Anthropic has redesigned Claude Code Projects into a beta system that coordinates parallel AI worker threads for long-running, multi-step engineering tasks, available initially to select Claude Pro and Max subscribers using cloud sessions without existing web or desktop projects. Instead of manually dividing work across sessions, users define a goal — such as reducing checkout p75 latency or retiring a deprecated v1 endpoint across API, web, and mobile repositories — and a coordinator agent scopes the request, delegates tasks to independent threads, monitors progress, resolves merge conflicts when threads touch overlapping code, and assembles results. Each thread operates as a full Claude Code cloud session on its own repository branch, capable of opening pull requests, running tests, reading documents, and drafting changes; threads can further subdivide work using subagents, loops, and workflows. Users steer progress via a main project chat or by diving into individual threads, and can configure the cloud environment, connectors, plugins, instructions, and model per project. A shared memory accumulates context over time — capturing decisions like release dates, architectural rationale, and communication preferences — reducing prompt engineering overhead, while a library collects user-provided files and Claude-generated artifacts for reuse. Usage limits may be reached faster due to concurrent threads; project-specific usage dashboards let users select models and effort levels for both coordinator and workers. Local execution alongside user tools and behind their network is planned soon.
Read original
hackernews