Chris Lattner used to be somebody. 🫠

it is a shitty compiler that is made to pass tests instead of not being shitty, but it is still a massive accomplishment. Generating a speed or code performance improvement in an existing compiler project would be more useful and impressive. LLMs are optimized to write readable simplest code that has poor performance. The reasoning frontier is making performant working code.

Before diving in, here are my main take-aways:

AI has moved beyond writing small snippets of code and is beginning to participate in engineering large systems.
AI is crossing from local code generation into global engineering participation: CCC maintains architecture across subsystems, not just functions.
CCC has an “LLVM-like” design (as expected): training on decades of compiler engineering produces compiler architectures shaped by that history.
Our legal apparatus frequently lags behind technology progress, and AI is pushing legal boundaries. Is proprietary software cooked?
Good software depends on judgment, communication, and clear abstraction. AI has amplified this.
AI coding is automation of implementation, so design and stewardship become more important.
Manual rewrites and translation work are becoming AI-native tasks, automating a large category of engineering effort.
AI, used right, should produce better software, provided humans actually spend more energy on architecture, design, and innovation.
Architecture documentation has become infrastructure as AI systems amplify well-structured knowledge while punishing undocumented systems. 

I find it hard to square his experience with my own. Whenever I use LLMs, admittedly only the free versions because fuck paying these scrapers, they fail miserably to write code that even just runs, let alone makes use of good coding practices, when I ask for more than one specific example of code. Broaden the prompt even a little bit and the answers I'm getting are thought-starters at best and unusable garbage at worst.