In since post-Xcode Agentic coding we’re all having the “are LLMs compilers” conversation again…

It’s really helpful to think of LLM’s as databases of _someone else’s code._ That pretty quickly establishes good bounds of what they are and what they are not. They’re a bit more than that but still fundamentally constrained by the data.

That also gives you a good sense of the ethics, and the progress curve. They can really progress only as long as they have data to gobble up.