Theo Kitsberg wants you to fire your recruiter. Not hire a new one—fire the concept entirely. His pitch arrives via Slack message, not sales deck: Brief us once on the role you need filled, then step back. AI agents will handle the rest.
It sounds like the kind of promise that saturates the venture-backed corners of hiring tech these days, except Kitsberg's four-person startup has published something most competitors haven't: a benchmark showing their system outperforms established AI recruiting tools by more than 20 percentage points.
Whether that technical edge translates to actual placements—and whether clients will pay a 15% success fee for software to do what human headhunters charge 20% for—remains an open question. But Prism, as the London and San Francisco-based company is called, is betting that businesses don't want more recruiting tools to learn. They want recruiting done.
The distinction matters more than it might seem.
A Benchmark Test They Didn't Write (But Passed)
On May 3, Prism released results from PeopleSearchBench, an industry standard created by their competitors to measure how well AI systems surface the right candidates from conversational queries. Think: "Find me a senior Rails developer in Berlin who's worked at a Series B startup."
Prism scored 89.64 on the recruiting subset. Lessie, which helped author the benchmark, came in at 68.23. Juicebox scored 65.73. Exa managed 64.67. Claude Code—just 50.50.
The company flagged the obvious conflict in their technical report: they're a vendor grading themselves on someone else's test. They limited the comparison to recruiting queries specifically, ran seven scoring iterations, published confidence intervals, and documented the self-evaluation caveat upfront. The methodology looks sound, if not exactly independent.
Still, a 21-point lead is hard to dismiss outright. And for a startup that launched publicly just weeks earlier, it's an unusually concrete claim in a market thick with vague promises about "AI-powered talent intelligence."
The Whole Search, Not Just the Software
Here's where Prism diverges from the pack. You don't log into their platform. There is no platform—not for clients, anyway. Instead, you describe the role in Slack. Their system then hunts across LinkedIn, email, WhatsApp, a pre-screened candidate pool they've assembled, and your applicant tracking system if you're already using one.
Autonomous agents handle the initial outreach and screening. According to the company's Y Combinator launch post from mid-May, they can run "dozens of searches in parallel," each one heavily instrumented to track what's working and what isn't. Candidates who clear the automated screens land in a ranked shortlist. Only then do you get involved—for interviews, final calls, offers.
Kitsberg, who studied philosophy and psychology at Cambridge before running recruiting for a UK crisis helpline, frames it plainly: "No software, no team to build. Brief us once."
It's an agency model dressed in AI agent clothing. You pay a $500 retainer to start, then 15% of first-year salary if someone accepts. That's on the low end of what traditional headhunters charge—usually 15% to 25%—but it's still a human-recruiter fee structure, just with fewer humans involved.
During a Product Hunt launch on May 9, where Prism landed as the third-ranked product of the day, they waived the retainer for early customers. A classic growth tactic, perhaps more aggressive than one might expect from a company claiming the tech speaks for itself.
Two-Sided Ambitions
Prism also built something called Ray—an "AI career agent" for job seekers. Candidates upload CVs, get matched to roles, access upskilling courses and flashcards. The company is trying to create a two-sided marketplace, essentially. Details on how many candidates have actually signed up, or whether Ray is driving meaningful placement volume, remain sparse.
By early June, Prism's website displayed client logos: KelAI Capital, Stint, Cars24, Corgi. These are self-reported testimonials. No one has verified them independently, and the company declined to share metrics on successful placements or average time-to-hire.
The team itself is still tiny. Four people listed on the Y Combinator profile. Job postings for founding and junior recruiters suggest they're scrambling to scale. Co-founder Axel La Pira—who studied computer science and philosophy at Oxford and worked at French healthtech unicorn Alan—handles the technical side as CTO. Tom Blomfield, the Monzo founder turned YC partner, is their main contact within the accelerator. That's a strong sponsor to have, though not a guarantee of anything.
A Crowded, Noisy Space
The AI recruiting market is, to put it mildly, saturated. Lessie sells its people-search tech as a standalone tool for in-house recruiters. Juicebox markets PeopleGPT with access to over 800 million profiles and "agentic outreach" capabilities. Other YC companies—Contrario, Symphony, Spott—are attacking adjacent parts of the hiring funnel with their own flavors of automation.
What most of these companies have in common: they sell software. Tools that require someone on your team to operate them, interpret results, manage outreach cadences.
Prism is wagering that companies would rather outsource the whole mess. Don't learn a platform, don't assign headcount to wrangle it—just pay on results. That's a harder business to scale than pure SaaS. Even with agents doing the grunt work, someone at Prism still needs to coordinate searches, handle edge cases, manage client expectations. It's a service layer that doesn't evaporate.
But it's also lower friction. If you're a 30-person startup without an in-house recruiter, you don't want software. You want hires.
The Harder Test Ahead
A benchmark score is one thing. Running live searches at scale—across multiple roles, industries, geographies—without quality collapsing is another. Proving that a semi-automated service justifies a 15% fee when clients could theoretically use Juicebox or Lessie themselves for a fraction of the cost? That's the real test.
Prism's website footer lists San Francisco, though YC has them in London. They're likely straddling both markets, which is common enough for early YC companies trying to access U.S. venture capital while keeping engineering costs lower in Europe.
Founded in 2025, launched publicly in May 2025, still a team of four as of mid-2026 trying to prove that autonomous agents can replace an industry built on relationships, judgment calls, and the occasional lucky LinkedIn message.
The technology might be there. The benchmark suggests as much, anyway. Now they need to show the service actually works—and that companies will pay for it.
