Research scientist
The one chasing an answer nobody has yet
An AI research scientist works on questions nobody knows the answer to, and a negative result still counts as a result, like a cartographer who also marks where the road stops.
- Floor 1The Model
- P4Writes the model
- Established titleIt has been in adverts long enough to mean something.
What they answer for
The deliverableA result others can reproduce, and the question that made it possible. If they leave, a line of research stops, not a service.
The profile
Five activities, scored 0 to 3- ResearchProducing knowledge that does not yet exist.
- BuildShipping a system that runs, deploys and breaks.
- OperateKeeping it in production: cost, incidents, drift, on-call.
- VerifyMeasuring, testing, attacking. Producing a verdict that holds.
- LeadDeciding, persuading, driving adoption, answering to others.
What the work is
The open question
The trade is recognisable by one thing: at the moment the work begins, nobody knows whether it will work. Neither a paper nor an existing product holds the answer. That is what separates research from implementation, and it is also what makes its schedule impossible to hold.
The success criterion
What separates it from engineering is not technical depth, which can be identical, but the success criterion. An engineer succeeds when the system runs. A researcher succeeds when the question is settled, and a negative answer settles it as well as a positive one.
The trap
The trap is the number of posts. Labs employ two to four research engineers for every scientist, and many adverts carrying the word "research" describe the former. The sign that settles it: does the advert ask for first-author publications, or for experience at scale?
A week in the role
- Reading, a great deal of it, including what was published yesterday.
- Designing experiments, often smaller than you would imagine, to isolate one variable.
- Writing: a paper, an internal note, a proof. That is the deliverable, not a progress report.
- Talking to research engineers, who run at scale what worked small.
Ways in
- A doctorate, or first-author publications at the major conferences. This is the one trade in the corpus where the requirement is real and not copied.
- A move across from physics or mathematics, commoner than people think.
- What the role does not require: knowing how to keep a service running in production.
Reading an advert
2 signs- The advert asks you to "deploy models at scale".
- That is a research engineer post. Both are good; they are not prepared for in the same way.
- No mention of publication or of any conference.
- The lab may not publish, which is a legitimate choice but changes everything: your work will not leave the building, and your next application will rest on what you can tell of it.