Mercor hits $500M ARR run rate in 17 months: Brendan Foody on AI talent and the future of expert data

Summary

Mercor CEO Brendan Foody founded the company with high school friends at 19, scaled to a $1M ARR run rate, dropped out of college, and has now reached $500M ARR in 17 months working with AI labs. The company has 250 full-time employees across the US, India, Latin America, and the UK, plus a broader contractor network.

From academic to professional evals

Mercor's early work centred on sourcing elite academic talent — Foody's first memorable project was placing 25 Olympiad medalists in 24 hours to help labs push models toward superhuman performance on competition math. That phase is largely closing. The frontier demand is shifting toward professional-domain evaluation: what does it mean for a model to build production software, conduct financial analysis, or segment a market? These tasks resist clean scoring, and Foody argues humans remain essential for defining the success criteria that make automated reward signals possible at all.

How the eval pipeline works

The model Foody describes involves humans constructing rubrics and verifiers rather than simply labeling outputs. For a well-scoped task like laundry folding, a vision model can serve as the reward signal directly. For something like scientific experimentation, Mercor contractors run physical experiments and report results. For browser-based workflows, a human expert writes the unit test that checks whether the model changed the right state. The common thread is that human judgment defines the reward boundary; the model then learns to optimize within it.

Foody expects computer-use tasks to be largely solved within two to three years. Beyond that sits a longer, harder problem: agentic trajectories spanning 30 or 90 days, using ten or more tools, interacting with employees across an organization. Building evaluation environments for those workflows is where he sees the most significant near-term opportunity.

What labs still buy

RHF hasn't disappeared, but Foody says data flywheels from real-world product usage are increasingly more efficient for basic preference signals. Where expert human labor commands a premium is in the design work — a contractor who spends five hours constructing a structured evaluation framework for a specific professional capability in a way that product users reliably cannot.

Talent and growth culture

Mercor's contractor network grows primarily through referrals, with a $250 fee per successful placement. On the full-time side, Foody says the early team worked seven days a week without mandated hours and eventually needed to shift toward output-based rather than input-based intensity to accommodate people with families — while keeping what he describes as a high-pressure, customer-focused culture.

Foody previously tweeted "unicorn by end of year" as a seed-stage company and made it happen after raising a $40M seed round. He followed that with a joking IPO tweet. No public IPO timeline has been set.