For years, Jupyter Notebook has been the default tool for data analysis in Python.
But it assumes one thing:
👉 you start with code.
What if you didn’t?
I’ve been experimenting with a different workflow.
Instead of writing Python, you describe what you want in plain English — and the system generates and runs the code for you.
The flow becomes:
prompt → LLM-generated code → auto-execution → results
Code is still there.
But it’s no longer the starting point.
Why this might matter
The idea came from a simple observation.
Someone told me:
“I don’t care about the code. I care about the results.”
And honestly, that stuck with me.
As engineers, we often treat code as the main interface.
But for many people, it’s just a tool to get answers.
Old problems... less important?
There has been a lot of criticism of notebooks over the years:
- hidden state
- mixing code and outputs
- hard to review in git
These problems are real.
But I’m starting to wonder if AI changes which of them actually matter.
If:
- code is generated
- execution is automated
- results become the main interface
then some of these issues feel... less central.
New problems appear
Of course, we are not removing complexity — we are shifting it.
New challenges show up:
- trusting LLM-generated code
- debugging when something goes wrong
- less visibility into what is actually happening
In some cases, these might be even harder than the original problems.
A small but important detail
One thing I care about a lot:
👉 everything is still saved as a standard .ipynb file
So you can always:
- open it in a classic notebook
- inspect the code
- edit anything manually
Nothing is locked in.
So what’s really changing?
Maybe the biggest shift is this:
We are moving from:
code → results
to:
intent → results
Code becomes an implementation detail, not the main interface.
Open question
Are we solving notebook problems — or just hiding them behind AI?
I wrote a longer version with screenshots and implementation details here:
👉 https://mljar.com/blog/reimagine-python-notebook-in-ai-era/
I’d love to hear your perspective:
- Would you use a prompt-first notebook?
- Does this make notebooks better — or just different?
United States
NORTH AMERICA
Related News
UCP Variant Data: The #1 Reason Agent Checkouts Fail
7h ago
Amazon Employees Are 'Tokenmaxxing' Due To Pressure To Use AI Tools
21h ago
How Braze’s CTO is rethinking engineering for the agentic area
10h ago

Décryptage technique : Comment builder un téléchargeur de vidéos Reddit performant (DASH, HLS & WebAssembly)
17h ago
How AI Reduced Manual Driver Verification by 75% — Operations Case Study. Part 2
4h ago