The author discusses the limitations of traditional notebooks like Jupyter, which create hidden state, leading to issues with reproducibility. To address this, they present marimo, a new approach where each notebook is modeled as a directed acyclic graph (DAG) to control the flow of variables. This allows for reproducibility and maintainability, as marimo notebooks are pure Python files that can also be executed as scripts or web apps. The author emphasizes the importance of simplicity, extensibility, and staying true to project pillars when designing systems. Although rejecting certain feature requests, they consider feedback to improve marimo’s efficiency while retaining its core principles.
https://marimo.io/blog/lessons-learned