True its is like a junior dev with brain damage and insane ego (especially Fable who thinks its a rockstar programmer for some reason), but you have to use plans, anything else turns into a turd way to quickly. Write a summary of what you want to happen, have it plan it (this is where the architecture decision gets made) then iterate on that plan until its detailed enough and the design is sound. Let it work in auto mode on the plan then review the changes. This only really works if you have a good mental model of the codebase.

Its still a lot of work, just less than having to code literally everything yourself. I agree though it feels to me 99% of discussions on LLMs is either marketing bullshit or junior devs talking out of their asses about LLM coding, who anyway feel that

> "different implementation of security for every endpoint in your app"

is perfectly reasonable. It also seems as if people haven't sandboxed claude already (WTF), that says a lot too about the competence of people speaking here.