The Future of Notebooks - with Akshay Agrawal of Marimo
Latent Space: The AI Engineer Podcast
2025/07/18
The Future of Notebooks - with Akshay Agrawal of Marimo
The Future of Notebooks - with Akshay Agrawal of Marimo
Latent Space: The AI Engineer Podcast
2025/07/18
In this episode, Akshay Agrawal joins host Alessio to discuss Marimo, a next-generation Python notebook platform designed for AI and data science workflows. The conversation explores how Marimo reimagines the traditional notebook experience by introducing reactivity, interactivity, and seamless integration with AI tools, all while maintaining compatibility with existing Python ecosystems.
Marimo is an open-source Python notebook platform that offers reactive execution, built-in UI components, and seamless integration with AI tools. It allows users to build interactive, app-like notebooks that update dynamically based on code or UI changes. The platform supports Jupyter notebook conversion, dependency tracking via integration with the UV Package Manager, and AI-driven code generation. Demonstrations include a machine learning workflow with MNIST, a PS5 controller-based annotation interface, and AI-assisted code creation. Marimo notebooks are stored as pure Python files, enabling execution as scripts or deployment as web apps. MoLab, a cloud-hosted version, provides scalable environments for notebook execution. The team is also exploring AI agents and WebAssembly support, aiming to push the boundaries of traditional notebook-based development while fostering an active open-source community.
00:47
00:47
Marimo is stored as pure Python and supports rapid prototyping
03:40
03:40
Marimo reimagines notebooks by combining interactivity with reproducibility.
04:26
04:26
Marimo notebooks are reactive to UI and code, automatically updating outputs based on interactions.
07:45
07:45
Marimo's reactivity and flexibility allow users to create unique machine learning workflows.
10:08
10:08
Marimo allows generating code like plotting a sine curve using matplotlib with an AI button
11:36
11:36
Marimo's tight integration with UV Package Manager enables better dependency management.
13:06
13:06
Interactive data annotation using a PS5 controller in Marimo for quick model comparison.
18:31
18:31
The AI understands column names and performs data queries faster than using API docs.
19:57
19:57
Marimo can run on a laptop or in WebAssembly through a browser playground.
21:07
21:07
MoLab enables users to use any package and configure CPU, RAM, and data uploads
23:23
23:23
Marimo promotes functional programming and breaks old coding habits
25:43
25:43
Marimo Agents introduces an agent cell that generates Python code from English prompts.
26:45
26:45
The speaker thinks it's great and asks about call to action, such as hiring and seeking more users.
26:56
26:56
Marimo needs a dedicated team to reinvent notebook workflows.
