Every scale-up hits the same wall. The first big hire, the one that is not inside anyone’s network and getting it wrong is not a slow quarter. It’s the company.
Run a tight process. Score against a rubric. Don’t hire on vibes. And then founders hire the wrong person anyway. Not because the process failed. Because for top talent, focusing on the interview process is solving the wrong problem.
The real problem isn’t selection. It’s visibility.
A structured interview does not tell the founder whether they are talking to the right person in the first place. For leading AI hires, that is the question that matters. The pool of people who can lead a training run or solve a specific class of alignment problem is not thousands. It is dozens.
The Stanford HAI 2026 AI Index, drawing on Zeki’s dataset of 658,000 frontier AI researchers, shows US net inflow of this talent collapsing from a peak of 324.6 in 2022 to just 26.0 in 2025 — a fall of roughly 92% in three years.
Switzerland and Singapore now lead the world on a per-capita basis at around 110 frontier researchers per 100,000 inhabitants. Denmark, the Netherlands, and Finland rank above Germany and the UK on the same measure. The map of where the right candidate lives is being redrawn faster than most hiring pipelines can adjust.
At this level of scarcity, the question is not how do I evaluate the candidate in front of me? It is am I even looking at the right twelve people?
A founder running a flawless interview process on the wrong shortlist ends up with a flawless interview of a mediocre candidate. The founder will not know, because the only reference point is the other candidates who replied to posting.
What actually moves the needle
Three things separate the founders who make this hire well from the ones who don’t. None of them are interview techniques.
They know who the candidates actually are, from outside the CV. Not where they worked — what they contributed. The research record, the models they touched, the code they shipped, the citation footprint that ties them to real work. A candidate’s own telling of their career is a marketing document. The verifiable trace they leave in the field is not.
They see the network, not just the individual. Frontier researchers do not work alone. Every senior hire brings a second-order question: who has this person built with over the last five years, and which of those people would follow them? This is invisible in any interview and invisible on any CV. It is only visible in longitudinal data that tracks who actually collaborates with whom.
They benchmark against the real market, not the inbox. The most dangerous moment in a first frontier hire is when the founder, with no frame of reference, mistakes a middle-of-the-distribution candidate for an exceptional one because the candidate is clearly smarter than anyone who has come through the process so far. That is not a signal of quality. It is a signal of a narrow pipeline. Knowing where a candidate sits in the actual global distribution of people who do this work — not the distribution of people who responded to a recruiter — is the difference between a hire and a guess.
The honest checklist
For a first or early senior hire, the checklist is short.
- Run a structured interview loop. Necessary. Not the hard part.
- Use a real work sample. Necessary. Also not the hard part.
- Verify the candidate’s contribution to the field from external signal. What did they actually ship, publish, release, or move forward? Independent of what they say they did. This is where most hiring processes go dark, and it is the single biggest source of avoidable mistakes.
- Map the candidate’s collaborator network. Who do they build with, and what does that network say about the team they would bring with them — or leave behind?
- Know the shape of the real market. How many people in the world could plausibly do this job? Where are they employed today? Which of them are showing signals of movement? A founder who cannot answer these questions is running a process against whoever replied, not against the market.
The first two items raise the floor. The last three raise the ceiling. The last three are what talent intelligence is for. This is the work Zeki exists to do. Zeki identifies and tracks the world’s frontier AI researchers — recognised not by self-reported titles but by the research, data, and models they have actually contributed.
The first frontier AI hire is the biggest bet a scale-up founder makes. A good process is how the bet is placed carefully. Talent intelligence is how the founder knows the bet is on the right horse — and that the horse exists in the first place.
Sources: Sackett et al., Journal of Applied Psychology (2022); Schmidt & Hunter (1998); Stanford HAI 2026 AI Index.