Thoughts on Python vs Rust | Chirag Patil
Rust-red Mars seen from orbit, with dark canyons and a sunlit planetary rim.

3 min read

Thoughts on Python vs Rust

Chirag Patil · Python, Rust, and AI agents

TL;DR: If you no longer write most of your code yourself, why not use a language that gives you much better performance?

This is specifically about how I see things right now, based on my own experience.

My first programming language was C, and I hated it. At the time, it was simply too hard for me.

Then I was introduced to Python. It was dead simple to learn, and it became my language of choice for years.

But my workflow has changed.

I now use AI agents to write most of my code. I barely write code myself anymore, and honestly, I barely read all of it either. Instead, I rely heavily on tests and actually using the application to verify that it works.

I maintain strict 100% test coverage, along with integration tests. I understand the architecture of what I am building, and I know how the system works at a higher level. In a sense, I have abstracted away the code itself.

That made me wonder: if an agent is writing the code anyway, why am I asking it to write Python? Why not ask it to write Rust and get the performance benefits?

So I ran an experiment.

I had a personal AI agent called Blacki, written in Python using Google ADK. I migrated the same idea to Rust using Rig and a custom harness. That project became Athena.

The memory difference was ridiculous.

Blacki used roughly 500 MB of RAM on my VM.

Athena uses around 12 MB in production.

That is more than a 40x reduction in memory usage.

Now I can comfortably run production and staging deployments on the same small server. Tell me that does NOT make my 4 GB Hetzner VM very happy.

So I am now on a bit of a Rust spree.

A common pattern in software companies has been to build something in Python first because development is fast, then rewrite performance-sensitive parts in Rust, C++, or another lower-level language once the product stabilizes.

But "development speed" used to mean human development speed.

As of October 2026, for my workflow, development speed increasingly means the throughput at which an AI agent can understand a task, write the code, run the tests, fix its mistakes, and produce a working PR.

If the human is no longer paying most of the complexity cost of using a harder language, the trade-off between Python and Rust starts to look very different.