tldr; The resume was never a measure of competence. It was a proxy that survived because faking it took effort, and AI removed the effort from both sides at once. We now screen with AI and act surprised when candidates apply with AI. That is not an ethics problem, it is a prisoner’s dilemma, and you cannot ethics your way out of one.
“We’re drowning in AI-generated resumes. Every application looks perfect and they all read identical.”
Ten minutes later. Same meeting. “Okay, next item. The demo of the new AI screening tool.”
Nobody laughed. Nobody was supposed to.
That meeting is happening everywhere
I want to be careful here, because it would be easy to write this as a story about dumb decisions, and it isn’t one. Everyone in that meeting is behaving rationally. Hold that thought, because it’s the whole point.
First, the numbers, because this stopped being a vibe a while ago.
LinkedIn is processing roughly 11,000 job applications per minute, up 45% year over year, per reporting in the New York Times. Recruiters describe a “1,000 applicant threshold” where a single competitive req clears a thousand applications in days. Around 90% of HR managers say their workload has gone up because of the AI application surge.
And here’s the number that should end every “AI is making hiring more efficient” sales call: per SHRM’s benchmarking, time-to-hire and cost-per-hire have both risen for three straight years. Right alongside the rollout of generative AI tooling on both sides of the funnel.
Read that again. Everyone is spending more money to hire more slowly. That is the arms race’s actual output. Not a side effect. The output.
The resume was already dead. AI just called the time.
Here’s the thing: the resume was never a measure of competence. It was a proxy. A lossy, self-reported, unverifiable proxy that survived for one reason only. Faking it took effort.
That friction was the entire security model. Not verification. Not validation. Just the assumption that a person wouldn’t spend forty hours tailoring lies for one application.
AI removed the friction. In both directions, at the same time. Candidates can now generate a keyword-perfect resume for any req in the time it took you to read this paragraph. Employers can generate a plausible-sounding rejection for ten thousand of them before lunch.
So both sides are now optimizing a document that neither side believes.
You cannot out-detect a dead signal. There is no screening tool sophisticated enough to extract truth from an artifact that no longer contains any. The vendors selling you AI to “cut through the noise” are selling noise-cancellation for a room where the signal already left.
We did this to ourselves, and I can prove it with one sentence
Here’s the sentence: We screen candidates with AI, and we are shocked, shocked, when candidates apply with AI.
That’s not an ethics problem. Run the symmetry test on any rule in your hiring process. If a rule doesn’t survive being reversed, it was never a principle. It was just leverage. “AI is fine when we use it to reject you, but cheating when you use it to reach us” fails that test so hard it should come with a laugh track.
And once both sides are armed, the escalation logic takes over. This is a textbook prisoner’s dilemma. Every individual move is rational. The TA leader can’t stop posting where the candidates are. She can’t hand-review a thousand applications with four recruiters. The candidate can’t unilaterally disarm either, because he knows a bot reads his resume before any human does. Each player’s best individual move sustains the worst collective outcome.
And the defining feature of a prisoner’s dilemma is this: you cannot ethics your way out of it. Responsible-use pledges don’t beat incentives. They never have. Which is an awkward thing to say out loud in a year when “Responsible AI” is on every conference agenda. I’ll say it anyway.
The part that actually turns me green
It gets worse, because employers aren’t just trapped in this equilibrium. We fund it.
Every recruiter seat license. Every ATS module. Every screening add-on, every sourcing tool, every “AI-powered talent intelligence platform.” All of it is revenue flowing to an intermediary economy whose business model depends on the resume staying exactly as ambiguous and unverifiable as it is today.
We are paying protection money to the arsonist. And buying our smoke detectors from them too.
(Who profits from the ambiguity, who quietly built the fix and let it die, and what LinkedIn’s own verification badge accidentally confesses: that’s Part 2. Bring a shovel. We’re touring a graveyard.)
The concession, because this newsletter isn’t a whine
If you run TA and you just signed a PO for a screening tool, I’m not calling you a sucker. The volume is real. The four-recruiters-and-a-thousand-apps math is real. The tool genuinely stops your team from drowning this quarter, and this quarter is when your requisitions are due.
That’s what makes it a trap. Traps built out of stupid decisions are easy to escape. Traps built out of rational ones are a different deal entirely.
And one more concession, ahead of Part 2: the research on what actually predicts job performance has been sitting in plain sight for decades. Observed work samples. Structured interviews. The Schmidt-Hunter line of research puts work sample validity miles ahead of resume screening, which doesn’t even crack the top predictors. (Not take-home assignments, before anyone writes in. Those fell to the same tools the resume did. Part 2 deals with that.) The better instrument exists. It’s just slower, and nobody got fired for buying the fast one.
So no, the arms race isn’t a failure of technology. The technology is working great. It’s a failure we’ve collectively decided to subscribe to. Annually. With auto-renew.
Back to that meeting from the top. The person who described the flood and the person who approved the flood-management tool were the same person. That’s not irony.
That’s the system. Working as designed.
— Mike.
Director HR Technology | Battle-Bot referee
Coming soon: the graveyard tour. Everyone who built the fix, including a name you know very well, and why the fix keeps dying.
P.S. Yes, I used AI to help research a piece about AI ruining hiring. The difference is I’m telling you.
I am curious if there will be a solution or if it will just exponentially get worse. Having hired a few roles this last year, wading through the hundreds of mass produced applications was a real thing to find the few genuine people. 🥺