In my previous full time gig, I did a bit of work implementing a set of APIs using FastAPI and deploying them into AWS. I needed to get a picture of API usage from external partners and AWS didn’t have anything to easily use straight out of the box. So I set about doing some dashboard development and initially thought about using Jupyter but decided to take a sidequest into marimo which had been popping up quite a bit on my podcast radar. Worked like a charm.
Also as part of my job, I built a little NLP model training platform on top of Coiled running in AWS. Highly recommend Coiled if you need to scale compute on AWS but have minimal in-house cloud and ops expertise and staffing. Especially coiled batch, which let us effectively use AWS GPU nodes. We were running super lean and didn’t have time to really drill down on all that AWS had to offer.
And now Coiled has illustrated running marimo notebooks on coiled.
My spider sense tells me marimo is having a moment and building towards escape velocity. It won’t dislodge Jupyter so much as provide a complement in the notebook ecosystem, similar to how polars complements pandas in the dataframe ecosystem.
Here’s some data points about marimo being on my radar. These are all from podcasts I subscribe to and where I consume episodes regularly:
- PythonBytes July 23, 2024
- The Real Python Podcast, November 29, 2024
- PythonBytes March 24, 2025
- Talk Python Podcast April 14, 2025
- The Real Python Podcast, Jun 13, 2025
- The Data Engineering Podcast, July 28, 2025
- The Data Exchange Podcast, August 7, 2025
A Listen Notes search would seem to confirm my intuition that marimo is making a push to increase visibility, possibly due to a round of venture funding at the end of last year. Podcasts aren’t the only place marimo has been popping up. The founder, Akshay Agrawal, did a PyCon US 2025 talk which is available on YouTube, and the tech features prominently in the TalkPython course, “LLM Building Blocks in Python”.
I’ve found Agrawal to be pretty engaging and thoughtful in all of these conversations. He seems to be coming from a pragmatic place of hard won experience. It’s giving me confidence that this project might have legs.
My limited experimentation with marimo has shown promising shoots although it’s a fast moving target. I really like the reactive execution design choice and the underlying usage of plain Python as the notebook storage format. They’re integrating agentic AI features, of course, but there’s interesting possibilities for agentic co-development of an interactive computational artifact along with the user. Probably worthy of doing some digging into the CS research literature for comps.
Here’s to marimo finding traction and enduring.