Why Invisible Technologies Hires From Ivy League AI Groups

Picture an email landing in the inbox of a physics department at a top-tier university. It says, roughly: is anyone here interested in training one of the world's leading AI models? Here's the rate card. Here's what you might make.

That's a real email? Not a recruiter DM, an actual cold outreach to a department?

Francis Pedraza described exactly that on the Masters of Scale podcast in May twenty twenty-five. He's the founder of Invisible Technologies. And the reason I love it is it tells you something counterintuitive about how AI companies actually hire from universities... because it's not the job you'd assume.

Meaning what, they're not hiring these people to build the models?

Not even slightly. And that's the loop I want to close by the end. Because the data says the engineering jobs are going AWAY from campus.

Okay, hang on. Everyone says AI labs are raiding campuses. Business Insider ran that exact story.

They did, in April twenty twenty-three. Mid-six-figure offers, students vanishing mid-degree. Kavita Bala, the dean of computing at Cornell, put it beautifully: quote, they dangle a lot of resources and money in front of young undergrads who at twenty-one are like, this is great, and they vanish.

Brutal.

But then look at the numbers. The VC firm SignalFire ran its Beacon data and found the share of AI hires coming from top schools like Harvard, Princeton, Yale, Stanford, Caltech, or holding doctorates, peaked around thirty-five percent in twenty fifteen. By twenty twenty-three it was closer to eighteen percent.

Halved. So the campus pipeline is shrinking, not growing.

Shrinking for engineers. SignalFire literally published a piece in January twenty twenty-four titled: hire from these nine AI-vy League companies, not Ivy League schools. Their argument is that a Harvard diploma or a PhD is no longer the best signal. They list OpenAI, Anthropic, Cohere, Hugging Face, Midjourney and others as the actual talent pool.

So if you want someone to build your training pipeline, you poach from a lab, not a lecture hall.

Right. And you can see it in Invisible's own job ads. Their listing for VP, Head of AI Training Research asks for eight-plus years in applied ML, three-plus leading teams, and it says flatly, quote, this is not a purely academic role.

They put that in writing.

In writing. And a posting from October sixteenth, twenty twenty-six, Head of Frontier AI, wants ten-plus years in enterprise sales and existing senior relationships at foundation labs. That is a Rolodex job, not a dissertation job.

Then why email the physics department at all?

Because of what Invisible actually sells. Reuters reported in September twenty twenty-four that AI model trainers now need advanced degrees, and that Invisible employed five thousand specialized trainers globally. Inc. reported last November that the expert network had grown to roughly twenty thousand.

Twenty thousand. So the subject-matter experts ARE the product.

They're the product. Reuters framed it as: it takes smart humans to stop models hallucinating. If you need someone to grade a model's answer on graduate quantum mechanics, you can't hire a prompt engineer. You need a physicist. And where do physicists live?

Oh. The department IS the talent marketplace.

And Pedraza gave a second example on that podcast. Suppose a lab needs Claude to be excellent at Tibetan. How do you hire a cohort of Tibetan speakers? You do what he called a very targeted search. He was careful, by the way, to say he wasn't claiming Invisible did that particular work with Anthropic. He used it as an illustration.

Good of him. What's the scale of applications, though? That sounds like a flood.

Three hundred sixty-six thousand people applied in a year, he said. They hired about a thousand. And roughly half of hires came through referrals from existing team members.

So the real pipeline is a group chat.

Partly, yes. Meanwhile the company around it is conventional and large. One hundred million dollars in growth funding announced September sixteenth, twenty twenty-five, led by Vanara Capital, total raised one forty-four million. Inc. reported a valuation above two billion dollars. CEO Matthew Fitzpatrick came from running McKinsey's QuantumBlack Labs. Team of three hundred fifty.

Three hundred fifty staff. Twenty thousand experts.

And that's the payoff. The Atlantic reported in July twenty twenty-six that across OpenAI, Meta, DeepMind and Anthropic they counted more than eighty current or former professors, likely an undercount. So the frontier labs hire the faculty. Companies like Invisible hire the expertise, by the hour, from everyone else in the building.

Two completely different doors into the same university.

And only one of them asks you to leave.