Transforming the programming environment from a static folder of files into a continuous, dynamic dialogue with a coordinating agent is akin to hiring a technical operations manager who runs a full team of programmers.
Article Index:
- Reinventing the workspace from folder to chat
- Parallel thread engineering and code conflict resolution
- Shared cumulative memory and output repository
- Daily consumption limits and cloud cost controls
- Team availability plan and the future of local execution
- The fierce battle over code agent orchestration
- Frequently asked questions
Reinventing the workspace from folder to chat
Anthropic announced a major upgrade on Thursday to the project management feature in its programming platform, Claude Code, introducing a revolutionary concept based on a persistent smart agent that takes on the role of coordinator and distributor of engineering tasks. This upgrade—titled "Project Redesign: From Folder to Chat"—replaces the old model based on static files and a single chat window with a continuous interactive dialogue that analyzes major programming goals and smartly distributes them to sub-workspaces.
Anthropic compared this experience to appointing an "executive operations manager" for the developer, where the engineer drops multiple varied functional tasks at once and in any order they prefer, allowing Claude to sort them, determine execution priority, open parallel work sessions, and periodically monitor progress.
Parallel thread engineering and code conflict resolution
Each session or sub-work thread operates as an independent cloud programming environment with its own version of the code repository and an independent branch in the Git version control system. The independent agent can write code, run automated tests, call verification tools, open official pull requests to review modifications, and immediately push fixes if continuous integration tests fail.
When two different programming threads coincidentally modify the same part of the code, the system handles the result as a standard merge conflict resolved and managed using the same criteria and tools familiar to human programmers, enabling weeks’ worth of tasks to be completed in a few hours.
Shared cumulative memory and output repository
Anthropic equipped the system with a central, shared memory layer above all individual programming threads. This layer allows Claude to permanently retain key engineering decisions agreed upon, such as postponing a specific launch date, canceling a particular feature, or committing to consult a specific technical expert before modifying a complex service architecture.
The interface also includes a dedicated library tab that gathers all uploaded files, documents, and digital outputs produced by the threads during their work, forming a cumulative knowledge base that subsequent sessions rely on without needing to re-explain the project context or work requirements from scratch.
Daily consumption limits and cloud cost controls
Because each programming thread represents a full inference session consuming computational resources within Claude Code, the consumption rate of available quotas in subscription plans rises faster than traditional single sessions. To prevent resource exhaustion, Anthropic set a maximum limit allowing the creation of up to 200 new programming threads daily across a user’s various projects.
Idle threads can wake up autonomously and consume additional credit upon receiving a notification of a failed test or a new code review registration from the team, with user control to choose model types and define computational effort levels consumed by the coordinating agent and working threads separately.
Team availability plan and the future of local execution
The beta rollout of the feature began on September 17 for Claude Pro and Max plan subscribers using cloud sessions in Claude Code, and will expand gradually to all subscribers over the coming days in preparation for arriving at team and enterprise plans at a later stage.
Threads currently operate exclusively via cloud servers, but Anthropic confirmed that local operation and execution support on personal devices is coming "very soon," a crucial step needed by companies and institutions implementing strict security policies that prevent sharing their software packages or closed networks with the public cloud.
The fierce battle over code agent orchestration
This launch places Anthropic at the heart of a heated race alongside major companies like OpenAI, GitHub, and Google, where competition is no longer confined to mere code completion quality, but has shifted to the cleverness of orchestration and coordination of autonomous agents.
Although tools like OpenAI’s Codex and GitHub agents allow concurrent sessions, Anthropic bets that having a smart, persistent conversational coordinator represents the most appropriate software layer to manage long-term software projects, giving developers unprecedented capability to build integrated systems with minimal manual effort.
Frequently asked questions
Q: What is the core feature in the new Claude Code Projects system?
A: The presence of a persistent smart agent that coordinates tasks and divides them into independent programming threads operating in the cloud even after closing the computer.
Q: How does the system handle code conflicts resulting from parallel threads?
A: It treats them as regular merge conflicts in Git and opens official pull requests to review and resolve them automatically or programmatically.
Q: Does the new projects tool support working on local servers and networks currently?
A: It currently works in the cloud only, but Anthropic confirmed that local execution support is coming soon to support businesses and restricted environments.
