•
AI will not fully replace data analysts. Agents can query and chart, but not decide the question. After-work Python for the job you already have.
AI will not fully replace data analysts. It will hollow out the part of the job that is routine extraction, formatting, and baseline charts. If that is all you do, software is already cheaper. If you own definitions, review what an agent produced, and still have to answer for the number, you are in a scarcer seat.
This is for employed people who want to stay competitive in the job they already have. It is not a bootcamp, not a get-rich plan, and not a subsidised certificate.
In a 17 July 2026 TechTarget opinion, Donald Farmer revisited the question a year and a half after arguing that generative AI would not replace analysts. Model capability has moved. Agents are commonplace. An agent can now query governed tables, write a formulaic summary, and produce a much better chart than in early 2025.
What it still cannot do, Farmer writes, is decide which question to ask, or which answer to trust. That needs business context, doubt, and a person who will be held to the number. Software can retrieve brilliantly and still be blind to the financial, ethical, and strategic consequences of its own output.
A marketing example from the same piece: an agent reports that paid search drove last quarter's revenue and recommends a bigger budget. The arithmetic can be flawless on the model's terms. A human who knows the brand campaign and the price cut can still see that last-click attribution is taking credit for demand that was already there. The agent does not know to reopen the definition.
Farmer's verdict is not "analysts are safe" or "analysts are finished." It depends on the type of analyst.
People who spend the day on routine extraction, layout, and baseline charts are the ones whose work software already does well. People who maintain a semantic layer — the shared definitions that keep humans and models talking about the same metrics — who audit agents, and who can take responsibility for numbers, remain scarce. Chief data officers, he notes, cannot hire enough of those people.
That split is the after-work problem. You do not need a new job title. You need to stop being only the person who pulls the dump and makes it look tidy.
This is not a Hong Kong salary figure, not a count of Hong Kong job ads, and not a local pay claim.
365 Data Science's 28 April 2026 job-market note analysed a sample of US data-scientist postings (they started from 1,000 Monster Jobs ads and narrowed to 827 relevant ones). In that US job-ad sample, Python was explicitly mentioned in 57% of 2025 data-scientist offers, down from 78% in 2024. The language still appears in a majority of that sample. Treat it as a US hiring-ad snapshot, not as a Hong Kong wage table and not as a ranking of courses.
The same note is not a reason to quit your job and rebrand as a data scientist. It is a reason not to assume Python has vanished from the work next to AI tools.
The World Economic Forum's Future of Jobs Report 2025 (published 7 January 2025) puts AI and big data among the fastest-growing skills over the next five years, alongside other technology skills such as networks and cybersecurity and technological literacy. That is an employer-survey outlook, not a promise that any particular course will place you.
For someone already in an analyst or analyst-adjacent seat, the practical reading is narrow: the scarce work is judging automated output, not only formatting it. After-work Python is one way to keep a file, a definition, and a chart you can defend — instead of only pasting what an agent returned.
Stay in the job you have. Learn enough Python to read a file, trust a number, and reproduce a chart without waiting for a chatbot to remember last week's prompt.
A few evenings a week is the format. We have already written Learn Python after work: 19 short lessons, a few evenings a week. If the interpreter is not on your machine yet, Set up Python after work in 30 minutes (Windows, Mac, VS Code). The first useful artefact is not another print: Your first Python program that uses a library (not another print). If your daytime work is still Excel, Make one chart tonight: from Excel to a Python file is the evening that moves one series into a .py you own.
That path is literacy for employed people. It is not a career switch, not a trading income, and not a subsidised certificate.
A later data-analysis path is an outline we are building. It is not 39 live lessons today. Do not enroll expecting that catalogue to already exist as a finished course. We also will not ask you to enroll in unpublished quant or macro material from this page.
If you want short after-work lessons for the job you already sit in, open Introduction to Python Programming. It currently has 19 published lessons. Check that course page for the current enrollment window — windows move, and this article will not freeze a date into the title or the meta description.
Do one sitting on an evening you actually have. The point is still to be useful at the desk you return to in the morning — including when an agent has already drafted the chart.