Pointing twenty coding agents at one goal is a management problem before it is a compute problem, and Anthropic is now selling the manager.
Claude Code projects are being rebuilt around a coordinator and a set of addressable threads. Claude takes a request, breaks it up, runs parallel threads, checks the results and assembles them, and Anthropic says the work continues after a laptop closes. Threads carry a shared memory that accumulates project knowledge, which the company says cuts down on the prompt engineering needed to keep related work consistent.
Each thread is a full cloud session with its own branch and its own copy of the repository it edits. Naming them matters: a user can drop into a running thread, see where it is heading and redirect it, instead of burning tokens on a result that is visibly wrong.
Anthropic advises briefing Claude the way you would brief a chief of staff and letting it route work outward. The same architecture, it says, is what the feature uses under the hood, so persistent threads are less a new parallelism than a way to reopen and discuss the subagent work developers already fan out.
Availability is limited for now, going to select Pro and Max subscribers on cloud sessions who have no existing web or desktop projects. Anyone already using projects keeps the current behaviour until a later upgrade.
ZDNET’s David Gewirtz writes that the feature does not appear fully autonomous, since threads seem to begin when a user presses Start on each suggested one, and he flags unresolved questions about migrating existing setups. Local developers are excluded, though Anthropic says local execution arrives very soon and the terminal stays.
