Merry Shell(e)y [weekly-2025-w50]
- December 15, 2025
- Patrick Ferris
Aside from a serious amount of admin, last week I decided to spend a majority of my time thinking about shells.
What is a shell?
- December 15, 2025
- Patrick Ferris
There is no single definition of what a shell is or isn't.
AI Blues [ai-blues]
- December 15, 2025
- Patrick Ferris
Undoubtedly last week, I suffered a severe case of "AI blues" (for lack of a better description).
I have been doing some thinking, trying to distil my own thoughts on what aspects of vibecoding seem to irk me so much. Ultimately, it seems it is more about attitudes than anything else.
Exploitative
- December 15, 2025
- Patrick Ferris
Increasingly, I am finding that other people's vibecoded work is my problem. I have been an active participant in OCaml's outreachy work for going on five years now, and a problem mentors have to face is deciding the appropriate amount of help to give to your intern when they come to you with questions. More than likely, you could answer with specifics and solve the problem at hand. However, this doesn't scale very well and doesn't help your intern learn where else they may look for answers (e.g. Dune's documentation, a project's README page, OCaml's package explorer, heck even just using some simple CLI tools!) or how they might go about debugging a certain problem.
LLMs, in my experience, make this problem worse. When they come up short, they can leave their programmer higher and drier than they would be had they hit a problem on their own. There is an inertia to understanding problems now that there's lots of code lying around. Turning to your colleagues has a few problems that it didn't before:
Your colleagues don't know whose understanding they are correcting: yours or the LLMs?
If it is the latter, there seems to me to be an exploitative quality to even asking for help until you yourself know what's going wrong.
Technological Accelerationism
- December 15, 2025
- Patrick Ferris
There is no doubt that LLMs are pretty good at writing large amounts of okay code. The generation of this code, and often subsequent publishing of it, completely outpaces the ability to maintain it.
Programmers may be able to release ten times more libraries than they could before, but at the moment, they certainly cannot maintain ten times more libraries than they could before.
Scale --