Your comment makes it sound like what Anthropic and friends are doing is a marketing scam of little value.
I agree the long-term effects of AI are seriously worrying, but what these models can produce is truly mind-blowing. I can’t say for certain that “coding is solved”. But I sure as hell can’t say it isn’t.
It feels like we’re close to a tipping point where generating code matters less than being able to clearly articulate an idea in natural language and knowing how to validate it.
Solving coding makes no sense if you're a business that relies on that code.
Every layer of the stack has had decades of eyeballs, research, and actual validation.
Binary machine circuits, run by assembly that has been hand refined from LLVM (and GNU, etc.), generated from higher level languages that have been through so many brilliant engineers over decades that made conscious decisions how their language gets turned into assembly.
I don't need to know what Intel or Apple Silicon circuits do with the C++ or Swift code I write. As an engineer I entrust that these corporations employee other engineers who connect those dots, so that I can focus on C++ or Swift.
Natural Language (already imprecise and clunky) that non-deterministically spits out high level code works because every layer below it is doing the right thing. People love getting LLMs to generate rust because rustc is so crazy strict that LLMs can't cheat.
English on the other hand, every word can have so many meanings and developers who use LLMs to generate things they aren't familiar with, judge models based solely on whether these things can properly capture intent. It's one thing to say we can finally break free of dealing with shoddy, imprecise code, but it's another to say natural language is code.
They do not need to make any technical changes. They could make the change overnight: $1 per repo per month (or whatever).
This would stop the aggressive growth that they clearly cannot handle anyway.
The question is: are they willing to compromise growth for QoS. So far the answer is a resounding: "no". They are deliberately sacrificing the experience of existing users to gain new ones.
No way I pay $1/repo/month. No way I pay $1/repo/year. I still host and maintain my old code on GitHub, but any free tier pricing update and I jump ship immediately.
There are tons of existing free users, they'd be sacrificing those too. And yeah they don't want to sacrifice growth either, so the remaining is an engineering question.
I agree the long-term effects of AI are seriously worrying, but what these models can produce is truly mind-blowing. I can’t say for certain that “coding is solved”. But I sure as hell can’t say it isn’t.
It feels like we’re close to a tipping point where generating code matters less than being able to clearly articulate an idea in natural language and knowing how to validate it.
reply