Anthropic upgrades Claude Code Projects to run parallel AI workers

Anthropic launched an overhaul of Claude Code Projects, turning the developer tool into an autonomous coordinator directing parallel cloud sessions across complex workflows. According to claude.com's report, users specify goals while Claude scopes requirements, delegates tasks across background threads, reviews pull requests, and merges changes. The system operates across connected codebases, running persistent cloud routines even after engineers step away from their workstations.

This report details the coordinator architecture, branch isolation, shared memory persistence, beta tier access, and quota limits for engineering teams deploying autonomous agents. Technical leads evaluating multi-repository workflows will see how the redesigned platform handles concurrent endpoint profiling, cross-repo deprecations, subagent loops, and planned on-premises deployments.

01Coordinator architecture and parallel cloud threads

Starting a project begins by choosing an engineering goal and linking relevant code repositories or technical documentation. Claude immediately analyzes the workspace and suggests starter tasks without requiring manual task division. Developers can tailor the cloud runtime by configuring target models, reasoning effort levels, system instructions, environment connectors, and installed tools.

The central coordinator acts like an engineering chief of staff, translating instructions into tasks for concurrent execution threads. Developers can guide progress from the main chat or inspect individual threads to review command logs, file modifications, and test execution details. Autonomous worker threads follow through on assignments by opening pull requests against codebases and drafting documentation.

02Branch isolation and performance metrics

Parallel thread coordination replaces the friction of juggling terminal tabs, managing context handoffs, and manually stitching branches. For performance optimization, teams can instruct Claude to profile endpoints and reduce checkout p75 latency across distributed services. The agent profiles routes concurrently, evaluates optimizations on isolated branches, runs test suites, and opens pull requests.

Cross-repository maintenance benefits from identical parallelism during major architectural changes, such as retiring a deprecated v1 endpoint across services. The coordinator provisions distinct threads across API, web, and mobile repositories simultaneously to update callers and validate test suites. After opening the required pull requests, Claude specifies the exact merge order to prevent breaking changes.

03Shared project memory and artifact library

Projects are engineered for long-running agentic workflows that span multiple days, complex steps, and distributed team handoffs. Instead of forcing developers to repeat instructional prompts across sessions, Claude automatically persists project knowledge into a centralized workspace memory. Every worker thread contributes to and queries this shared context, ensuring architectural decisions remain consistent throughout extended delivery cycles.

This context layer remembers technical decisions such as delayed release targets, dropped features, and designated service owners for critical modules like billing. Claude also adapts to individual communication preferences, allowing users to tune check-in intervals, thread creation rates, and update verbosity. Additionally, an integrated library organizes uploaded files and agent-generated artifacts so future assignments build directly on past deliverables.

04Beta tier access and setup commands

The upgraded project environment is available in beta to select Claude Pro and Max subscribers using cloud sessions who do not have legacy projects. Anthropic plans to broaden access to additional Pro and Max developers over the coming week before expanding to Team and Enterprise tiers. Existing projects remain active on current plans and will migrate as rollout reaches chat and Cowork.

Developers update the client, switch models, and establish execution defaults using standard terminal commands and in-app directives: ``bash npm install -g @anthropic-ai/claude-code claude update ` `text /model claude-fable-5-1 /model fable ` `bash claude --model fable claude config set model fable `` Project dashboards enable teams to monitor token consumption, calibrate coordinator reasoning effort, and assign distinct models across worker threads.

05Cloud token limits and self-hosted agents

Because each background thread runs a complete Claude Code cloud session, dispatching multiple concurrent workers accelerates token consumption and reaches plan limits faster. Engineering teams must monitor project-specific usage meters and adjust model selections between the coordinator and individual threads. Configuring lightweight models for routine test passes preserves subscription quotas while reserving advanced models for complex architectural coordination.

While worker threads currently run in the cloud, local execution alongside private tools and internal networks is scheduled for an upcoming release. Anthropic also announced self-hosted Claude Managed Agents on November 12, 2026, allowing enterprises to run autonomous agents within private infrastructure. This deployment architecture lets security-sensitive organizations host agent credentials, session histories, and data routing internally.