AI Career Roles Compared: Who Does What?
Nine AI and data roles side by side, from data analyst to AI governance: what each does day to day, the skills involved and how to pick one.
“I want to work in AI” covers dozens of jobs that look very different from one another. A data analyst’s day has little in common with a research scientist’s, and an MLOps engineer’s week looks nothing like that of an AI governance specialist. Job titles also vary from one organisation to the next, which adds to the confusion. In this post I put the nine roles I come across most often side by side and describe, in plain terms, what each one does day to day and which skills it calls for. These descriptions reflect general patterns; every organisation draws its own boundaries.
A map of the roles
It helps to picture these roles along a line. At one end, data is collected and organised; in the middle, it is analysed and modelled; at the other end, models are deployed, turned into products and overseen. Research sits slightly off that line, on the side of inventing new methods.
| Role | Core of the daily work | Key skills |
|---|---|---|
| Data analyst | Answering questions with data; building reports and dashboards | SQL, spreadsheets, visualisation, basic statistics, business knowledge |
| Data scientist | Extracting insight from data; building and evaluating predictive models | Statistics, Python/R, SQL, experiment design, communication |
| Data engineer | Making sure data is collected, moved and stored reliably | Data pipelines, databases, distributed systems, data modelling |
| Machine learning engineer | Turning models into components that run inside real products, at scale | Software engineering, ML libraries, performance optimisation |
| MLOps engineer | Automating training, deployment, monitoring and retraining | CI/CD, containers, cloud infrastructure, monitoring |
| AI / LLM engineer | Integrating large language models into applications; building retrieval (RAG) systems | API integration, evaluation, prompt design, safety |
| Research scientist | Developing new methods, running experiments, publishing | Advanced maths, deep learning, experiment design, academic writing |
| AI product manager | Deciding which problem to solve, how to measure success and what comes first | Product management, user research, ML literacy, communication |
| AI governance specialist | Running risk assessment, policy, regulatory compliance and audit processes | Risk management, legal and regulatory knowledge, ML evaluation, documentation |
Close to the data: analyst, scientist, engineer
A data analyst usually works on “what happened last quarter, and why?” Much of the day goes into SQL queries, reports and meetings with decision-makers. It is one of the most accessible ways into the field.
A data scientist adds “what happens next?” and “what if we change this?” The U.S. Bureau of Labor Statistics (BLS) describes the occupation as covering data collection and cleaning, building and testing models, visualisation, and making recommendations to decision-makers.
A data engineer is the unsung hero. They make sure the data analysts and scientists rely on arrives on time, complete and consistent. Most data quality problems are either solved or amplified at this layer. In my experience, a good data engineer shapes a project’s fate at least as much as the model does.
Getting models into production: ML engineer, MLOps, LLM engineer
There is a big gap between a model that runs on a laptop and one that reliably serves millions of users.
A machine learning engineer closes that gap, making the model fast, scalable and maintainable. Here software engineering skills matter at least as much as statistics.
An MLOps engineer focuses on process. Google Cloud’s architecture guide describes MLOps as an engineering culture and practice that unifies ML system development and operation. To the familiar software ideas of continuous integration and continuous delivery it adds a step specific to ML: continuous training, where models are retrained automatically when new data arrives or performance drops.
The AI or LLM engineer title is newer. These engineers often do not train models from scratch; they connect existing large language models to products, build question-answering systems over an organisation’s documents and, crucially, measure how accurate and safe the outputs are. Evaluation skills matter far more in this role than many people expect.
The research side
A research scientist develops new methods, designs experiments and publishes results. According to the BLS description of computer and information research scientists, these roles typically require at least a master’s degree, and some employers prefer a PhD. The days are spent reading papers, setting up experiments, analysing results and writing. Tolerance for uncertainty and long timelines is one of the least discussed but most decisive traits for this role.
Product and governance
An AI product manager turns “can we do this with AI?” into “should we, how will we measure success, and what happens when it gets things wrong?” They may not write code, but they need to understand what models do well and where they fail.
The AI governance specialist role is one I follow closely. People in it identify the risks of an AI system, write organisational policy, track regulatory compliance and make sure systems are documented, tested and monitored. The U.S. National Institute of Standards and Technology (NIST) released its AI Risk Management Framework in January 2023, built around four functions: govern, map, measure and manage. It is a good starting point for understanding what the job involves day to day. There are also professional certifications in the area, such as the IAPP’s AIGP. Because the role bridges law, technology and ethics, it is open to people from many backgrounds.
Moving between roles
These roles are not separated by hard lines. Plenty of people move from analytics to data science, from data science to ML engineering, or from ML engineering into governance. What they share is the same foundation: understanding where data comes from, and measuring honestly whether a system really works. Investing in those two skills pays off whichever direction you take.
Key takeaways
- "Working in AI" is not one job but a spectrum, from data and research to product and governance.
- Titles vary between organisations, so read the duties in each posting.
- Production roles such as ML engineering and MLOps need strong software skills.
- Governance and product roles are open to people from other disciplines too.
- Data quality and evaluation skills are common ground across almost every role.
Frequently asked questions
Can I work in AI without coding?
Yes, especially in product management and governance. Even there, understanding the basic concepts and the limits of models makes a big difference.
Which role is the best entry point?
Data analysis is generally seen as one of the most accessible. SQL, basic statistics and visualisation are a good place to start.
Do I need a PhD to be a research scientist?
Not always, but postgraduate study is often expected in research roles, and some employers prefer a doctorate.
Will "LLM engineer" still be a job in a few years?
Hard to say. The title may change, but skills such as integrating language models safely into products and evaluating their outputs look set to stay valuable.
Sources
- U.S. Bureau of Labor Statistics – Data Scientists ↗
- U.S. Bureau of Labor Statistics – Computer and Information Research Scientists ↗
- Google Cloud – MLOps: Continuous delivery and automation pipelines in machine learning ↗
- NIST – AI Risk Management Framework ↗
- IAPP – Artificial Intelligence Governance Professional (AIGP) ↗
