Guide
AI researcher, applied AI engineer or AI product engineer
Three profiles, three different searches, and one question that tells you which one your problem needs.
The short answer
Hire an AI researcher when you do not yet know whether the capability your product needs can be built. Hire an applied AI engineer when the method exists and the risk is your data, your latency budget and your inference bill. Hire an AI product engineer when the model already works and the open question is whether anyone will use the thing you wrap around it.
Most companies putting AI into an existing product need the third profile first and believe they need the first. The supply leans the same way: 48% of the 721 620 researchers in the Founding Hires index with a known affiliation are early career, and among the 378 954 whose current position has been read, only 17% name a company. Getting that wrong costs about a year: three months to notice, three to be sure, six to recover. If the question is what the titles mean rather than which job you need, the titles are decoded here.
Not sure which one you need
Describe what you are building, we will name the profile.
Most of the expensive hiring mistakes we see are a profile picked wrong at the start, not a search run badly. Tell us the product, the stage and who is already on the team, and we will say which of the three you are actually short of and how hard that one is to find right now.
| AI researcher | Applied AI engineer | AI product engineer | |
|---|---|---|---|
| Hire when | You do not know whether the capability can be built | The method exists, you do not know whether it holds at your scale | The model works, you do not know whether the product does |
| What you get | An answer to an open question, and the papers behind it | A system in production at a viable cost | A product people keep using when the model is wrong |
| Strongest signal | First-authored papers at NeurIPS, ICML, ICLR and their peers | A dated move from research into an industry lab or product team | Shipped repositories other people depend on |
| Where you find them | The published literature, months before they finish a doctorate | Declared employment records and recent subject matter | Public code, releases and issue threads |
| Hardest part of the search | Reaching them inside the short window they are reachable | Telling capability tiers apart under one job title | Convincing your own team the role is senior |
| Expensive mistake | Hiring one for a solved problem | Asking them to invent a new architecture | Paying them like a junior and losing them in eight months |
When should you hire an AI researcher?
An AI researcher earns their keep on open questions. Your product needs a capability that nobody has demonstrated at your constraints, and someone has to establish whether it is reachable at all. That work has no delivery date, and paying for it means accepting months with nothing shippable at the end.
The signal that identifies a strong one is first authorship at a venue with a hard review process. Author order carries meaning in machine learning: the first author did the work. Publication count rewards a large lab where everyone appears on everything, and citation count rewards seniority, since citations take two to four years to accumulate. A doctoral student finishing this year has published their best work too recently to be cited for it, so ranking candidates on citations sorts the people you can hire below the people you cannot.
The cost of this search is access rather than identification. The people you want are employed, contacted constantly, and reachable during a window of a few months at the end of a doctorate. Our full method is in how to find and hire AI researchers.
| What the index holds | Count |
|---|---|
| Papers indexed | 561 727 |
| Researchers on those papers | 1 048 947 |
| Profiles enriched from OpenAlex, ORCID and DBLP | 1 040 229 |
| Profiles with a classified research area | 996 673 |
Updated 2026-10-26
When should you hire an applied AI engineer?
An applied AI engineer takes a method that already works in a paper or a notebook and makes it survive contact with your traffic. Serving, latency, quantisation, retrieval, evaluation, the retraining loop, and above all the judgement about which failure modes matter and which are noise. Hire one when the science is settled and the engineering risk sits in your data and your bill.
The strongest signal here is a career move that has already happened. Someone who published research and then joined an industry lab or a product team carries model intuition from the first half of their career and production constraints from the second, and an organisation with a demanding interview process reached that conclusion about them at its own expense. Researchers declare that move publicly, with a start date, because their scholarly profile exists to establish authorship.
The cost of this search is identification. One job title covers people who fine-tune a model on a laptop and people who run inference for millions of daily requests, so keyword search returns noise. You filter on what their recent work is about. Full method in how to hire an applied AI engineer.
When should you hire an AI product engineer?
An AI product engineer builds everything around the model. Orchestration and tool use, evaluation harnesses, guardrails, cost control, and the interface. A system that is wrong five percent of the time is either unusable or excellent depending on what the product does when it is wrong, and designing that behaviour is this person's job.
They publish little, so the literature will not find them. Their record sits in repositories: code other people depend on, release histories, issues opened by strangers, tests where the hard parts are. Reading that record well takes a different search entirely, covered in how to find and hire AI engineers.
This is the profile teams skip. They hire two researchers, no product engineer, and then wonder why nothing reaches customers. It is also the profile most often paid a tier below the others because the title sounds junior next to Research Scientist.
How do you decide which one you need?
One question settles it. What is uncertain about your problem?
- Uncertain whether it is possible. You need a researcher, and you should budget for a period with no shippable output.
- Certain it is possible, uncertain it will hold in production. You need an applied AI engineer. The method exists and your risk is data, latency and cost.
- Certain it will hold, uncertain anyone will use it. You need a product engineer. The model works in a notebook and the question is whether the thing around it is any good.
- You cannot tell which of the three you are in. Run a two-week spike with a contractor before writing a job description. The answer changes what you hire, what you pay and how long the search takes.
Common questions
- What is the difference between an AI researcher and an applied AI engineer?
- An AI researcher establishes whether a capability is reachable, and is measured by published work that survived peer review. An applied AI engineer makes a known method hold on real data at production scale and cost, and is measured by a system that stays up. The two searches use different signals: first authorship at a top venue for the researcher, and for the applied engineer either a documented move from research into industry or a record of public code, depending on which way they came into the work.
- What is an AI product engineer?
- An AI product engineer builds the software around a model without training it: orchestration, tool use, evaluation harnesses, cost control, and the interface and failure paths that make a probabilistic system usable. Most of the work in shipping an AI feature sits here rather than in modelling, and this is the role teams most often skip.
- Do I need to hire an AI researcher?
- Only if your product depends on a capability that does not exist yet at your constraints. If your problem can be solved by combining existing models with your own data, an applied AI engineer will ship it sooner and a researcher will run out of open problems within a few months and leave.
- What job title should I put on the posting?
- Whichever title the people you want already hold, which is usually not the one that describes the work. Someone doing applied AI work at a large company is often titled Software Engineer or Research Engineer. Write the posting around the problem and the constraints rather than the title, and search on evidence of the work instead of on the words in a headline.
Settle the question
A call is the fastest way to get this right.
A short call on your product and your team, and you leave with the profile named and a read on where that person is likely to be found. If you would rather hand the whole search over afterwards, you pay us only when someone joins.