@yegor256 Acually I like Claude Code in desktop app, thread management is convenient enough, and it has almost all features for common use. Not a fan of tmux mess
Spent some time on a skill prompt, then had a subagent read it cold. It flagged two rules that quietly contradicted each other.
Your prompt has these too. You just can't see them, because in your head statements cover different cases and never collide.
Subagent = QA for skills
@theo I've been migrating the other way: pulling things out of CLAUDE.md into skills. Most instructions are contextual - they shouldn't live in a global prompt that ships with every conversation.
On the engineering side of this one. The chart is the outcome; the part I care about is the system behind it β the architecture that lets growth like this happen without the whole thing groaning. Thatβs where the real work lives.
From ~500 dev responses: the real AI-adoption blocker isn't dev disinterest β it's corporate policy.
Many maintain personal Claude/Cursor subs alongside bans at work. The employee is ahead of the employer.
If your company is slow on AI, your engineers already know.