Almost every Head of Research brief that comes past Spectrum Search this year describes a single role and lists, in the requirements section, the qualifications for four. The mismatch is not always obvious to the hiring board, because two years ago the role was in fact one job and the requirement list was correct. It has fragmented since.
The four jobs hiding inside one title
The first is a research director: someone who owns the lab's bet-portfolio, decides what gets attempted and what gets killed, and signs off on publication. This is the seat the founder usually has in mind when they say 'Head of Research'. The second is a research-engineering leader: someone who runs the training stack, the pre-training reliability, the eval infrastructure. The third is a model-team lead: someone who runs a single model from spec to launch, and is measured on that model. The fourth is a research-to-product translator: someone whose entire job is taking research output and shaping it into something an applied team can ship.
All four of these jobs are now full-time at any serious lab. Three years ago two of them did not exist as standalone seats. The labs that have separated them have moved faster than the labs that have not. That is the operational point hiding under the hiring point.
How the briefs go wrong
The most common version is a Head of Research brief that, on reading, describes job one (research director) and screens for job two (research-engineering leader). The candidate the board ends up wanting is a senior research scientist with eight years of publishing, who in the interview is then asked how they would scale the training cluster. Those candidates exist. There are roughly twelve of them in the world and they are not moving for a Head of Research title.
The second-most-common version is a brief for job three (model-team lead) being interviewed as if it were job one. Model-team leads are usually six-to-eight years out of a PhD, have led a single major model release, and are not yet ready to set portfolio-level strategy. Asked in the third interview to articulate a three-year research agenda, they freeze, and the board reads it as a lack of vision. It is not a lack of vision. It is a category error.
“Asked to articulate a three-year research agenda, model-team leads freeze. The board reads it as lack of vision. It is a category error.”
What to do before opening the search
Two questions resolve almost all of this before a search opens. The first: in eighteen months, is this person measured on a portfolio of bets or on a single model? If it is a portfolio, you are hiring job one. If it is a model, you are hiring job three. The second: who currently does the work this person will do? If the answer is 'the CTO' you are hiring job one. If the answer is 'a senior IC who has plateaued' you are hiring job two. If the answer is 'no one yet' you are usually hiring job four and should say so out loud.
The label is going to follow the org
I expect within eighteen months the labs that have moved fastest will stop using 'Head of Research' as an external title for any of these seats. They will use 'VP Research', 'Director of Research Engineering', 'Head of [Model]', and the title will start to do the work the brief used to do badly. Firms still running 'Head of Research' searches in 2027 will, I think, be telling on themselves.
