Maybe this is why you see the Anthropic/OAI people going like "You should be speccing research loops", because in the use case of trying lots of research directions fully autonomously, this makes a lot more sense. But not so much in normal software development. It's overkill.
Key 🗝 🦊✅
My feeling about Fable is that it's a massive leap over everything, but people who already had the dynamic of "Write spec, give spec to LLM, LLM implements spec" will not see much improvement over Opus or GPT, except on extremely difficult problems.
