I’ve continued using Claude Code after its recent addition to the Pro Plan. One thing that’s jumped out to me is the difference in experience between the terminal-only UI vs. the side-by-side in Cursor. With Claude Code, you can’t see all of the code, only snippets. And those views are all mediated by the AI. Setting its effectiveness as a coding agent aside, the fact that you can’t see the code forces you to trust and accept what it’s giving you.
A lot has been said about how good or bad these agents are, but I think the mental shift here is the interface. You’re no longer working side-by-side with the AI, you’re fully depending on it.
“AI as middleman” has many implications I don’t think we fully understand and appreciate yet.