Python for AI Engineering: Why Coding Beats Prompting Alone

Published by Kirill Eremenko

November 4, 2026

Tom is a quant analyst who writes Python all day. 

But when it comes to building AI, he does it the same way my non-technical friends do.

He’s 56 and hopes AI skills will keep him gainfully employed until 80. The bigger dream is to build AI tools of his own on the side and leave the 9-to-5 behind.

His method is simple. He writes instructions in markdown files and expects AI to follow them. But as he puts it, “sometimes it goes off on a tangent.”

So he wrote even *more* detailed MD-file instructions. When that didn’t work, he set up a whole team of AI agents, each with its own job and one agent in charge…

…The whole team went off on a tangent together.

That cost Tom a real project when the results didn’t make any sense. 

The AI hadn’t varied the parameters he’d asked for, and the only option was to start over.

The irony is that Tom also teaches data science to graduate students. 

In his machine learning class, AI isn’t allowed, because students need to understand the concepts first. 

Hearing that was a light bulb moment for me 🤔

He makes students learn the foundations before they touch AI. Yet about his own AI work, he’d admitted earlier in the call, “I don’t think I’m interacting with it the way I should be.”

So I was candid with him: 

“Markdown files are made for people who can’t code.”

Tom writes Python every day at his day-job, and Python is the language of AI. It lets you give AI precise instructions and test its answers with Evals, the same way you test normal code with unit tests. 

That’s how mistakes like the one in his project get caught, and markdown files simply can’t do it. 

I told him he was sitting on a huge missed opportunity: he knows Python BUT he doesn’t use it for his AI projects!

Tom didn’t argue. “You caught my attention,” he told me. 

Then came the admission: “So I’m going about it the wrong way, and I knew I was.”

As I told Tom, he’s doing the right thing, just in the wrong way. With the Python he already knows, building AI properly would come easily. The world’s his oyster.

P.S. If you write Python every day but build AI the way non-coders do, hit “Book an appointment” on my profile. I’ll tell you what I’d do in your shoes.

Share on

Related Blogs