I’m just subtooting all of the hot takes this weekend:

Neither LLMs nor traditional OSS projects solve the problem of enabling users to control their own digital experiences. Open source is better than proprietary, but I’ve been in this field for 15 years and I am still acutely aware of my limitations when it comes to customizing the software I use.

How is the codebase organized? Do I know the language? Can I even get it to build? There’s a bunch of hurdles, and then we hit the big one: Do I want to maintain these patches for the rest of my life?

The vast majority of software is not built to be extensible by third parties at all, and the vast majority of the software that does limits itself to constrained plugin systems.

When the only option for customization is working inside a monolithic and unfamiliar codebase every time the software ever does a release, yeah, I think it’s unsurprising that people would turn to LLMs. This doesn’t mean they’re a good solution! But if the counterpoint is “just spend all your time writing and maintaining patches” then that’s just unrealistic for most professional programmers, let alone the median computer user. Seriously, most of us don’t even change our own oil.

We desperately need to find ways of structuring systems such that they can be hacked on and customized incrementally. I think this would look a lot more like building with extensible abstractions that new functionality can be glommed onto or injected into than it would look like downloading and editing the whole program’s source code.