Raw LLM Responses

It can be true (though I agree it's the most generous possible interpretation of the actual numbers - assuming this is based on real measurements). But it's hard to say what this really means and what the impact really is. Writing 75% of the code is not doing 75% of the work. There's a lot of surrounding work that's not just "writing code" - but AI tools also assist in that work, which is hard to quantify. An LLM could also write 100% of the code for a feature but not actually make you any faster due to a thousand different reasons. As always, measuring developer productivity is difficult so I hate latching on headlines like this.