A hiring manager we spoke to had 400 applications for one Director of Finance role.
Every resume was clean. Every cover letter was tailored to the posting. A lot of them echoed the exact phrases from his job description, because the candidates had fed that job description to a model and asked it to mirror the language back.
On paper, dozens looked qualified. He had no way to tell which ones could actually do the job.
That's the honest version of the AI conversation in recruiting. AI has made large parts of hiring genuinely faster. It has also made candidates much harder to read.
Start with what AI actually does well
We use it. It would be strange to write about AI in recruiting and pretend otherwise.
It finds people. A good sourcing setup surfaces candidates across a far bigger pool than anyone works through by hand, and it does it in an afternoon instead of a fortnight. It enriches company data, so we know headcount, systems, and which businesses are mid-implementation on an ERP before we ever pick up the phone. It preps research, so we walk into a client call already understanding their business. It handles the scheduling and the notes and the follow-ups.
All of that was the slow part of the search, the busywork around the actual decision. AI compressed it, and that's a real gain for clients, because it means more of the week goes into the work that decides the outcome.
What it quietly broke
The same tools reached candidates at the same time, and that changed what a resume is worth.
Roughly two thirds of job seekers now use AI to write or polish their resume. Cover letters, the same. Increasingly, so are interview answers, rehearsed against a model that has read the job description and knows exactly which competencies to hit.
Meanwhile, on the other side of the table, most companies now run some form of AI screening on inbound applications. So a machine writes the application, and a machine reads it. Somewhere in that exchange, the signal a hiring manager used to rely on disappeared.
Polish used to be a rough proxy for competence. Someone who wrote a sharp, specific resume had usually done sharp, specific work. That correlation is gone. Anyone can produce a document that reads like a strong operator wrote it, for free, in a minute.
The damage shows up in the evaluation layer
At high-volume junior hiring, this is an expensive annoyance. You wade through more noise to find the same people.
At Director-to-C-Suite, it's a different problem, because the cost of being wrong is a year of momentum and somewhere north of a quarter of a million dollars.
A candidate who can describe leading an ERP implementation in fluent, confident, well-structured language is now hard to separate from one who actually led it. On paper they are the same person. In a first conversation, often still the same person, because the first conversation is the one a model can rehearse you for.
“AI can tell you who to talk to. It still can't tell you who can do the job.”
What still separates people
The texture of work that was actually done.
A model can generate a confident account of a month-end close. It can't generate the memory of the close that fell apart at 2am, what broke, who you called, and what you changed so it didn't happen again. It can produce a tidy summary of an ERP rollout. It can't tell you which department refused to move off their spreadsheets and how you got them there.
So we go after the parts of the job that don't survive summarising. What went wrong. What it cost. Who pushed back. What you'd do differently. And then we keep going, because AI-prepared answers tend to be strong on the first pass and thin by the third. Depth is the thing that can't be pre-generated, and depth only comes from having been in the room.
It works because the person asking the questions has done the job and knows where the hard parts are buried.
The honest answer to “will AI replace recruiters”
It already replaced part of the job, and that part deserved to go. Building lists, chasing calendars, keyword-matching a stack of resumes: if that was the service, the service is now a subscription.
What it can't carry is the judgment. Whether this person holds up in this company, with this board, in this quarter, reporting to this CEO. That call is a bet made with context, and context is the one thing a model doesn't have about your business.
Recruiters whose value was access to a database should be nervous. Recruiters whose value is judgment on senior hires are more useful than they were two years ago, precisely because the paper signal everyone used to lean on has stopped working.
What to ask your recruiter about AI
Three questions worth asking anyone running a senior search for you.
Where do you use AI, and where do you refuse to? A recruiter with a real process will have a clear line and be able to tell you where it sits.
How do you verify a candidate can do what their resume claims? If the answer is vague, what you're buying is a filtered list.
How did you get from the long list to this shortlist? You want to hear about judgment calls and rejected candidates, and the reasons behind both.
The tooling matters less than whether there's a human at the end of it who can tell the difference between a good answer and a true one.
About ICA
I'm April Ben-Sabat, founder of Inner Circle Agency. I built my career in finance, accounting, HR leadership, and industrial operations before starting ICA. ICA recruits for the roles I used to hold, so our team evaluates candidates the way a board would, because that operational experience is baked into how we work.
ICA is boutique by design. We take on a small number of searches at a time so every client gets senior-level attention. We specialize in Director-to-C-Suite placements across finance, ERP, and operations for mid-market companies in the US.
If you're hiring for a role where the wrong person costs you a year of momentum, let's talk.
Frequently asked questions
Will AI replace executive recruiters?
It has already replaced the mechanical parts of the job: list building, keyword matching, scheduling. What it cannot do is make the judgment call on a senior hire, which depends on context about your business, your team, and your board that a model does not have. Recruiters whose value was database access are exposed. Recruiters whose value is evaluation are more useful now that resumes have stopped being a reliable signal.
How can you tell if a candidate used AI to write their resume?
Usually you can't, and trying to is the wrong game. Most strong candidates now use AI to write or polish their application, and penalising them for it would rule out good people. The useful move is to stop treating the document as evidence and verify the substance in conversation instead, by probing for the specifics that only come from having done the work.
Where does ICA use AI in the search process?
We use it for sourcing, data enrichment, company research, and the administrative load around a search. We don't use it to decide who is worth putting in front of a client. Every candidate on an ICA shortlist has been evaluated by a person who understands the function they are being hired into.
Is AI screening reliable for senior hires?
For high-volume junior roles it can be a reasonable first filter. For Director-to-C-Suite roles it is a poor one, because the qualities that matter at that level (judgment, business context, how someone handles pressure and resistance) are exactly the qualities that don't show up in the text a screening model is reading.