Hey everyone, Neil here. You're reading High-Signal Hiring. Hiring systems from 20+ years of global recruitment experience and 500+ technical hires. Zero noise and instantly actionable.
This week is about the standard you set in your own head before a single person applies. You rarely write it down, but every CV that lands gets measured against it, and that's where a lot of good hires get lost.
Somewhere in the last two years the bar for a "good" engineer drifted into fantasy. Founders have started holding out for someone who ships several times faster with AI, and turning down solid engineers who don't put on that show in a one-hour interview.
You'll learn where that bar came from, why the 10x-with-AI engineer is mostly a story you've been sold, and how to reset the standard to something that predicts a good hire rather than a good performer.
Not a subscriber yet? Sign up here
| Where the bar drifted
Start with your inbox. The average engineering role now pulls in more than 300 applications, roughly three times what the same role attracted in 2021, and candidate pools have grown around 340% since AI sourcing tools arrived. A large share of those CVs were written or cleaned up by a model. Resume Genius put the number of job seekers using AI on their applications at 38% this year, so the tidy, keyword-perfect CV in front of you is worth very little as a signal.
When you're staring at a pile that size, something understandable happens. You quietly raise the bar to cut it down. Not on purpose, but the standard creeps up until only the most dazzling few get through, and "dazzling" in 2026 means someone who talks fluently about agents, ships a side project over a weekend, and makes the whole thing look effortless. At that point the bar has stopped describing the job you're hiring for. It's describing whoever wins the cull.
| The 10x-with-AI engineer is “mostly” a story
The archetype you're now filtering for doesn't hold up under much scrutiny.
METR ran a randomised controlled trial in 2025 that every hiring founder should know about. They took 16 experienced developers, gave them real tasks on codebases they knew well, and let them use AI tools on half the tasks and not the other half. The developers using AI were 19% slower. What makes it worth your time is the gap in perception. Beforehand they expected AI to speed them up by 24%, and even after living through the slowdown they still believed it had made them 20% faster. (the full study is here, and it's worth your time.)
That was a hard setting for AI, mature repositories the developers already knew inside out, and the tooling has moved on since. So don't read this as "AI doesn't help." METR's own follow-up in 2026 surveyed 349 technical workers and landed on a self-reported uplift of roughly 1.4 to 2x, while openly cautioning there are reasons to doubt even that number. Put the two together and the honest picture is a real but modest and uneven gain, one that people are bad at measuring in themselves. Nobody in that data is 10x. And the person you're interviewing has no better way of proving the number than the developers who were wrong by forty points about their own work.
| What you reject when you filter for the show
When the unspoken test becomes "impress me with AI," you start selecting for the wrong thing. You reward the candidate who confidently demos an agent, and you cool on the one who says "I'd be careful there, I've watched these models produce very convincing nonsense." The second person is usually the better hire. I made this case back in Issue 14, that your strongest engineer often isn't the flashiest coder, and again in Issue 15, that a bit of scepticism about AI is a feature, not a gap on the CV. A funnel tuned for dazzle filters both of them straight out.
That's the real cost, and it's invisible on a dashboard. It isn't the bad hire you dodged. It's the good hire you never called back, sat in a pile of 300, passed over because they didn't perform on cue.
| This is not an argument for lowering the bar
I want to be clear here, because in Issue 16 I argued the opposite. Pay more, hire fewer, don't settle. I stand by all of it. A high bar is correct. The mistake is aiming a high bar at the wrong target. Filtering hard for AI theatre while judgment, ownership and taste go completely untested isn't a high standard. It only feels like one. Raise the bar on the things that predict good work, and drop it on the things that predict a good demo.
| How to reset it
Before you open the role, write down three or four things a great hire would get done in their first 90 days. Real outcomes tied to your product, not "10x output." That short list is your bar. Interview against it and nothing else.
Then test the one thing that matters. Hand the candidate a piece of AI-generated code with a subtle flaw buried in it and ask them to walk you through what they'd change and why. You'll learn more in those ten minutes than in an hour of tool talk. The engineer who catches the problem and explains their reasoning is showing you judgment, which is the exact thing the model can't hand them on the job.
And stop treating fluency with AI tools as the headline qualification. It's table stakes now, closer to knowing Git than to a superpower, and no reason on its own to hire anyone.
The bar isn't too high. It's pointed at a person who doesn't really exist. Move it back onto judgment and you'll stop turning away the engineers who would have been your best hire.
Cheers
Neil