Super Data Science: ML & AI Podcast with Jon Krohn 991: Pair Programming with AI in Your Python Notebook, with Dr. Trevor Manz
35 snips
May 12, 2026 
Dr. Trevor Manz, researcher and engineer behind AnyWidget and Marimo, talks about teaching coding agents to control reactive Python notebooks. He explains Marimo Pair, how skills teach agents to read and run notebook cells, and the promise of recursive language models. He also covers bridging Python and web front ends and why deep curiosity fuels impactful tool building.
AI Snips
Chapters
Books
Transcript
Episode notes
Prefer Declarative Skills Over Imperative Scripts
- When authoring agent skills, prefer declarative descriptions of tools over dogmatic step lists so models can flexibly plan.
- Trevor explains he shifted Marimo Pair from imperative instructions to teaching tool semantics instead of prescriptive steps.
Terminal First Engineer Adopts Agentic Tools
- Trevor codes from the terminal with NeoVim and uses Cloud Code agentic tools for tasks he delegates to agents.
- He kept his editor workflow but adopted agent tools to offload repetitive review and iterating on code changes.
Drive Your Notebook From An Agent Session
- With Marimo Pair you start an agent session (e.g., in Cloud Code) which programmatically starts or connects to a Marimo notebook in the browser.
- The agent then drives the notebook remotely: editing cells, running kernel code, and returning results to you.
