You skipped 7 steps (and Claude can’t fix it)

Hi John,

You take one look at the staircase. Eight steps.

Nah... one giant step to Claude.

Congratulations. You’re an AI Engineer. (lol)

The crazy thing? A lot of people are doing exactly this right now. And it looks like it works.

Until the data changes.
Or the agent invents an answer.
Or the app runs on your laptop and breaks the moment someone else uses it.

Every one of those failures lives on a step you skipped.

The data changed? That’s Step 02, Data Engineering.

The agent made something up? That’s Step 06, AI Applications: nothing checking what it can and can’t claim.

It broke for someone else? That’s Steps 05 and 08, Deployment and Infrastructure.

You still need to know what to ask for, what to check, and what to fix.

Claude can write the code. Your name is on the result.

Claude isn’t the top step.

It’s the handrail.

Use it to climb faster. Don’t expect it to climb for you.

Why this is good news for data scientists

If you’ve built in Python, SQL, and machine learning, you’re already standing on Steps 01 through 03. The people leaping straight to Claude aren’t.

You know what a slick AI demo hides: messy data, misleading results, and the gap between a prediction and a decision someone can act on.

But most data scientists stop at Step 03. They build the analysis, hand it off, and wait for someone else to turn it into action.

The AI Engineer role is everything above that. You’re closer to it than you think.

How to climb: One project, five stages

Take one business question:

“Which customers should our team follow up with this week, and why?”

Here’s how I’d build the answer, one step at a time.

1. Define what the team needs.

Produce a list of customers to review, the reason each one needs attention, and a suggested follow-up. Decide who will use it and what they’ll do next.

If you skip it: You spend weeks building an impressive agent nobody has a reason to use.

2. Build the data and analysis underneath. (Steps 01–03)

Use Python and SQL to prepare customer histories, check missing values, and calculate useful signals. If a predictive model helps, add it and check its performance. This is familiar territory.

If you skip it: The agent confidently explains numbers that were wrong before the LLM ever saw them.

3. Give the agent a specific job. (Step 06)

Have it use your calculated results to explain which customers stand out and to draft follow-ups for a person to review. Keep the underlying numbers next to every recommendation.

If you skip it: You get fluent answers nobody can check.

Now you have Data → Analysis → AI explanation → Suggested action. This is where your data science work becomes an AI system.

Want to watch these steps come together in a live Python build?

On Wednesday, November 11th @ 2:00 PM Eastern, I’m hosting a Free AI + Data Science workshop.

I’ll build an AI/DS Copilot LIVE, show how the data analysis connects to decision-ready next actions, and explain the skills behind the workflow.

👉 Watch the AI + Data Science workflow come together live—reserve your free seat.

4. Give someone a way to use it. (Steps 06–07)

Build a simple app or API so a user can run the workflow, inspect the results, and review the suggested follow-ups.

If you skip it: Every request still comes back to you: “Can you open the notebook and run that again?”

5. Test it, deploy it, and learn from the failures. (Steps 05 & 08)

Try missing data, an empty customer list, a failed tool call, and an answer the agent can’t support. Decide how the app should respond to each.

Then deploy it, protect the credentials, and log enough to investigate problems.

If you skip it: You’re back at the top of this email. It works on your laptop and breaks for everyone else.

Notice where Claude shows up: At every stage.

It writes the pipeline, explains an unfamiliar function, suggests test cases, debugs, and packages the app.

You run it, inspect the output, and check that it does what the business needs.

Same tool, two outcomes.

One person leaps and gets their reputation tied to a failed project.

The other climbs with a handrail and gets cheered by their company.

How the AI Bootcamp helps you climb faster

You could spend months piecing this roadmap together yourself.

Or you could shortcut it in 8 weeks:

• Live coding with me in Python.
• Build 3–4 portfolio-ready agents.
• Get the AI Engineering Toolkit: 100+ curated libraries (LangChain, LangGraph, OpenAI, Polars, XGBoost, and more).

My student Lannon landed an AI Engineer role after interviewers were impressed by a project he learned in Clinic 1.

Your next step: Start the roadmap today.

👉 Or shortcut the process and join me in the 8-Week AI Bootcamp.

Enjoy!

Your Friendly Neighborhood Generative AI Data Scientist,

Matt Dancho

P.S. - On November 11th, I have a free AI for Data Scientists workshop that will take your AI skills to the next level. I'll show you how to build an AI Customer Segmentation Agent that automatically segments customers and provides marketing recommendations (Live Python Code). You can register for free here.

P.P.S. - AI Bootcamp Cohort 12 starts February 1st, 2027. Kick off your 2027 with a new AI career with the fastest path from Data Scientist to AI Engineer. Join Cohort 12 here.