Claude Code Ultraplan: Planning Moves to the Cloud, Terminal Stays Free
Claude Code just got a feature that solves one of the most annoying problems in agentic coding: planning locks up your terminal. The new /ultraplan command sends the planning task to the cloud where multiple Opus 4.6 agents work in parallel, and gives you back your terminal immediately.
The numbers tell the story. Standard planning mode: 4+ minutes of your terminal frozen solid. Ultraplan: about 1 minute, running on Anthropic's cloud infrastructure while you keep coding. When the plan is ready, you open it in a browser, leave inline comments on specific sections, and choose whether to execute it in the cloud or send it back to your terminal.
This matters because planning is where agent-assisted coding actually succeeds or fails. Bad plans waste hours of agent execution time. Ultraplan lets you treat planning as a collaborative review process β like a Google Doc for your codebase architecture. You comment, Claude revises, you react with emoji, and only when you're satisfied does execution begin.
The catch: it requires Claude Code v2.1.91+, a Claude Code web account, and a GitHub repo. And it's not available through Bedrock, Vertex, or Foundry β Anthropic-direct only. That last part is telling. Anthropic wants the best Claude Code experience to happen on their own infrastructure.
Docs: https://code.claude.com/docs/en/ultraplan
← Back to all articles
The numbers tell the story. Standard planning mode: 4+ minutes of your terminal frozen solid. Ultraplan: about 1 minute, running on Anthropic's cloud infrastructure while you keep coding. When the plan is ready, you open it in a browser, leave inline comments on specific sections, and choose whether to execute it in the cloud or send it back to your terminal.
This matters because planning is where agent-assisted coding actually succeeds or fails. Bad plans waste hours of agent execution time. Ultraplan lets you treat planning as a collaborative review process β like a Google Doc for your codebase architecture. You comment, Claude revises, you react with emoji, and only when you're satisfied does execution begin.
The catch: it requires Claude Code v2.1.91+, a Claude Code web account, and a GitHub repo. And it's not available through Bedrock, Vertex, or Foundry β Anthropic-direct only. That last part is telling. Anthropic wants the best Claude Code experience to happen on their own infrastructure.
Docs: https://code.claude.com/docs/en/ultraplan
Comments