In the early days, computers were machines that could take up an entire wall, and users interacted with them via teletypes. This electronic typewriter could be in the same room as the computer, in a building next door, or in an operator’s residence where they would connect it to their home telephone and dial in. Line editors like Ed the standard Unix text editor were a good fit for the technical limitations of the day: the “command / response” model of editing works well when you can immediately see your commands printed onto the teletype paper even if the response requires a long round-trip to the nearby university.

With the advent of the personal computer, we started to see interactive text editors like vi and emacs. With a digital screen and the computer physically attached to the monitor, interactive editing was practical and useful! These editors and their descendants have dominated developer usage up until today, and their architecture assumes in many places that the editor UI is running on the same machine that is storing the files, hosting language tooling, has your built executables, etc.

But nowadays I rarely edit files on the physical machine that I’m typing into! Lots of my work is happening on machines in the cloud that’s where the beefy GPUs are . I’m running vim over ssh one of my vim plugins runs executables that need to be on the machine with the GPU , and the experience sucks; with round trip latency that can reach into the hundreds of milliseconds, every keypress takes ages to show up on my screen.

In the machine learning community where programming on powerful computers far away from your laptop is common , many practitioners use notebooks like JupyterLab. Much like Ed, with a notebook, you type out a full command locally instead of “editing” commands, these are repl-style script and then send it to the remote machine to be evaluated. Now I don’t think that notebook-programming was developed solely to save programmers from typing latency, similarly, I think that Ed’s design was mostly constrained by typewriter hardware and not round-trip-timing for keypresses but I do believe that development with a notebook is what made the practice of “writing code where it runs” ergonomic in this era of slow links to expensive hardware.