Claude Code’s Secret Upgrade Is an Orchestration Beast

Claude Code Projects: Multi-Agent Orchestration in the Cloud

This video introduces Claude Code Projects, a powerful hidden feature that transforms Claude into a cloud-based project manager. Instead of manually juggling multiple terminal sessions, you interact with a single orchestrator agent that controls a fleet of full Claude sessions—each with its own context window, memory, and persistence. These sessions run 24/7 in Anthropic’s cloud, share context with each other, and draw from your Max plan usage (not API costs). The feature is ideal for complex, long-running builds that need constant attention without requiring your local machine to stay on.

Access and Setup

• Getting access: Submit an application on the waitlist page (link in description). As of now, approval typically comes within 24 hours. Once granted, a “Projects Beta” tab appears in the Claude desktop app.
• Creating a project: Name it, optionally set a goal and provide context (e.g., documentation). In Project Settings, you configure:
  • Coordinator model (default: Opus 5.5 on low effort—fine for delegation) and Thread model (workers; default Opus 5.5 on medium).
  • Auto-continue when usage resets weekly.
  • Memory: Project instructions (like claude.md) and auto-memory for persistent sticky notes.
  • Environment: Link a GitHub repo (workers send PRs there) and optionally add a cloud environment with API keys/secrets.
  • Local settings: Allow workers to access your device and designate pre-approved folders for local tasks.

When to Use (and When Not)

• Use for: Complicated builds that can be split into parallel parts, long-running streams of work you don’t want to babysit, and projects that benefit from cloud persistence (e.g., scraping, dashboards, content pipelines).
• Avoid for: Quick single‑chat tasks, work that must manipulate local files (unless connected via GitHub), or team/enterprise projects (not yet supported).

Demo Walkthrough

The creator demonstrates building an AI Creator Radar—a dashboard that scrapes X (Twitter) for trending AI news, checks YouTube coverage gaps, and suggests content formats. After giving a one‑shot prompt, the orchestrator splits the work into five parallel threads:

1. Dashboard & hosting
2. X scanner
3. YouTube coverage
4. Content packs
5. Claude Code news

Each thread operates in its own session, sharing context via the orchestrator. Blockers (e.g., missing API keys) are resolved by updating the cloud environment with secrets. One thread uses a local session (via the in‑app browser) to scan X because cookies are stored locally. The final dashboard shows trending stories, coverage gaps, content pack ideas, and channel stats—all refreshed automatically.

Key Differences from Sub‑Agents

• Persistence: Workers in Projects are full sessions that never shut off; they own their own pull requests and can be messaged individually.
• Management: You talk to one orchestrator; Claude handles delegation. In normal sessions, you manage everything.
• Context: Workers maintain their own context windows and share memory with the orchestrator, unlike sub‑agents that dump summaries back into a single session.

Key Takeaway

Claude Code Projects turns Claude into a persistent, cloud‑based project manager that runs 24/7, uses your subscription plan, and lets you control a fleet of agents from a single chat. It’s a game‑changer for anyone building complex, long‑running applications who wants to avoid babysitting multiple sessions or paying API costs. The waitlist is now fast to clear, making this feature accessible to all Max plan users.