Anthropic has fundamentally restructured how developers interact with Claude Code, moving away from static folder-based projects to a dynamic, conversation-driven model. The new architecture treats a single project as an ongoing dialogue where Claude acts as a coordinator. Instead of you manually managing branches or switching contexts, you describe the work, and the system automatically spawns separate threads for each task. These threads operate as full cloud sessions, running on their own branches and repository copies, allowing them to execute code, open pull requests, and even fix CI failures in parallel while you step away.
This isn’t just about multitasking; it’s about autonomous delegation. Each thread can further split its workload using subagents, loops, and workflows. For instance, if you ask Claude to reduce checkout latency by profiling endpoints, it can launch simultaneous threads to test optimizations across different services. When multiple threads touch the same codebase, conflicts are handled through standard git merge processes. Crucially, every thread inherits shared context, including up to 16,000 characters of project instructions and a MEMORY.md index that tracks vital details like release schedules or key stakeholders. This ensures consistency without requiring repetitive prompts.
However, this power comes with a steep resource cost. Because every active thread counts as a full session, usage limits burn significantly faster than in traditional single-session workflows. Anthropic caps users at 200 new threads per day, and idle threads can wake up to spend more credits if CI checks fail or review comments arrive. The default configuration runs the powerful Opus model for both coordination and high-effort execution, though users can adjust these settings. Currently, this beta feature is restricted to select Pro and Max users on web and desktop platforms, excluding CLI access.
The shift suggests a future where AI agents manage complex, multi-repository tasks with minimal human oversight. By automating the grunt work of branch management and cross-repo synchronization, Anthropic is positioning Claude not just as a coding assistant, but as a project manager. Yet, the financial implications for heavy users remain unclear. With threads consuming plan limits aggressively, teams will need to carefully monitor their usage tabs to avoid unexpected overages while leveraging this new parallel processing capability.
