Claude Code adds cloud ultraplan planning

This is less about raw agent power than about workflow design: Anthropic is turning Claude Code into a two-stage system where planning and execution can happen in different places. That makes the product feel more like a coordinated coding workspace than a terminal bot.

• –The browser review surface is the real upgrade here; targeted comments and outline navigation are much better than iterating in a CLI prompt.
• –Keeping the terminal free while the cloud session drafts the plan is useful for parallel work, especially on larger refactors or migrations.
• –The handoff choice is smart: teams can approve and run in the browser or bounce the plan back locally when they need full environment access.
• –The feature depends on Claude Code on the web plus GitHub access, so it adds capability, but also more moving parts and more surface area for operational friction.
• –Research preview status matters; this looks like a strong direction for agentic coding, but the boundaries and reliability are still likely to move.