close the intake, review them properly, see what your filters got right and wrong, then open the door again. It's slower on paper and it gets you better hires.

Where we landed

We didn't solve it, and nobody has. It sits across the whole chain at once: candidates gaming the filters, TA building filters to catch them, leaders setting policy, and vendors selling all of us tools to fight each other. It's AI against AI, with a human in the middle still hoping to hire a real person. The best we can do is keep comparing notes honestly, which is what this community is for.

A few things from the call to follow:

• The member who built the job-application agent may share how he did it on Slack. If you're hiring or job hunting, keep an eye out.

• We might run a workshop on this, one side for employers, one for candidates. Tell me if you'd come.

• The blog will keep building on these “all hands” conversations. This is the first.

🫶🏻 And the usual asks, because I'm one person doing the work of five:

• Get on the newsletter if you're not already: tinyurl.com/tcupdates26.

• Come say hello on Slack - apply here

• Help me reach 10,000 humans (not bots) in the talent and people space on LinkedIn this year. Bring the people who see things differently to you