Is data science still a good career in 2026, now that AI writes the code?

Yes, if you mean the part of data science that decides things, and less so if you mean the part that writes the code. AI has made the code cheap; it has not made the judgement cheap. What has changed is the size and shape of the hiring: in our count of 315 AI job ads from July to September 2026, 9 were titled data scientist, against 48 titled AI engineer. The job still exists, it is smaller in these sources, and the data scientists employers do post for are expected to work with AI, not compete with it.

How many data scientist jobs are there compared with AI engineering?

Fewer, on the evidence we have. Our headline sample of 19 September 2026 counted 315 AI and ML postings from Hacker News "Who is hiring?" (July to September), Remotive and Arbeitnow. By title:

A re-run on 29 September over a wider fetch (3,610 postings, 417 of them AI or ML) found 17 postings titled data scientist. Seventeen is a small number, and we will give counts rather than percentages below so nobody mistakes it for a market.

The caveat matters here more than usual. These sources are startup- and Europe-heavy, Australia barely appears, and established companies that hire data scientists in volume tend to post on LinkedIn and national job boards we did not sample. So read the table as "startups building AI products hire AI engineers far more often than data scientists", not as "data science is over".

What do data scientist job ads ask for now?

Among the 17 data scientist postings on 29 September:

• Python in 12, SQL in 8.
• LLMs named in 6, agents in 5, evals in 5.
• Experimentation or A/B testing in 5.
• Classical ML frameworks were thin: scikit-learn in 3, PyTorch and TensorFlow in 2 each, pandas or NumPy in 3.
• Stakeholders or cross-functional work in 6.
• A degree was mentioned in 4. Seven stated years of experience, with a median of four.

Two readings hold up even at this size. First, Python and SQL are still the floor; no amount of AI assistance has removed them from the ads. Second, 6 of the 17 name LLMs and 5 name evals: the role is being asked to work with language models, not only with tables. The data scientist who can check whether a model's output is right is doing a job the AI engineer also needs done.

The counts file is public, with the method in it. Some Arbeitnow postings are duplicated, which is another reason to treat 17 as indicative only.

Does AI writing the code make data scientists redundant?

It removes the part of the job that was typing. A capable model will now write the pandas, the SQL join and the plotting code in seconds, and it will often be right. What it will not do reliably is:

• Choose the question. Most analyses fail because they answer something nobody needed decided.
• Notice the data is wrong. Duplicated rows, a changed definition, a tracking bug from March. A model will happily compute an average over all of it.
• Get the statistics right under pressure. Peeking at an A/B test, a confounded comparison, a sample too small to say anything. These are judgement errors, and a model that writes fluent code makes them faster, not less often.
• Make a decision maker act. A finding nobody believes has no value. The memo, the dashboard and the conversation are still human work.

So the job has shifted from "person who can write the analysis" to "person who can be trusted with the conclusion". That is a harder bar for a beginner and a better one for anyone who likes the thinking more than the syntax.

Should you train for data science or AI engineering instead?

It depends on whether you want to build products or find answers. We compared the three roles in detail in AI engineer vs ML engineer vs data scientist. The short version:

• If you come from software and want the largest number of openings in these sources, AI engineering is the shorter route to more ads.
• If you come from analysis, finance, operations or research and you enjoy the question more than the build, data science is still the natural fit, and it is the better base for moving into AI work later.
• If you are undecided, learn Python and SQL first. Every one of these paths starts there.

The move from data scientist to AI engineer is common and well trodden. The move from nothing to either is the hard one, so pick the one that matches the work you like and get the first job.

What skills make a data scientist hireable in 2026?

Five, in order:

1. SQL and Python, fluently. Enough to pull data from a system you have never seen and trust the result.
2. Statistics you can defend. Uncertainty, experiments and their traps, and the difference between prediction and explanation.
3. A model you validated honestly. Held-out performance reported as it is, not as you hoped.
4. Communication. A one-page finding, a decision memo, a dashboard a stakeholder uses without you.
5. AI in the loop, checked. Using LLMs for analysis and code, and being able to say where the output could mislead. This is the newest line in the ads and the one most candidates cannot yet show.

Our data scientist role page lists the day-to-day in more detail.

Where do you start this week?

Pick one public dataset in a domain you know and answer one question a manager would care about. Write the SQL yourself, use an AI assistant for the plotting if you like, and then do the part that matters: list three ways your finding could be wrong and check each one. Publish the notebook and a one-page summary with its caveats stated. If that felt hard in the SQL or Python rather than in the thinking, fix the tools first. If it felt hard in the thinking, you have found the job.

The free data science skill check takes about three minutes and tells you which of the five skills above is your gap.

Which Square 1 programme fits?

The Data Science Bootcamp is twelve weeks, live on Zoom with one instructor, about 15 hours a week. It is six blocks, each ending in a deployed project and a gate you must pass: a reproducible finding, an experiment with a decision memo, a validated predictive model deployed as an API, a stakeholder dashboard, an automated reporting pipeline with an AI step, and an employer brief in the final block alongside a hiring sprint. There is a recorded viva in which you defend the analysis and say where the AI could mislead. The entry bar is comfort with spreadsheets; Python is taught from block one. If you want the foundations first, Python for AI and SQL and Data for AI are on-demand courses recorded by an instructor, with the exercises graded by Nova, the AI tutor. All three are taking waitlist places today.