Why Python for AI?

A strange fact

Python is a slow language, and its own documentation says so. Yet almost all of AI is built with it. At most of my workshops, students ask me about this.

• Is AI Python?
• Is Python AI?
• Why not use the other languages we already have, for AI?
• What is so special about Python?

These are good questions, and there is a puzzle inside them. Languages like C and Java run faster than Python, while AI runs billions of calculations.

It reads like English

Let's start with the person, not the computer. Say you want the computer to print one word: Hello.

Both programs do the same job, but most of the Java lines have nothing to do with hello. Try the Python one: press Run, change the word on the line marked # edit here, and run it again.

Why does that matter for AI? Because AI work is not written from fixed rules. You start with an assumption, run an experiment, look at the results and build on them. Then you go round again, many times.

Python reads close to plain English, so you can change one line and run it again straight away. Every try is quick, and in AI, quick tries matter most.

Who does the heavy work?

That still leaves our puzzle: if Python is slow, who does all the heavy work?

Say you ride a scooter to college. You turn the key and twist the accelerator, and that is all you do. Underneath, the engine does the hard work, and you never touch it. You still reach on time.

Python is your hand on the accelerator. You twist it, and an engine of fast code, written in C and C++, does the work. On graphics cards, that engine is CUDA. The heavy maths runs there, not in Python.

Fast code packed so that Python can call it is what we call a library. NumPy and PyTorch are two of them: NumPy's work is done by pre-compiled C code, and most of PyTorch is written in C++.

Let's time it

Say a shop has 10 million bills, and you want price times quantity, all added up. I asked an AI to write this two ways: a plain Python loop, and the same job handed to NumPy.

I ran both on my own laptop, using only the processor, with no graphics card.

Both gave the same answer, but NumPy was about 100 times faster, because the work moved out of Python and into C. So yes, Python is slow, but AI in Python is not.

What you should see: "same answer: True", and NumPy many times faster than the loop. This page runs a smaller Python inside your browser, so your seconds will differ from my laptop's. Look at the gap, not the seconds.

How it became the language of AI

Python was not built for AI. Guido van Rossum started it in 1989, in the Netherlands, as a simple everyday language. And no, it is not named after the snake: it is named after a comedy show, Monty Python.

Then the tools for AI arrived, one after another.

Each new tool made the next one easier to build, so more people came and built still more tools. Many tools, together with the people who build them, form what we call an ecosystem.

The numbers

So where does Python stand today? Here are the numbers, checked in October 2026.

But is Python the most used language in the world? That depends on who you ask. The TIOBE index, which counts web searches, puts it at number 1. Stack Overflow's developer survey puts it at number 4, and GitHub at number 2.

The honest cost

That is the good side. Now, what does Python cost?

First, when you write your own loops over big data, you feel the slowness: about 100 times, as the timing showed.

Second, Python is not the answer for every job. Android apps are mostly written in Kotlin, and websites in your browser run on JavaScript.

From my own work

I have lived this myself. In 2019, I worked on a deep learning project in the automotive world. Cars with driver assistance, called ADAS, already had cameras as their eyes, and we were building their ears.

When you drive, you don't only see the things that need your attention; you also hear them. A horn is a warning, and a siren means an emergency for someone.

We trained the model in Python. But to run it inside the car's controller, we wrote the libraries underneath in C and flashed them onto the chip.

So, is AI Python? No. AI itself does not belong to any one language. Python wins for other reasons:

1. It is easy to read, and easy to learn.
2. It is open source.
3. It has libraries for almost everything.
4. A huge community keeps helping each other.

So use Python to build and try AI, and other tools where the job needs them.

When AI writes the code

One last question: if AI writes the code anyway, does the language even matter?

In this year's Stack Overflow survey, Python came first as the language people use AI tools with, just ahead of JavaScript. And the best-known tests of AI coding, such as SWE-bench, are in Python.

So very often, the code an AI hands you will be Python. Someone has to read it, check it, and fix it when it breaks. That someone is you.

The takeaway

1. Python is not chosen for speed. It only looks fast, because engines written in C run under its hood.
2. It wins because it is easy to read and quick to try, and AI work is many quick tries.
3. AI does not belong to one language: Python to build and try, other tools where the job needs them.

That leaves you with a question. When an AI hands you a Python program, can you read it?

Related