What recruiters do that agents cannot

Sourcing candidates who match stated criteria, reading every application rather than the first fifty, scheduling, sending updates, chasing feedback from hiring managers, keeping records accurate. These are volume tasks with defined correct answers.

Automating them is good for candidates too. Most complaints about recruiting — no response, slow scheduling, no feedback — are failures of capacity rather than of intent.

It is worth dwelling on that, because it reframes the whole question. The recruiter who never replied was usually not indifferent; they had four hundred applications, six open roles and a hiring manager who would not write feedback. Capacity problems have technical solutions in a way that motivation problems do not.

• Sourcing against stated criteria
• Reading and triaging every application
• Scheduling and coordination
• Status updates and record keeping
• Chasing the feedback nobody wants to write

Persuading a strong candidate who is not looking. Working out that the role as written is not the role the team actually needs. Reading hesitation in a conversation. Managing a hiring manager’s unrealistic expectations. Knowing that two candidates would not work well in the same team.

These are judgement and relationship tasks, and they are where good recruiters create most of their value — while being the least visible part of the job.

The second item on that list is the most undervalued. A surprising share of failed hiring processes fail because the written specification describes a person who does not exist, and the only fix is someone with enough standing to go back to the hiring manager and say so.

As the administrative half automates, the role concentrates on the judgement half: fewer recruiters handling more roles, spending their time on assessment, persuasion and advising hiring managers.

Purely transactional recruiting — forwarding CVs matched on keywords — is the part most exposed, because it is exactly what an agent does well and cheaply.

The agency market is where this shows first. A contingency agency whose product was access to a database and the willingness to send fifty CVs has a weak position once the client can search competently themselves; an agency whose product is a specialist network and a genuine view on the market does not.

More applications will be read, which is good. More initial screening will be automated, which means clear evidence against stated requirements matters more than a well-written narrative.

And the recruiters you do speak to will have more time, which makes those conversations more consequential. Preparing properly for one recruiter call is worth more than it used to be.

The practical adjustment is to write for two readers. Make the requirements verifiably met in plain language near the top for the screening pass, then give the human something worth reading — because the screen is no longer the last stage at which a person forms an opinion of you.

The skills that appreciate are the ones that were always hardest to hire for: assessing capability rather than matching keywords, persuading someone who is not currently looking, and having credible conversations with hiring managers about what the market will actually supply.

The skills that depreciate are the ones built on throughput — knowing the boolean search syntax, working the database faster than the next person, being the one who responds within the hour.

Getting closer to the hiring decision is the most reliable move available. A recruiter who is consulted about whether the role should exist in this shape occupies a different position from one who receives a specification and fills it, and the second position is the automatable one.

• Appreciating — assessment, persuasion, market advice, specialist networks
• Depreciating — search speed, throughput, response time, database access
• Durable — being trusted by hiring managers before the requisition is written

Automation should not be allowed to remove the human decision. A recruiter who approves an automated shortlist without reading it has added nothing while appearing to provide oversight.

The version of this that works is a person making decisions with better information and more time. The version that fails is a person rubber-stamping a system nobody can interrogate.

There is a simple diagnostic for which one an organisation has. Ask how often the human overrules the system, and what happened the last time. If the answer is that it never happens, the oversight is decorative, and everyone involved should stop describing it as a safeguard.