Train for the AI engineer role unless you have a specific reason not to. It is the one being hired for: in our count of 315 AI job ads from July to September 2026, 48 postings were titled AI engineer and a further 62 were software-engineer roles at companies building AI products, against 13 for machine learning engineer and 9 for data scientist. The three jobs are genuinely different — different daily work, different maths, different interview — and the differences below are what should decide it. But the market has moved, and the default in 2026 is the role that builds with models rather than the one that builds them.
What does each role actually do all day?
An AI engineer builds software that has a model in it. The day is API calls, retrieval, tool use, prompts under version control, evaluation harnesses, latency and cost, and the same deploy-and-monitor loop as any backend engineer. The model is usually someone else's — Anthropic's, OpenAI's, Google's, or an open-weights model behind an inference server. Success is a feature that works for users and a number that proves it.
A machine learning engineer builds and ships models. The day is data pipelines, training runs, feature engineering, experiment tracking, model serving, and the infrastructure that keeps a model retraining on fresh data. In 2026 this role has split: at the large labs and a few well-funded companies it means training and fine-tuning large models; everywhere else it means classical ML on tabular and time-series data — fraud, forecasting, ranking, pricing — where a gradient-boosted tree still beats an LLM on cost and accuracy.
A data scientist answers questions with data. The day is SQL, notebooks, statistics, experiments and A/B tests, and the presentation to the person who has to decide something. Models appear as tools for the analysis rather than as products. In many companies the data scientist is closer to the business than either engineer, and the best ones are as much analysts and communicators as modellers.
Which one is hiring?
The AI engineer, and the software engineer with models — by a wide margin, on the evidence of what employers post. The skills those ads name are consistent with the job: Python 38%, TypeScript 19%, agents 39%, evals 23%, RAG 11%, AWS 16%, Docker 10%, Postgres 10%. PyTorch appears in 5% of ads and TensorFlow in 3%, which is the ML-engineer stack's footprint in a sample dominated by AI-product hiring.
That does not mean ML-engineer and data-science roles are gone. It means they are hired for through different channels — more often via LinkedIn and Seek at established companies than in the startup-heavy sources we counted — and in smaller numbers. It also means the entry bar for those two is higher: they are more likely to ask for a degree, and the two ads in our sample that wanted a research scientist wanted a PhD.
Which one suits me?
Answer three questions honestly:
Do you want to build products or find answers? Products → AI engineer or ML engineer. Answers → data scientist. This is the biggest split and people get it wrong by choosing the fashionable title over the work they actually like.
Do you enjoy the maths, or do you tolerate it? Enjoy it → ML engineer or data scientist, and the research path is open to you. Tolerate it → AI engineer. The AI engineer needs an intuition for probability and enough statistics to read an eval; the ML engineer needs linear algebra and optimisation; the data scientist needs experimental design and inference.
Are you coming from software or from analysis? Software → AI engineer is the shortest bridge; ML engineer is the second. Analysis (SQL, Excel, a bit of Python, a business domain) → data scientist first, then AI engineer if the product side pulls you.
What does each path take to become hireable?
- AI engineer: six to nine months for a working programmer, twelve to eighteen from zero. Three deployed projects, an eval set, a container and a deploy pipeline. How long it takes to learn machine learning well enough to get hired has the week-by-week. The AI Engineer role page lists the modules that map to it.
- ML engineer: nine to eighteen months, and a degree helps more here than for the AI-engineer role. Two classical ML projects on real tabular data with honest validation, one fine-tuning project, and the serving infrastructure for at least one of them. The Machine Learning Engineer role page and the ML Engineer bootcamp are the track.
- Data scientist: six to twelve months for an analyst, and the portfolio is analyses rather than apps — a decision made better because of your work, documented end to end. The Data Scientist role page and the Data Science bootcamp cover it.
Can I move between them later?
Yes, and people do — the boundaries are porous and the first job matters less than people fear. AI engineer to ML engineer is a maths-and-infrastructure step. Data scientist to AI engineer is a software-engineering step, and the most common move we see described in ads: companies want the analyst who can now also ship the thing. ML engineer to AI engineer is the easiest of all and is happening at scale as classical-ML teams pick up LLM work.
What does not move easily is nothing to any of the three. Pick one, build three things, get the first job, and then decide.
The short answer
Most people reading this should train as an AI engineer: it is the role being hired for, the shortest path from software or adjacent work, and the one whose skills — agents, evals, retrieval, deployment — are the ones the ads name. Train as an ML engineer if you enjoy the maths and want to build models rather than use them. Train as a data scientist if you would rather find the answer than build the product, and you are coming from analysis.
The free skill checks are one per subject and take three minutes; taking the Generative AI, Machine Learning and Data Science ones back to back is a fast way to feel the difference.