An Open Letter to Cory Doctorow: Ollama is part of the enshittification!

To start, like a lot of people in this space, I'm a big fan of Cory Doctorow's work. He goes up to bat for initiatives I firmly believe in, such as Creative Commons. He coined the term enshittification, and I've used it often in my own writings.

Which is why I, like many others, was surprised to see him dedicate a rather large portion of his [6th year celebratory post](https://pluralistic.net/2026/02/19/now-we-are-six/) to performing apologetics for generative AI. He laments that anybody who is wholly against genAI is participating in "purity culture" "of the neoliberal ideology".

Is this not the same neoliberalism that, you know, gave way to OpenAI et al. receiving billions of dollars in funding despite being wholly unprofitable? [This political ideology which is defined](https://plato.stanford.edu/entries/neoliberalism/) as prioritizing rapid expansion, deregulation, and free-market capitalism. Where the fuck is puritanism anywhere in there?

## ...a spell checker?

But I don't want to get into the weeds of politics or ethics here, there are far more intelligent people to do that. What I want to do is get into the weeds of his bizarre use case that he's defending in the first place.

Cory declares his use-case for LLMs saying he uses an offline model to "run the text [of his writing] through Ollama as a typo-catcher."

...why? I don't see how an offline LLM would be better at checking for spelling errors than, you know, spell check? He says that "catching these typos at the start of the process is a huge time-saver" which is why this is his preference, but this is a solved problem. Every text editor, ranging from OpenOffice to Sublime Text to NeoVim has that built-in or available as a plug-in.

There's something much more important about his choice here, though. He says technology should be liberated through "free and open versions". But Cory, Ollama is _part_ of enshittification. It's built on [llama.cpp](https://github.com/ggml-org/llama.cpp). There are [license violations where Ollama's releases didn't include required MIT license copyright notices](https://github.com/ollama/ollama/issues/3185) for over a year, and [the absence of any mention of llama.cpp in Ollama's README despite heavy dependence on it](https://www.arsturn.com/blog/is-ollama-stealing-from-llamacpp-the-open-source-ai-controversy-explained).

Beyond attribution issues, [Ollama forked the ggml inference engine without upstream coordination, resulting in implementations incompatible with standard GGUF files and 20-30% slower than running llama.cpp directly](https://www.nijho.lt/post/llama-nixos/), while also introducing proprietary model packaging that fragmented the ecosystem.

Ollama built a venture-backed commercial product on llama.cpp's foundation while not actively