Previously I’ve noted my appreciation for marimo notebooks, especially how their reactive cell model was different from Jupyter notebook. Marimo has also been developing an interesting narrative around integration with agentic coding.
In this blog we explain why agentic coding tools like Claude work exceptionally well with marimo, especially when compared to other notebooks such as Jupyter. We also share tips on how to best use Claude when working with marimo. While this blog focuses on Claude Code, you don’t have to use your terminal if you don’t want to: in a future blog post we’ll describe how marimo provides a batteries-included AI-native editor, with a best-in-class experience for working with LLMs and your data in a single development environment
However, Jupyter is an established, robust, and large ecosystem. With a lot of smart people at the forefront of data science and machine learning. So I should have known the Jupyter team would not stand still in the face of AI advances.
Enter Jupyter AI
Welcome to Jupyter AI, which brings generative AI to Jupyter. Jupyter AI provides a user-friendly and powerful way to explore generative AI models in notebooks and improve your productivity in JupyterLab and the Jupyter Notebook. More specifically, Jupyter AI offers:
An %%ai magic that turns the Jupyter notebook into a reproducible generative AI playground. This works anywhere the IPython kernel runs (JupyterLab, Jupyter Notebook, Google Colab, VSCode, etc.).
A native chat UI in JupyterLab that enables you to work with generative AI as a conversational assistant.
Support for a wide range of generative model providers and models (AI21, Anthropic, Cohere, Gemini, Hugging Face, MistralAI, OpenAI, SageMaker, NVIDIA, etc.).
That’s a solid marketing pitch. A few years ago I had started poking at the precursor to Jupyter AI for some work activities. One of me or the framework, maybe both, wasn’t mature enough to intimate all the downstream implications. Also, agentic coding wasn’t really a thing at that stage.
However, I became really smitten when I started working through a course at Deeplearning.AI: Jupyter AI: AI Coding in Notebooks. The course features Brian Granger, one of the originators of the Jupyter project, in addition to Andrew Ng.
About this course
Learn to use Jupyter AI as your notebook coding partner in this short course, taught by Andrew Ng and Brian Granger, co-founder of Project Jupyter.
Coding practices are shifting from manual coding to AI-assisted development, and Jupyter AI allows you to integrate AI coding into all your notebook development workflows. Many AI coding assistants struggle to function well within the notebook environment, the Project Jupyter team has introduced Jupyter AI, which is an open-source framework that deeply integrates AI coding and collaboration into Jupyter notebooks and JupyterLab.
Jupyter AI provides a chat interface that you can use to generate new cells in your notebook. You can also drag existing cells into the chat for debugging, attach files for context, and save chat histories to reuse later as additional context for your work.
Even about two thirds of the way through the course I’ve found Jupyter’s conversational assistant approach to AI integration intriguing. In essence, a user leans on an open AI chat window in parallel with an open notebook. A little bit of UI chrome makes it easy to transfer AI content to cells and cells into AI prompts. It’s not as silky smooth as I would like, but the notion of working on a notebook in “immediate mode” as a computational artifact feels promising as a conceptual model. I’ll have to cross check, but this seems like a distinctly different style from the marimo way of pointing the AI at the code underlying the notebook implementation. For marimo that makes total sense since their notebooks are just Python code. Thankfully, this doesn’t have to be a competition. Let a thousand flowers bloom.
Personally, I’d like to see Jupyter AI ease the path of having the AI
manipulate the notebook from natural language prompts. Currently you
have to click buttons to do many of the content exchanges between the
two sides. Since a notation for referring to cells already exists,
courtesy of IPython, maybe a handful of tools could be added to make
this easier. I wonder how far you could get with read_cell and
write_cell tools, baked in?
So much room for experimentation. Part of the reason I’m skeptically optimistic about AI and coding are examples like this of the exploration space. There’s lots of possible ways to invent the future and lots of folks exploring as they please. Exciting times!