In hiring, AI should touch the admin, not the decision.

Most of hiring is admin, not assessment. The smart move is to let AI carry the volume (job descriptions, acknowledgements, summaries) while people keep every decision that affects whether someone gets the job.

A people lead told me last month that her team spends more time acknowledging applications than assessing them. Three hundred CVs for one role. Most never get a reply, not out of rudeness but because there simply aren't enough hours in the day.

This is where AI earns its place in hiring, and also where it gets dangerous if you're not careful.

The dull half is the easy win

A lot of recruitment is administrative, not evaluative. Writing the job description. Sending acknowledgements so candidates aren't left in silence. Pulling twenty applications into a consistent summary so a human can actually compare like with like. Drafting interview questions tied to the role's real requirements. None of that needs judgement, and all of it eats the week.

Hand that part to AI and your people get their time back for the bit that matters: talking to candidates and making a call.

The decision stays human, deliberately

The temptation, once the summaries start looking good, is to let the tool rank people and quietly filter some out. Don't. Not because the model can't produce a list, but because hiring decisions carry obligations that a confident-sounding summary will paper over. Australia's privacy reforms are tightening expectations around automated decisions that affect people, and a candidate rejected by a black box is exactly the sort of thing that comes back on you.

The line we draw with clients is simple. AI prepares the work, people make the decision. Anything that affects whether a human gets a job, a loan or a service stays with a human who can explain it.

How to start without creating a problem

Pick the least consequential part first. Job descriptions and acknowledgement emails are a safe place to build the habit, because nothing about them decides anyone's future. Once the team trusts the output there, move to application summaries, with the rule that the summary informs a person rather than replacing them.

You get most of the time back from the admin, and you keep the judgement, and the accountability, where it belongs. That is usually the whole game: let AI carry the volume, and protect the decisions that deserve a person.