Let’s talk about the hiring loop where the machine writes the application and the machine reads it — and the human becomes paperwork.
There is a pattern forming that most people in hiring do not want to name out loud because it implicates everyone in the room.
It goes like this.
A candidate opens ChatGPT and types “write a cover letter for a senior product role at a fintech startup. Make it sound like I have experience with compliance-heavy environments.” The output is generic but polished. The candidate tweaks two sentences, pastes it into the portal, and hits submit.
On the other side, the recruiter’s ATS runs the application through an NLP filter trained to score keyword density, formatting patterns, and structural signals that correlate with successful past hires — most of whom also used AI to write their applications.
Nobody in this transaction has read anything written by a human.
The candidates who make it past the filter get an asynchronous video interview. The platform scores their eye contact, vocal tone, sentence complexity, and hesitation patterns. The candidate has practiced with an AI interview coach that told them exactly how fast to speak, where to pause, and which keywords to hit.
The machine scores the candidate. The candidate performed for the machine. The recruiter gets a score out of a hundred and schedules the next round.
Here is the part nobody building these tools wants to admit.
Every layer of AI in the hiring pipeline measures compliance, not competence. The filter rewards the candidate who knows how to talk like the system expects. It penalizes the person whose resume does not conform to a template, who speaks with a rhythm the scoring model was not trained on, who has a work history that does not fit the clean narrative arc the model prefers.
The confident generalist who cannot be bothered to optimize for keywords gets screened out. The mediocre operator who knows how to game the AI interview coach gets through. The system selects for the people who are best at talking to the system — which is not the same thing as being good at the job.
There is research on this now, though it does not get much attention because the companies selling the tools have no incentive to publish it. A 2025 study by a European labor institute found that AI-screened hiring pipelines produce candidates who score higher on standardized interview metrics but lower on peer-reviewed work output after six months. The system selects for presentation, not performance.
The operator lesson is uncomfortable.
If you are a candidate and you refuse to use AI for your applications, you are not making an ethical stand. You are making yourself invisible. The filter does not know you are a principled human. It knows you did not use the right keywords in the right structure. You get binned before any human sees your name.
If you are a hiring manager and you rely on AI scoring without understanding what it actually measures, you are not saving time. You are outsourcing judgment to a system that optimizes for the wrong signal and has no way to know it is wrong.
The deeper pattern is this.
Every industry that inserts AI as a decision layer between two humans ends up training both sides to optimize for the machine. The platform captures the interaction. The incentives bend toward what scores well. The actual human judgment — reading a resume and feeling whether this person has the thing, sitting across from someone and sensing whether they will work — gets replaced by a proxy that feels objective but is not.
This is not a critique of individual tools. The individual tools are fine at what they do. The problem is the loop. When both sides are talking to AI, the human disappears from the transaction entirely.
The fix is not to ban AI from hiring. That ship has sailed and the return journey is not coming.
The fix is to demand transparency about what the scoring actually measures, blind spots the model has, and cases where the AI’s preferred candidate demonstrably underperforms. And then, critically, to build a human override that is actually used — not a checkbox that hiring managers skip because the score already told them the answer.
Because the real question is not whether AI can screen resumes faster than a human.
It is whether, in a system where machines write to machines about humans, there is still a point where a person reads something written by another person and decides.
If that point disappears, we have not improved hiring. We have built a machine that hires itself.
You have got this.