Agentic job search

Conventional job searching is mostly logistics: scanning listings, discarding most of them, rewriting the same document slightly differently, retyping the same details into another form. Judgement is maybe a tenth of the time.

Agentic job search reverses that ratio. The logistics are delegated, and what remains is deciding what you want, judging whether a role is worth pursuing, and doing well in conversations with people. Whether that is an improvement depends entirely on whether you were any good at the part that is left.

That last sentence is the uncomfortable one. For someone who was applying widely and thinking little, removing the busywork exposes rather than solves the problem — there is now nothing between them and the parts of a search they were avoiding.

The clearest wins are in work that is repetitive and verifiable. Scanning hundreds of listings against fixed criteria, spotting that three postings are the same role, re-ordering a CV to match a posting’s vocabulary, keeping track of what you sent where.

These are genuinely tedious and genuinely mechanical, and nobody was ever hired because they did them especially well. Removing them recovers hours that were pure overhead.

The less obvious gain is the mental load. Knowing nothing will be missed — no deadline, no follow-up, no application left in limbo — removes a continuous background cost that people rarely count but feel throughout a search lasting months.

• Filtering listings against criteria you can state
• Deduplicating the same role across sources
• Reformatting and re-emphasising the same material
• Tracking what was sent where, and when to follow up

When applying becomes cheap for everyone, the application stops being a signal. An employer receiving three times the volume filters harder, and the things that survive harder filtering are the ones automation cannot manufacture: a referral, demonstrated work, a specific reason you want this role.

So the counter-intuitive consequence is that agentic search raises the value of the non-automatable parts. The candidate who uses the recovered hours to talk to people and build something visible is in a much stronger position than the one who uses them to send four hundred applications.

This is the same pattern every efficiency gain produces. When a thing becomes cheap, its scarcity value disappears and whatever remains scarce becomes the differentiator — and in hiring, what remains scarce is evidence that someone vouched for you or that you made something.

The obvious failure is treating cheaper applications as a reason to send more. It feels productive, produces a satisfying number, and is close to useless — because the constraint was never how many applications you could send.

The second trap is losing track of what was sent in your name. Agentic search only works if you can still answer, in an interview, why you applied and what interested you. A candidate who cannot is worse off than one who applied to a tenth as many roles deliberately.

The third is quieter: the search becomes passive. Reviewing a feed someone else assembled is a different activity from forming a view about where you want to work, and a candidate who never did the second has no answer when an interviewer asks why this company.

It is worth understanding the other side, because the employer response is what determines whether any of this helps you. A team that received eighty applications for a role and now receives three hundred has not gained three hundred candidates; it has gained the same number of good ones buried in more noise.

The predictable reactions are already visible. More screening questions on the form, more work-sample requests, more weight on referrals, and more scepticism about applications that read as generated. Each of those is an attempt to restore a signal that volume destroyed.

That loop is the reason the strategy advice inverts. Every tool that makes applying cheaper accelerates the shift towards signals it cannot produce, so the candidate who leans hardest into volume is optimising for the thing that is actively losing value while it is being optimised.

• More screening questions, answered before a human reads anything
• Work samples and short paid exercises in place of CV screening
• Referral schemes weighted more heavily
• Detection of templated covering letters
• Smaller, less visible channels where volume is naturally limited

Set the agent to be selective rather than exhaustive. Ask it to surface ten roles worth your attention rather than a hundred that match, and require a reason attached to each so you can correct its judgement.

Then spend the recovered time on the parts it cannot do. That is the whole strategy, and it is why the people getting the most from these tools are often applying to fewer jobs than before, not more.

Keep one habit manual on purpose: read the posting yourself for anything you actually want. It takes three minutes, it is where the specific detail for a covering note comes from, and it is the difference between applying to a company and applying to a row in a list.