What might that next level be? 
dimitarvp
linusdm
I enjoyed the conversation too. Some of the tips I got out of this discussion are:
- the code in each cell should be very small (think in terms of what you’d execute as one expression in IEx)
- rethink how you use modules in Livebook: anonymous functions assigned to variables work just fine (this basically another take on keeping your codecells small)
- mix prose and code liberally
- Livebook allows “dual learning” where you get explained a concept in one way (prose) and then yet another way (code), which facilitates learning a lot
- the process dictionary is not(?) forwarded from cell to cell, which might impact random number generation (in favor of reproducibility)
- Livebook takes away a lot of the expert knowledge needed to get started with a live environment (in contrast with: step 1, install a compiler, etc.)
I especially liked the first part where James explains a model of problem solving that explicitly calls out three stages: (1) exploration, (2) modelling and (3) delivery. Where there sometimes is too much emphasis on the delivery part, and not enough on the first two parts that don’t involve coding. I totally share the thrill of throwing away code, or not using code after all even though you’ve just spend a day writing it 
Go listen to the episode to get more details and inspiration, it’s explained very well 
And thank you @JEG2 for the reference to the book you mentioned: Stuck? Diagrams Help about that three-step process.
2
JEG2
Author of Designing Elixir Systems with OTP
On minor correction here: It probably wasn’t very clear, but I was discussing a historical change to Livebook. Today, it does forward the process dictionary and it special cases random number seeds so everything will likely just work as you expect.
2