AI for Data Scientists: What to Try, What It Costs, and What Stays Yours

Data scientists rank 98 out of 100 for how much AI can do, and the job is still growing fast. The code is quick now. Asking the right question is the work. Here is what to try, what it costs, and a four-weekend plan. One rule comes first. Keep client and company data out of a personal AI account.

98 of 100rank for how much AI can do in this job. 100 is the most. Microsoft Research.
$120,230typical pay, May 2025. BLS.
275,600jobs in 2025. BLS.
+35%more jobs expected by 2035. BLS.

How Data Scientists Can Use AI Today

A data scientist's week. Higher means you do it more often. Further right means AI can do more of it. Start in the top right.

  • Query and code drafts. The first version of the query in seconds. You check the joins. (AI does it)
  • Data cleanup. The dull part, written for you. (AI does it)
  • Charts. A clear chart from a table. (AI does it)
  • Write-ups. It drafts the summary. You own the claim. (AI does it)
  • Method choice. It lays out the options. You know the data. (AI helps)
  • Checking your analysis. It looks for the hole before your boss does. (AI helps)
  • Model building. It writes the first model. You judge whether it is any good. (AI helps)
  • Paper summaries. It reads the paper. You decide if it applies. (AI helps)
  • Asking the right question. The right question is most of the work. (You do it)
  • Knowing the data. You know which column lies. (You do it)
  • Saying what it means. The number is the easy part. (You do it)
  • Saying no to a bad claim. Sometimes the data doesn't support the story. (You do it)

Where this comes from. Where each task sits is my estimate, guided by Microsoft Research's study of how people use AI at work.

12 AI Prompts for Data Scientists

Copy any of these into ChatGPT, Claude, Gemini, or Grok.

  1. Query and code drafts. Here are my table definitions. Write a query for monthly active users by plan for the last year. Explain each join.
  2. Data cleanup. Write Python that reads this CSV, fixes the date column, drops repeats, and reports how many rows it removed and why.
  3. Charts. Here is a table of signups by week and channel. Write the code for the clearest chart and one sentence on what it shows.
  4. Write-ups. Here are my results. Write a one-page summary for a leader who has two minutes. Say what we found, how sure we are, and what to do.
  5. Method choice. I want to know if a new feature raised retention, and I have no test group. Give me three ways to estimate it with the weak spots of each.
  6. Checking your analysis. Here is my analysis plan. List five ways the result could be wrong or misleading.
  7. Model building. Write a simple model that predicts churn from these columns. Hold out test data and report how well it does.
  8. Paper summaries. Sum up this paper in ten lines. Say what they did, what they found, and what would stop it from working on my data.
  9. Asking the right question. Ask me ten questions that turn a vague request from my boss into something I can measure.
  10. Knowing the data. Give me a checklist for the first hour with a table I have never seen.
  11. Saying what it means. Play a doubtful VP and question my result.
  12. Saying no to a bad claim. Help me tell a leader the data does not back up their claim. Ask me questions first.

Will AI Replace Data Scientists?

Data Scientists rank 98 out of 100 for how much AI can do in the job. 100 is the most. The government expects 35% more jobs by 2035.

  • Roofers: 2 out of 100 for how much AI can do
  • Carpenters: 15 out of 100 for how much AI can do
  • Plumbers: 22 out of 100 for how much AI can do
  • Nurses: 42 out of 100 for how much AI can do
  • Electricians: 53 out of 100 for how much AI can do
  • Accountants: 68 out of 100 for how much AI can do
  • Software developers: 85 out of 100 for how much AI can do
  • Data scientists: 98 out of 100 for how much AI can do

AI Tools for Data Scientists and What They Cost

Switch things on and off and watch the total. Prices checked November 19, 2026. No affiliate links.

Not ready to pay? See every free way in →

  • One paid assistant ($20 a month). ChatGPT Plus and Claude Pro list at $20, and Google AI Pro lists at $19.99. Grok Premium lists at $30, with a Plus tier at $8. Pick one and use it by voice and camera.
  • Your phone's microphone and camera ($0). Talk instead of typing. Take a photo when words are slow.
  • What your employer already provides ($0). Ask which AI tools are approved at work. Anything with client and company data belongs there.
  • A notes folder you own ($0). Your checklists, templates, and study notes in one place the assistant can read.
  • A second assistant ($20 a month). Run the same question through two models when the answer matters. Leave it off until you need it.

How to Start Using AI as a Data Scientist in Four Weekends

Using it once or twice and failing doesn't count. It takes practice and it takes time.

You can do weekend 1 for $0. See every free way in →

Weekend 1: Talk to It ($0 to $20)

This weekend, try query and code drafts. Here are my table definitions. Write a query for monthly active users by plan for the last year. Explain each join.

  1. Put one assistant on your phone's home screen. A free plan works for this weekend.
  2. Say the prompt above out loud with made-up details.
  3. Ask it something you already know the answer to and see if it gets it right.
  4. Do ten things from your own week before Sunday night.

By Sunday night. You know what it is good at and where it gets things wrong. From my post Agency Beats Intelligence.

Weekend 2: Give It Your Work ($0)

This weekend, try data cleanup. Write Python that reads this CSV, fixes the date column, drops repeats, and reports how many rows it removed and why.

  1. Start one project in the assistant for your work. Keep client and company data out of a personal account.
  2. Add your table definitions, your code style, past analyses, and how your team likes results shown.
  3. Tell it to interview you. Ask for 25 questions about how you work and answer them out loud.
  4. Run the prompt above again. The answer should fit you this time.

By Sunday night. The assistant knows how you work, so its answers fit. From my post The Second Brain Finally Works.

Weekend 3: Save What Works ($0)

This weekend, try checking your analysis. Here is my analysis plan. List five ways the result could be wrong or misleading.

  1. Run the prompt above. Fix what it got wrong and tell it why.
  2. Save it as its own instruction so you never type it twice.
  3. Save the prompts from weekends 1 and 2 the same way.
  4. Use all three for a week and fix what annoys you.

By Sunday night. Three saved helpers you use every week. From my post How to Choose AI Projects Wisely.

Weekend 4: Make It a Habit ($0)

This weekend, try method choice. I want to know if a new feature raised retention, and I have no test group. Give me three ways to estimate it with the weak spots of each.

  1. Run the prompt above. Keep client and company data out of a personal account.
  2. Pick the helper that saved you the most time. Use it every day this week.
  3. Show one coworker what you built and where it got things wrong.
  4. Write down what you will never hand to AI, and why.

By Sunday night. AI is part of your week, and you know where your line is. From my post The Broken Pool Cue.

Check its work every time on every number, join, and claim. It will be wrong sometimes and it will sound sure when it is. I wrote about why in The 1% Problem of AI Errors.

My Read on AI for Data Scientists

Being entrepreneurial is a mindset, and that agency and urgency mindset is what's really going to win the day. It does require trying these things out, and using it once or twice and failing doesn't count. If you have kids, you know what I'm talking about. They can't throw, can't catch, can't bat. At first it takes practice and it takes time. This is a good starting place, with the guide as a coach.

I've spent my career in data, and what I keep running into is missing context. I wrote in Bigger Models Can't Fix What Nobody Wrote Down that a bigger model can't fix what nobody wrote down.

So I think a data scientist's value moves to knowing which column lies and what question the boss meant to ask. The query takes seconds now. I'd let it write the code, and I'd spend my time on whether the answer is true.

Questions Data Scientists Ask About AI

Will AI replace data scientists?

A lot of the work will change. Data Scientists rank 98 out of 100 for how much AI can do. The government expects 35% more jobs by 2035.

What can data scientists use AI for today?

Start with query and code drafts, data cleanup, and charts. Then try method choice and checking your analysis, and check the work yourself. Keep client and company data out of a personal AI account.

What does it cost to start?

About $20 a month for one paid assistant. You can start for $0 on a free plan.

How long does it take to get started?

Four weekends. Each one has one task to try and four steps, and the page keeps track of what you finish.

What are good AI prompts for data scientists?

This page has 12 you can copy. Start with query and code drafts, data cleanup, and charts.

Can data scientists start using AI for free?

Yes. The free plans from ChatGPT, Claude, Gemini, Grok, and Copilot are enough for the first weekend. A paid plan at about $20 a month lets you do more.

Sources and Method

What the research says

Microsoft studied how people use its AI helper, Copilot. Then it scored 785 jobs on how much of each job AI can do. Data Scientists rank 98 out of 100, where 100 is the most.

Where each task sits on the map is my own estimate. I placed each one by how often it comes up and how much of it an AI assistant can do today.

The government counted 275,600 jobs for data scientists in 2025. Half of them earned more than $120,230 a year. It expects 35% more jobs by 2035.

Sources
Method and limits

The rank is where the job falls among the 785 jobs Microsoft scored. Where each task sits on the map is my best estimate, guided by Microsoft's data. A high rank means AI can do more of the tasks. It does not mean the job goes away. The 'where it's heading' view is a guess at direction, with no date on it. Last checked November 19, 2026.

Know a data scientist?

Send them this. It takes four weekends, and they can start for $0.

AI Guides for Other Careers and Majors