Loading lesson…
AI Careers & Paths / Roles, Skills & Growth
Who does what in AI work.
Reviewed by Yuvaraj
Job titles in AI are famously slippery. The same work might be called "ML Engineer" at one company and "AI Engineer" at another, and a role labeled "Data Scientist" can range from writing SQL dashboards to training neural networks. Underneath the noisy titles, though, the actual work clusters into a handful of recognizable roles. This lesson maps those clusters, what each role really does day to day, and where they overlap, so you can read a job description and understand what it is actually asking for.
Start with the shape of the field. The table below maps each role to its core work and the thing it is ultimately judged on, its typical output.
| Role | Core work | Typical output |
|---|---|---|
| ML Engineer | Trains, evaluates, and deploys models; owns the data-to-production pipeline | A model running in production, plus the pipeline that retrains and ships it |
| AI Engineer / LLM Engineer | Builds products on top of existing models, prompting, RAG, agents, API integration | A working feature or app (assistant, search, summarizer) powered by a model API |
| Data Scientist | Analysis, experiments, and metrics; turning messy data into decisions | Reports, dashboards, A/B test results, sometimes a predictive model |
| Research Scientist / Research Engineer | Advances the methods themselves; designs and runs experiments; publishes | Papers, new techniques, open-source implementations |
| Data Engineer | Builds and maintains the data infrastructure everyone else depends on | Reliable pipelines, warehouses, and clean, queryable datasets |
| MLOps / Platform Engineer | Tooling and infrastructure to train, deploy, and monitor models at scale |
Ask about this lesson, or about anything in AI. Answers cite the lessons they draw on.
Finished this lesson?
Mark it complete to earn XP, keep your streak, and schedule a review.
| Deployment systems, monitoring, CI/CD built for models |
| AI Product Manager | Decides what to build and why; bridges users, business, and engineering | Roadmaps, specs, and a prioritized list of problems worth solving |
| AI Safety / Governance | Evaluates risks, tests for harms, sets usage policy | Evaluations, red-team reports, guardrails, and policies |
Notice that no two "typical outputs" are the same. That is the most reliable way to tell roles apart: not the title, but what you are held accountable for producing.
These two titles overlap heavily in job postings, but the day-to-day work is genuinely different. The ML Engineer's center of gravity is the model: getting it trained, evaluated, and reliably served. The AI Engineer's center of gravity is the product: composing existing models into something useful.
Neither is more advanced than the other, they are different jobs. Many people move between them over a career, and small teams often expect one person to do both.
The Data Scientist sits closest to the business question. The work is framing a problem ("are these users about to churn?"), pulling and cleaning the relevant data, running experiments, and communicating what the numbers actually mean. Some data scientists build models; many spend most of their time on analysis, statistics, and clear reporting. The defining skill is turning ambiguous questions into measurable answers that a decision-maker can act on.
Research roles push the methods forward rather than applying settled ones. A Research Scientist typically forms hypotheses, designs experiments, and publishes; a Research Engineer builds the fast, correct infrastructure that makes those experiments possible and often co-authors the results. These roles usually expect deep specialization and, for many scientist positions, a graduate degree, though strong open-source and reproduction work is increasingly a real path in.
Most AI systems are team efforts, and several adjacent roles make the core work possible:
Titles blur; team size decides how much
At a small startup, one "AI Engineer" may do data engineering, model work, deployment, and product thinking all at once. At a large company, each of those becomes a separate specialized role. The same title therefore means very different things depending on org size, always read the responsibilities, not just the heading.
The roles form a rough pipeline: Data Engineers supply the data, Data Scientists and Research folks figure out what is possible, ML and AI Engineers build and ship it, MLOps keeps it running, PMs steer it, and Safety/Governance keeps it responsible. Understanding the whole chain, even for a role you do not hold, makes you far more effective in the one you do.
Common mistakes