Scaling Hiring With AI Agents: 40 Hires, 5 Recruiters

Last fall, a 600-person software company ran the biggest hiring push in its history: 40 hires in a quarter, across engineering, sales, and support, with a talent team of five. Two years earlier, a push half that size had nearly broken the same team. I spoke with people across the company and its candidate pool about how the quarter actually went. What follows is their account, in their words, lightly condensed. Names changed, texture kept.

The recruiter

"Everyone asks about the volume, and the volume is real, but the thing I'd actually put on my tombstone is the Tuesday in October when I realized I hadn't apologized to anyone in three weeks.

You have to understand what a hiring push used to feel like from inside. It's an apology machine. Sorry for the delay, sorry we need to reschedule, sorry I'm only getting back to you now. You become a person who disappoints strangers professionally. Last fall, the follow-ups just happened. Candidates got told things before they thought to ask. The scheduling ran itself, and I mean that literally; there's an agent that negotiates with panel calendars, and when it can't resolve a conflict it pings me with the two options and I pick one on my phone.

My days went to screens and closes. The two things I'm actually good at. I did more closing conversations in that quarter than in the previous year, and we lost fewer accepted offers to cold feet than ever, because a candidate who's been kept warm for six weeks doesn't get cold feet."

The hiring manager

"I run a platform team, and I'll confess what hiring managers usually don't: I used to be the bottleneck. Scorecards sat in my queue. Debriefs started with me flipping through notes I couldn't read. I once ranked a candidate from memory nine days after the interview, which, said out loud, is malpractice.

Last fall the scorecard drafts landed before I'd walked back to my desk. My interview, transcribed, mapped against our actual criteria, with the quotes attached. I edited every single one, sometimes hard, because the draft would occasionally read charisma as competence and I've been burned by that exact confusion enough times to smell it. But editing a wrong draft twenty minutes after the conversation beats reconstructing from fog a week later. It's not close.

The part nobody warned me about: the question packets kept catching my team asking redundant questions. Round three would get a note, essentially, 'this was covered twice, here's what hasn't been probed.' My interviewers grumbled for two weeks and then started competing over who got assigned the interesting gaps."

The candidate

"I interviewed at four companies in September. I signed with this one, and I've thought a lot about why, because on paper another offer was stronger.

Two of the four companies ghosted me mid-process. Not rejected. Ghosted, then resurfaced weeks later like nothing happened. The third was fine. This one was different in a way I struggled to name at the time. Every message I sent got answered fast, and when the answer needed a specific person, I was told who and when, and the when held. Between rounds, someone, something, I honestly don't know and stopped caring, sent me a note about what the next conversation would focus on, so I never walked in guessing.

At one point I asked a hard question about equity refreshers at 10pm. The reply said, more or less, here's the standard policy, and the part of your question about refresh timing is above my pay grade, so Elena will call you tomorrow. Elena called at 9:15. I remember thinking: if this is how they treat people they haven't hired, I can work with how they'll treat me after."

The coordinator

"I'll be blunt because you promised anonymity: eighteen months ago I assumed I was watching my job end. Scheduling was my job. An agent schedules now, better than I did, and it doesn't cry in the stairwell during finals-week-style crunches, which I have personally done.

What I didn't predict is what I'd become instead. My title now has 'experience' in it, and the work is everything the machine hands off: the candidate who needs accommodations the standard flow doesn't cover, the panelist who keeps declining invites and needs an actual human conversation about it, the onsite that has to be rebuilt same-day because a flight got cancelled. I'm the exception handler. The exceptions are the human part, it turns out. There are fewer of us doing this work than before, I won't pretend otherwise. But the work that's left is work I'd choose."

The engineer on the panel

"I've done maybe eighty interviews across my career and I used to be bad at two parts of it: preparing, and writing it up. The middle hour I'm decent at.

Last fall, prep meant reading a packet the night before with questions drafted for me specifically, built off what earlier rounds already covered about that candidate. First time it happened I bristled, honestly. Felt like being handed a script. Then I noticed the packet knew things I'd never have checked, like that the candidate had already answered my favorite system-design question twice, and that nobody had probed the gap in her incident-response story. I asked the gap question. It turned out to matter a lot, in her favor.

Afterward, the scorecard draft was waiting with my own interview quoted back at me. I changed one rating and kept the quotes. Fifteen minutes, done same day. My old write-ups took an hour and happened, if I'm honest, about sixty percent of the time."

The VP of talent

"The board asked me in July whether we could do 40 in a quarter, and the honest answer was that I didn't know, because nobody had stress-tested the new way of working at that load. So the quarter was the stress test.

Where it held: throughput, obviously. Time-to-hire came down by more than a third against our last push, with a team of five instead of the nine I'd have needed before. Candidate withdrawal rates were the lowest we've measured. And I could see everything, all quarter, in the channels: every shortlist argued over, every funnel dip decomposed by Monday morning, every escalation landing with a named human. I've never had that visibility. Previous pushes, I found out what went wrong at the retro.

Where it strained, and I want this on the record because the vendor-conference version of this story never includes it: quality control on the agents' judgment is a real job that we under-staffed. In week five we caught the ranking reasoning drifting on one job family, overweighting a credential that didn't matter, and we caught it because one recruiter was paying attention, not because the process guaranteed it. We've since made that attention someone's explicit responsibility. The machines don't get tired, but they do get confidently wrong, and confidently wrong at scale is a new category of risk that talent leaders are only starting to build muscles for.

The machines don't get tired, but they do get confidently wrong, and confidently wrong at scale is a new category of risk that talent leaders are only starting to build muscles for.

Would I run the quarter the same way again? Yes, with that one fix. What I keep returning to is a moment in November, walking past our recruiting pod at 6pm during the heaviest week of the push. Two years ago that room at that hour was takeout boxes and thousand-yard stares. This time the lights were half off. Two recruiters were finishing calls. The channel was still scrolling, agents reconciling tomorrow's schedules with each other, and nobody needed to be there watching them do it.

The room was quiet. The pipeline wasn't."