Hiring needs a better experience, not an AI arms race

Hiring does not need an AI arms race, because the arms race is a symptom rather than the disease. Candidates reach for AI to write applications and employers reach for AI to sort them because the application itself asks nothing that tells a good candidate from a fluent one, and the way out is to fix that, not to buy a stricter machine.

How the arms race started

The pattern is easy to recognise by now. A candidate who has applied to forty roles and heard back from none of them does the rational thing: they use a tool that writes a covering letter in seconds and fires off forty more. The employer, now receiving far more applications that all read beautifully and say roughly the same thing, does the rational thing on their side: they buy something that screens harder, faster and more opaquely. Each side’s move makes the other side’s move look sensible. Nobody is behaving badly, and everyone is worse off.

What gets lost is signal. When every CV is polished by the same kind of model, the CV stops telling you much. The recruiter reading it cannot tell who actually wants the job, and the candidate who does want it has no way to show it.

The problem is what the application asks

We spend a lot of time looking at application forms, and most of them were designed for the ATS rather than for the person filling them in. They ask for a CV, a covering letter and a set of questions that are identical whether you are applying to run a kitchen or to answer phones. That is exactly the kind of input a language model is good at generating, and exactly the kind a keyword rule struggles to read.

Frontline hiring makes this sharper. A hotel group or a restaurant chain might be hiring thousands of people a year for roles where the CV matters far less than a handful of practical facts: can you work weekends, do you hold a food hygiene certificate, can you get to this site for a six o’clock start. A generated covering letter answers none of those. A short, well-designed application answers all of them in a couple of minutes, and it is much harder to game, because the answers are specific and checkable.

So the first move out of the arms race is on the employer’s side, and it is a design move. Ask for less, and ask for what the role actually needs. Conversational apply is how we do it: the application reads the role and the region and asks only the questions that job needs, in the order it needs them, on a phone, and the answers land in the ATS you already have. A candidate who is not eligible finds out before they spend twenty minutes, and a candidate who is gives you something far more useful than prose.

Reading every application, not ruling people out

The second move is about what happens after the application lands. The instinct in an arms race is to reject faster. The better instinct is to read everything, and to be able to explain what you read.

That is the line we drew for the Hiring Agent. It scores every application against the criteria your team chose and writes down its reasoning: which requirements the candidate met, how they showed it and why they ranked where they did. It puts applications in order and suggests who to look at first; turning someone down is not something it can do. A recruiter opens the queue with the strongest evidence at the top and a sentence explaining why, and the decision stays with them.

The difference matters for the arms race specifically. A black-box rule invites candidates to guess what it wants and to generate more of it. A score with visible reasoning against visible criteria rewards the thing you actually asked for. It also gives your team something to push back on when the reasoning looks wrong, which a silent rejection never does.

Closing the silence

The third move is the one candidates notice most. Much of the escalation on their side comes from silence. If nobody replies, you apply more widely and put less of yourself into each application. If someone does reply, even to say the role has closed, you have a reason to take the next application seriously.

This is admin, and it is precisely the admin agents should carry. The Candidate Agent answers questions about shifts, pay and locations from the employer’s approved content, on a Sunday as readily as on a weekday, and hands the conversation to a person when it needs judgment. Automations handle the handoffs and acknowledgements that used to depend on a recruiter having a free afternoon. None of this is clever. It is courtesy, done reliably, and it takes some of the heat out of the whole exchange.

What to do next

If you think you are in an arms race, look at your own application before you look at another AI tool. Count the questions. Ask how many of them a model could answer convincingly without knowing anything about the candidate, and how many would tell you something about fit for this role at this site. Then look at what happens after someone applies: who reads it, what they see, and how long the candidate waits to hear anything.

If the answers are uncomfortable, the fix is usually closer to home than a new tool. We are glad to walk through how conversational apply, the Hiring Agent and the Candidate Agent fit together on your ATS, and there is more on where AI does and does not belong in hiring in our trust in hiring collection. You can book a demo whenever suits.