Claude Projects now hands one coding goal to several parallel cloud sessions

Split a migration across API, web and mobile repositories, one of the examples Anthropic gives, and each thread runs as its own cloud session with its own branch and its own copy of the repository ^[14][5]. Those sessions can change code, run tests and open pull requests ^[6]. Four threads means four branches and as many as four pull requests against a single goal ^[17].

The pitch is less manual work. devops.com writes that the redesign could mean less time dividing tasks among multiple Claude Code sessions and keeping track of the results ^[21]. Review replaces that time. Anthropic says developers still have to resolve merge conflicts when sessions make conflicting changes ^[10]. A conflict between two of your own branches is routine. A conflict between two sessions that each acted as though it owned the schema is a judgement about which one was right, and it lands on the developer ^[10].

Each thread is a full Claude Code session, and Anthropic warns that Projects can consume usage allowances more quickly for that reason ^[13]. Four threads on a goal draw roughly four sessions' worth of allowance where one chat drew one ^[16]. Individuals meet that first: the beta opens to select Pro and Max subscribers on cloud sessions, with wider Pro and Max access over the following week, and Team and Enterprise plans after that ^[3][4].

Anthropic's own reading of usage points the same way. In a June analysis of roughly 400,000 sessions, the company found users generally kept control of planning while delegating more execution to Claude, with usage moving toward more end-to-end work, including deploying and running code, across the seven months studied ^[15]. The devops.com report does not include a figure for how many threads a developer runs at once ^[20].

Shared memory keeps the parallel threads coherent, and it also spreads the record out. New threads can draw on decisions and information from earlier work ^[7], so the reasoning behind a change can sit in a different thread from the diff ^[18]. Developers can follow the work from the main project conversation, or open an individual thread to watch its progress and redirect it ^[9].

I would sort the work before splitting it by whether two threads can touch the same interface. Anything sharing a schema, an API contract or a generated client belongs in one thread, because parallelism is only cheap where the file sets do not overlap and the merge is a person's job ^[10]. Profiling several application endpoints at once, Anthropic's other example, splits cleanly ^[14].

There is a naming problem for internal documentation to handle. Anthropic describes this as a new version of the existing Projects feature and not a separate Claude Code product, and existing Projects in chat and Cowork keep working until the rollout reaches them ^[12]. For the length of the rollout, the same word means two different things depending on which surface a user is in. Threads run in the cloud for now, and Anthropic says support for running them locally alongside developers' own tools and code is coming later ^[11]. Teams that cannot have a copy of a repository running outside their own environment have nothing to pilot yet ^[19].