Why it was never the AI course

(The 90-90/10-10 heuristic — and what’s actually worth learning now)

AI
Datacamp
Agentic scaffolds
learning
Author

Jon Minton

Published

June 5, 2026

Over the last few posts I made the same strange recommendation, three times running. The single most important Datacamp course for working with agentic AI is Introduction to Shell — not because it’s about AI; it isn’t. The second is Introduction to Git. And the third and fourth are SQL and APIs. None of these are badged as about AI, and were taught for decades prior to Agentic AI’s ascendency.

The underlying rationale for these choices is what I call 90-90/10-10. It comes from the 16 week software development course I took in 2023, thinking through Agentic AI changes the value of the course material, and assuming it generalises to software development teaching in general.

90-90/10-10 is shorthand for the following modest proposal:

What’s guiding my suggestions in picking Shell, Git, and SQL and APIs, is that I think they sit in the 10-10 side of the boundary, with the last two being suggestions from Claude itself.

Of course, training in these areas isn’t just something Datacamp offers. Indeed, there’s likely plenty of free material on these areas. Claude suggests the Pro Git book, which is free, and MDN (also free) for learning about the web more generally.

These are the kinds of materials that I’d suggest form the grounding for understanding enough about what Agentic AI does, and how it works, to be confident, competent and safe in using them.

That, in the end, was the whole argument. The most important course for the age of AI isn’t an AI course — it’s the small, surviving core of the un-AI ones.


Series: 1. Shell · 2. Git · 3 & 4. SQL & APIs · the pattern underneath