Скачать или смотреть Rust Machine Learning AI Framework with Google Jules

Today we're using Google Jules, which is a coding agent from Google that's actually been around for more than two years, even though I did not know about it until now. Other coding agents like Codex from OpenAI or Claude from Anthropic came out later and got a lot of attention, but Jules has been quietly doing its thing for longer. Basically, Jules runs in your web browser, links to your GitHub repository through a simple sign-in, and then does coding work for you in containers on the cloud, kind of like Cloud Code, but even more streamlined.

Right now, we have an empty repo, and I’m going to ask Jules to write a simple deep learning framework in Rust, specifically one that can handle an XOR training example, which is a common hello world for machine learning because the solution is not linear and needs a deeper network to solve. We want features like stochastic gradient descent, a few activation functions, and a straightforward structure. While the much-loved MNIST dataset is a typical place to start, I prefer XOR since it is small, synthetic, and you know exactly what you should get.

The basic difference between standard machine learning and deep learning is actually just the number of layers: deep learning adds more layers, not some wild magic, and all the math is just working with arrays of floating-point numbers, weights and biases that get adjusted as training goes on. After Jules does its thing (which is impressively fast, I saw it finish a whole Rust project for us, including model setup, various activation functions, loss calculation, and even a working XOR demo in only about eight minutes), we’ll review what it wrote and maybe tweak it a bit, since I have some ideas about how I like to structure code, especially in Rust where you can overload operators for slick matrix math. If you’re curious about AI, machine learning, or even just coding faster, it looks like Jules is worth checking out, especially because it does everything in the browser and is free for small projects or trial use.

I think it’s wild that this tool has been hiding in plain sight, Google really was ahead of the game here!