Claude Code Ships Ultraplan for Cloud-Based Dev Planning

Anthropic Ships Ultraplan for Cloud-Based Development Planning

Anthropic shipped Ultraplan as a research preview feature in Claude Code, the first of several previously leaked features to move into public availability. Ultraplan enables developers to offload implementation planning to cloud-based Opus 4.6 sessions with up to 30 minutes of reserved compute time, allowing developers to work on other tasks while Claude creates a detailed architectural plan.

The feature requires Claude Code v2.1.91 or later and a Claude Code web account linked to a GitHub repository. Once invoked via /ultraplan in the terminal, the feature creates a cloud-synced snapshot of the repository and spins up a dedicated Opus 4.6 compute session to analyze the codebase and generate a comprehensive implementation plan.

How Ultraplan Works: Planning While You Work

Developers invoke /ultraplan in the Claude Code terminal and describe the feature or refactor they want to implement. Claude Opus 4.6 begins working on the plan in the cloud — reading the existing codebase, understanding the architecture, identifying dependencies, and outlining a step-by-step implementation strategy — while the terminal remains free for the developer to continue working on other tasks.

Once the plan is complete (typically 10-20 minutes for complex codebases), it appears in the Claude Code web interface where developers can review it before execution. This separation of planning from implementation is a significant architectural shift, treating planning as a first-class, asynchronous workload.

Collaborative Plan Review with Inline Comments

The finished plan appears in a structured outline view within the Claude Code web dashboard. Developers can highlight specific plan sections and leave inline comments, use emoji reactions to flag approvals or request revisions, and navigate the plan via a collapsible outline sidebar. This is a major departure from one-shot planning where Claude generates a plan and developers immediately start coding.

The collaborative interface allows team code review of the plan itself — multiple developers can review, comment on, and refine the plan before any code is written. This upfront review reduces mid-project refactors and ensures the team agrees on the approach before implementation begins.

Auto-Environment Creation and New Commands

As of April 10, Ultraplan and other remote-session features now auto-create a default cloud environment instead of requiring manual web setup, significantly reducing friction for new users. Developers no longer need to navigate Anthropic's dashboard to create a cloud environment — it happens automatically on first use.

The same update added two new commands to Claude Code: /team-onboarding, which helps teams set up Claude Code repositories with shared context and guidelines, and support for OS CA certificate trust, allowing enterprise proxy and security scanning without Claude Code breaks.

Flexible Execution: Browser or Terminal

Once the plan is reviewed and approved, developers can execute it in two ways: directly in the Claude Code web browser interface (with real-time code generation and preview), or back in the terminal where the plan was created (preserving local development workflow). This flexibility allows teams to choose whether they prefer web-based or CLI-based execution.

Web-based execution is useful for pair programming and async handoffs — a teammate can review the plan and start implementation in the browser while the original developer continues other tasks. Terminal execution suits developers who want the full plan applied to their local repository immediately.

Philosophical Shift: From Code Generation to Planning

Ultraplan represents a fundamental shift in how Anthropic positions Claude Code. Rather than treating Claude as a code generator that writes implementations on demand, Ultraplan treats Claude as a planning engine that architects solutions upfront. This mirrors how experienced engineering teams actually work: plan thoroughly, code quickly, iterate if needed.

The 30-minute compute budget is a deliberate constraint. It forces Ultraplan to produce high-quality plans within that window rather than endless refining, and it keeps costs predictable for enterprises. Users with simpler features typically see plans in 5-10 minutes; complex refactors might use the full 30.

What This Means for Developers and AI-Assisted Coding

For developers using AI-assisted coding tools, Ultraplan signals a maturation of the AI-developer partnership. Rather than asking Claude to generate code and discovering mid-way through that the approach is flawed, developers can now ask Claude to plan first, review the plan collaboratively, and then execute with confidence. This upfront planning should reduce costly mid-project refactors and improve code quality. For engineering teams, the collaborative review interface means AI-generated code no longer bypasses team consensus — the plan is reviewed and approved before implementation begins.