Jupyter is the poor man's debugger

Provocative? Let me explain.

Jupyter's weaknesses

• Inconsistent state with the code (unless you restart the kernel every time)
• Limited view of the application state — you need print statements everywhere
• Production integration? A nightmare
• Advanced debugging? Forget it, everything is done block by block manually
• Overloaded memory, block-by-block execution = slowness
• Labyrinthine navigation in code that's 400 pages tall
• Tooling? Zilch, figure it out yourself
• Poor syntax highlighting to top it all off

The Python debugger: the professional solution

• Guaranteed repeatability at every execution
• Full visibility of the app state (even plugins to visualize image variables and tensors)
• Smooth integration into production with the right entrypoint
• Very advanced debugging: step-by-step, conditional breakpoints, remote debugging, stepping into sub-functions...
• Optimized performance
• Intuitive navigation through the code architecture
• Rich ecosystem: auto-completion, linters...
• Git friendly

So why do so many people still use Jupyter?

A. For one-shot data analysis focused on visualization

B. They don't know how to use the Python debugger

Those who answered B, don't panic! Learning to use a debugger is doable. And it's a game-changer.

Hot take

Jupyter is a one-shot tool that will never compete with a good debugger in the long run.

For quick exploration and occasional visualization, it's perfect. For everything else — development, debugging, production — take 30 minutes to learn how to use the VSCode or Cursor debugger. You'll never go back.

---

Originally published on LinkedIn.