Handshake CEO Garrett Lord: frontier AI labs are hungry for PhD-level expert data — and Handshake supplies it

Summary

Handshake, best known as the dominant early-career recruiting network for the 18-to-30 cohort, has quietly built a second business supplying PhD-level human data to frontier AI labs. Garrett Lord, founder and CEO, says the pivot began roughly 18 months ago when frontier labs and large annotation companies started approaching Handshake asking specifically whether it had access to PhDs and master's students. The answer was yes at scale: the platform carries 500,000 PhDs and 3 million master's students alongside tens of millions of undergraduates, all acquired organically over a decade of university recruiting infrastructure powering 1,600 institutions.

The Business Model

Handshake routes that credentialed talent to AI labs under what it calls the Move Fellowship Program, paying contributors $60 to over $100 per hour for expert annotation work. That rate compares favorably against the roughly $23/hour a typical teaching assistantship pays, and Lord argues it is sustainable precisely because Handshake carries zero customer acquisition costs for this supply. Competitors, he says, are spending tens of millions of dollars monthly on performance advertising to recruit the same physics PhDs and lawyers via Instagram, a structurally inferior approach.

The company is now working with six frontier AI labs, providing tens of thousands of expert contributors. Growth has accelerated sharply: Handshake reported more than 3x growth in the past month on the back of surging demand from labs.

Why PhD Data Matters Now

The demand thesis is straightforward. Models have already consumed the open web, books, and YouTube. The remaining performance gains in reasoning-heavy domains, specifically law, finance, medicine, mathematics, physics, chemistry, and biology, require structured expert input rather than generalist labeling. Lord draws a direct line from the original definition of a PhD (advancing peer-reviewed knowledge at the frontier) to the kind of step-by-step reasoning data labs need most.

He also flags where demand is heading next: audio data, tool-use trajectories, and agentic browser workflows. The example he uses is a travel booking agent that must sequence calendar checks, airport logistics, price sensitivity, and insurance decisions in real time. That class of task, highly niche but economically valuable, is not close to being solved by current models, and Lord expects a long tail of vertical-specific data contracts to follow, analogous to the SaaS micro-vertical wave.

Competitive Position and Outlook

The structural moat is the pre-existing audience. Because Handshake does not need to advertise to recruit contributors, it can simultaneously pay workers more and charge labs less than pure-play annotation vendors, improving gross margin at both ends of the marketplace. Lord frames the next several years as a sustained demand environment for human expert data, particularly as labs push into agentic and multimodal capabilities that current synthetic data pipelines cannot adequately cover.