Handshake

A job board grew up and started teaching the machines

Walk into a frontier AI lab today and ask where the training data comes from - the careful, expert-graded examples that teach a model the difference between a correct proof and a confident wrong one. There is a decent chance the answer is a company most people still file under "the app I used to find an internship." That company is Handshake.

For a decade Handshake was the place a sophomore at a state school went to find a summer job. Free to students, sold to employers and universities, quietly running the career fair when nobody was watching. Then, in early 2025, it did the thing companies talk about and rarely do: it picked up its own foundation and moved it. The student network became an expert network. The matching engine that once paired a finance major with a bank now pairs a physics PhD with a lab that needs someone to grade the model's homework.

Talent is everywhere. The network is not.

Here is the uncomfortable truth about early-career hiring: for a long time, the best jobs went to the students who already knew someone. Recruiters flew to a short list of brand-name campuses, the same internships circulated through the same dorms, and a brilliant engineer at a school nobody had heard of simply never got the email.

The founders had lived the wrong end of this. Garrett Lord grew up in Michigan, went to community college, then Michigan Tech - a fine engineering school that no recruiter was booking flights for. He landed internships at Los Alamos and Palantir anyway, and noticed the obvious injustice: he had the skills, but almost not the access. The talent was distributed evenly. The opportunity was not.

Three students, one stubborn idea

In 2014, Garrett Lord, Ben Christensen, and Scott Ringwelski - all computer science students at Michigan Tech - bet that you could rebuild college recruiting around what students could do instead of where they were sitting. Not a job board with better filters. A network where a student's profile, skills, and interests did the introduction that a personal connection used to.

It was an unfashionable bet. LinkedIn already existed; so did Indeed; so did every campus career office with a filing cabinet. The founders' wager was that none of them were built for the specific, anxious, first-time moment of a 20-year-old looking for a first real job - and that universities would happily hand over their career services if someone made software that actually worked.

One platform, three audiences, and a plot twist

The original Handshake was a three-sided machine. Students got a free profile, job and internship listings, virtual career fairs, and recommendations tuned to what they actually wanted. Employers got tools to find, message, and screen early-career talent without flying anywhere. Universities got software to run their career centers and, crucially, to prove their graduates were getting hired.

Then came the plot twist that no slide deck predicted. The AI boom created enormous demand for one scarce thing: humans who could produce and grade expert-level data - math, physics, code, the hard stuff models still get wrong. Handshake looked at its network of students, PhDs, and specialists and realized it was sitting on exactly that supply. In January 2025 it launched Handshake AI: a marketplace that turns vetted experts into a training-data engine for frontier labs.

The numbers stopped being about job boards

For most of its life Handshake's core platform was a healthy, growing SaaS business - around $190M in revenue by 2024. Respectable. The kind of number that earns a company a comfortable middle age. Then Handshake AI arrived, and the chart broke its own axis.

Eight of the top AI labs - reportedly including OpenAI - became customers. The experts grading data were earning $100 to $125 an hour. And to make sure all that human effort actually produced correct labels, Handshake acquired Cleanlab in early 2026, a startup whose software automatically flags data that humans got wrong. A quality-control layer for an industry that runs on trust.

The old business did not vanish. Students still log in, employers still post jobs, universities still run their fairs. But the growth - and the story - now lives in the part of the company that did not exist eighteen months ago.

Democratize access to opportunity - even when the opportunity changes

The mission statement never moved: democratize access to opportunity. What is quietly clever is that the AI pivot is the same mission wearing a different outfit. The original product gave students without connections a way into good jobs. Handshake AI gives experts - graduate students, specialists, people with deep knowledge and shallow networks - a way to earn real money from what they know, on their own schedule.

It is still, at heart, a company that finds the talented person the system overlooked and hands them a way in. The system just got a lot stranger.

Whoever owns the experts owns the next model

The competition is no longer just LinkedIn and Indeed. It is Mercor, Scale AI, Surge, and a field of data-labeling companies racing to supply the labs. Handshake's edge is the one it spent ten years building by accident: a vetted, motivated network of educated humans, and the matching machinery to deploy them. As models get better, the data that improves them gets harder to make - which makes the people who can make it more valuable, not less.

There are real risks. Early-career hiring has softened; job postings on the core platform have slipped. AI data labeling could commoditize. A company that re-founds itself once can be forced to do it again. Skeptics are right to keep one eyebrow raised.

Walk back into that AI lab. Ask again where the training data comes from. The answer - a company that used to run career fairs - is no longer a surprise; it is the point. Handshake spent a decade convincing the world that talent without a network still deserved the call. Now it is making sure the machines learning to place that call were taught by the people the old system overlooked. The handshake is the same. The room got a lot bigger.