It's part of why they love agents and tools like cursor -> turns a problem that could've been one prompt and a few hundred tokens into dozens of prompts and thousands of tokens ;)
It's be nice if I could solve any problem by speccing it out in its entirety and then just implement. In reality, I have to iterate and course correct, as do agentic flows. You're right that the AI labs love it though, iterating like that is expensive.
Comments
"Foundational AI companies love this one trick"
It's part of why they love agents and tools like cursor -> turns a problem that could've been one prompt and a few hundred tokens into dozens of prompts and thousands of tokens ;)
It's be nice if I could solve any problem by speccing it out in its entirety and then just implement. In reality, I have to iterate and course correct, as do agentic flows. You're right that the AI labs love it though, iterating like that is expensive.