Cursor 3 shipped parallel agents, but is any of it new?

Cursor 3 shipped on April 2. The demos look great: eight AI agents running in parallel, each in its own Git worktree, building different parts of your project at the same time. The Hacker News thread lit up. Product Hunt gave it the #3 spot for the day. Then I read the comments. One user reported spending $2,000 in two days on cloud agents. Another switched from $1,800/month on Cursor to roughly $200/month on Claude Code and Codex. A third said they had "zero interest" in forced agent swarms and were moving to VS Code with Claude Code instead. The coverage so far has been mostly feature recaps reprinting the press release. Nobody's asking the obvious questions: is parallel agent execution actually new? What does it really cost? And what happens when your agents need to share context? Here's the Thing Cursor 2 already supported parallel execution via worktree.json configuration. What Cursor 3 actually shipped is a UI layer (Agents Window sidebar, drag-drop tabs) on top of the same Git w