What AI actually changes about hiring

Editorial

Since I first took an interest in HR, I've read my fair share. I came to it sideways, starting my career as an investor covering HRtech theses rather than as an HR operator, which has its pros (a broader macro view) and cons (no lived HR-operator experience). One thing stuck with me from the start of my time studying HR: the people decisions that shape a company most (who it hires, who it promotes) tend to be the ones made with the most human bias, hiring on vibes and promoting whoever is best at playing politics, not whoever delivers the most real value. Closing that gap is a large part of what we at Comp have worked on since.

I'd confidently say that almost everything written about AI in recruiting describes a similar funnel, just running faster: more sourcing, quicker screening, résumés read in seconds. That is real, but it is the least interesting part of what is happening. The first issue of this memo argued that AI changes what the HR function is made of, its cadence, its data, its coverage, and the role of the human inside it. That was the general case; this issue puts it to work on one sub-function, talent acquisition.

Recruiting was built to optimize for speed and volume for one reason: a recruiter’s hours were the bottleneck. Every decision downstream of that, from keyword filters to sampled reviews to time-to-fill as the metric that mattered, was a workaround for scarce human attention. When the cost of sourcing and screening falls to near zero, the bottleneck relocates downstream, onto the one thing cheap compute cannot (yet) solve: judgment about fit and quality of hire. The whole pipeline then reorganizes around where the constraint now sits.

Three consequences follow, and none of them is about speed.

The first: recruiting finally gets to optimize for the number it always wanted. Time-to-fill was only ever a proxy, the metric recruiting could measure while the one that mattered (quality of hire) stayed out of reach because the loop was too long and too manual to close. When an agent sources and the loop runs one step further, a candidate is hired and then performs well three or six months in, the signal of what a strong hire looks like for that role feeds back on its own. The pipeline gets more targeted over time without anyone recalibrating it by hand.

The second: the screen stops matching keywords and starts evaluating the full applicant population against what actually predicted success in that role. That is exactly where bias gets audited out or quietly amplified at scale, depending on whether you can see how the model decided, which is why explainability stops being a compliance checkbox and becomes a design requirement.

The third is quieter and easy to miss. The same skills graph that powers external sourcing also indexes the people already inside the company, so every open role can be matched against internal candidates before it is ever posted. “Internal-first” stops being a mobility policy that everyone endorses (but almost no one follows) and becomes the default path, because the fastest and cheapest match is usually already an employee.

None of this is theory for us. We have been running recruiting at Comp with no dedicated recruiter team. Leaders do their own hunting and interviewing, and AI does the operational lifting. Everything about how we hire lives in one self-improving document the AI reads for context, from our talent philosophy to our pay policy. Assessments are binary (a 0 or a 1) with no fence-sitting; no interviewer sees another’s write-up before submitting their own; and after each round the AI drafts the assessment from the transcript for the interviewer to edit, learning from every edit. Offers generate against our salary table and anything off-standard routes to me.

Now, look at where the human sits in all of that. The operational parts that ate most of the week fall away, and the work that matters moves up rather than disappearing: to building relationships, to selling the opportunity, to the conversations that decide a hire, and to the judgment about fit. That judgment is the discipline that compounds. Every time one of us edits an AI-drafted assessment or approves a shortlist, that call either ends with that one hire or feeds back to sharpen the next recommendation, and only the second kind makes next quarter’s hiring better than this one’s.

Next in this series, the same lens on performance and development.

Once again, let us know what you thought of this piece. The newsletter compounds when you do.

Hope you enjoy what follows,

Chris Gerlach · Co-Founder & CEO, Comp