An AI research scientist invents new methods, running experiments and publishing or defending results that did not exist before, while an AI engineer builds products on top of models that somebody else already trained. One role is measured by whether it advances what a model can do. The other is measured by whether it ships.
Employers apply both titles inconsistently, so the balance of publishing versus shipping inside a listing tells you more than the words in the job title. This breaks down what each role actually involves day to day, what background each one expects, and which is the realistic way in.
What does an AI research scientist do?
An AI research scientist designs, tests and publishes new methods rather than applying ones that already exist, and that distinction is what separates the role from almost everything else in AI. The daily work is running experiments, reading and writing papers, and pushing a model’s architecture, training process or evaluation past where it currently sits. PyTorch and JAX are the standard tools, alongside an evaluation harness the team usually builds in-house, because a result only counts once it is measured and defensible.
The employers hiring for this title are a mix of frontier labs training their own foundation models and product companies that need a small research function to stay ahead. Live listings carry the title from OpenAI, Cohere, Aleph Alpha, Perplexity, Poolside, Reka and Luma AI on the frontier-lab side, and from companies like Spotify, DeepL, Deepgram and PostHog on the product side. The titles themselves tend to be narrow, naming a specific open problem such as AI safety evaluations, multimodal systems or model scaling, because a title that specific only exists once a team is already deep into one line of work.
What does an AI engineer do?
An AI engineer owns the product wrapped around a model rather than the model itself, building the retrieval, prompting, agent orchestration and guardrails that turn an existing model into something a user can rely on. LLM APIs, RAG pipelines and agent frameworks are the daily tools, and the deliverable is a running service with an uptime target, not a paper or a defended result.
The employers hiring for this title look like consultancies and product teams fitting AI to something that already exists. Live listings here carry the title from Doctolib, bunq, Databricks, ICEYE, AssemblyAI and Clarity AI, companies shipping AI features into an existing product rather than training a foundation model from scratch. LinkedIn’s Jobs on the Rise 2026 notes that employers list this title interchangeably with machine learning engineer, and names LangChain, RAG and PyTorch as its most common associated skills, so the boundary with the model-training side of the field is porous in practice.
AI research scientist vs AI engineer at a glance
| Dimension | AI research scientist | AI engineer |
|---|---|---|
| Core deliverable | A new method, published or defended | A running product built on an existing model |
| Typical tools | PyTorch, JAX, custom evaluation harnesses | LLM APIs, RAG, agent frameworks |
| Typical employer | Frontier labs, product companies with a research arm | Consultancies and product teams adding AI |
| Usual way in | A PhD or a strong publication record | A shipped LLM feature, no degree required |
| Closest to | Academic research | Software engineering |
Is an AI research scientist or an AI engineer role more in demand in Europe?
An AI engineer role is in far higher demand across Europe right now than an AI research scientist role, because the applied, product-facing side of the field is where employers outside a handful of labs are actually hiring.
LinkedIn’s Jobs on the Rise 2026 ranks AI Engineer as the single fastest-growing job title in the US for the year, while AI/ML Researcher sits fifth on the same list, both inside the top five but with AI Engineer clearly ahead. The same report’s European edition ranks AI Engineer first or second in every market it covers, and does not surface a dedicated research title in any of them, which is consistent with research roles concentrating at a small number of labs rather than spreading across the wider hiring market the way engineering roles do.
The board’s own category split points the same way. In our analysis of 500+ live AI roles from 188 companies hiring across Europe, 179 sit inside AI and ML engineering, against 59 inside data science and research, the category research scientist, applied scientist and member-of-technical-staff openings fall under. Engineering roles outnumber them by more than three to one, and that gap is the honest picture of where the volume is, even though the research seats that do exist tend to be the most visible and most discussed.
Do the two roles need different education or backgrounds?
An AI research scientist role and an AI engineer role expect genuinely different backgrounds, with the research side leaning far more heavily on formal education. The US Bureau of Labor Statistics lists a master’s degree as the typical entry-level education for computer and information research scientists, the closest official occupation to this title, with a median annual wage of $140,910 in May 2024 and 20% projected employment growth between 2024 and 2034, well above the average for all occupations. Frontier labs hiring for research scientist specifically tend to go further and expect a PhD or an equivalent publication record, because the job is to extend what is known rather than apply it.
Where that formal training actually ends up is shifting. Stanford HAI’s 2026 AI Index Report found that the number of new AI PhDs graduating in the US and Canada rose 22% between 2022 and 2024, and that all of that growth went to academia rather than industry.
Industry still absorbed close to two-thirds of new AI PhDs in 2024, so a doctorate remains a viable route into a research scientist role at a company rather than a lab-only path. But the direction of travel matters for anyone weighing the years a PhD takes: the fastest-growing destination for that degree right now is academia, not the industry research seats this article is about.
An AI engineer role asks for almost none of that. What gets someone hired is a shipped feature built on an LLM and a clear account of what broke and what was done about it, not a transcript.
Can you move between the two roles?
Moving between an AI research scientist role and an AI engineer role is possible in both directions, and it happens more often from research into engineering than the reverse. A researcher who has trained and evaluated models has already done the harder version of what an AI engineer’s tooling work asks for, so the jump mostly requires picking up production skills, serving, monitoring, and working inside somebody else’s product constraints.
The other direction is rarer because the research role’s entry bar is a specific one. Indeed’s Hiring Lab found that almost 90% of AI-related job postings in 2025 sat with just 1% of hiring firms, and the research scientist title concentrates inside that sliver even more tightly than AI hiring does overall. An engineer aiming for a research seat without a PhD or a publication record is competing for a small number of openings at a small number of employers, which is a realistic but narrow target rather than a normal next step.
Should you become an AI research scientist or an AI engineer?
Choosing between an AI research scientist role and an AI engineer role comes down to whether the appeal is inventing a method or shipping one, since the two ask for different training and sit in very different sized parts of the market.
Become a research scientist if the work that excites you is the open question itself, and you are willing to spend years in formal study or a strong independent publication record to get there. The seats are genuinely few, concentrated in a handful of frontier labs and research-heavy product companies, and the bar is closer to academic hiring than to a typical tech interview.
Become an AI engineer if you would rather build something that runs today with tools that already exist. It is the far larger market, it does not require a doctorate, and the way in is demonstrated production work rather than a credential. For most people weighing the two, that combination, more openings and a lower entry bar, makes it the realistic starting point even if research stays the longer-term goal.
Start with remote AI engineer jobs and remote AI research jobs. If the product side of the model appeals more, LLM and generative AI roles sit next to it, and the analysis-heavy end of the research path is closer to data science and AI research.