Anthropic has been released ColleagueA new feature that runs agentive workflows on local files for non-coding tasks is currently available in Research Preview inside the Cloud macOS desktop app.
What does peer do at file system level
Cowork currently runs as a dedicated mode in the cloud desktop app. When you start a coworking session, you choose a folder on your system. The cloud can then only read, edit, or create files inside that folder.
Anthropic gives concrete examples. Cloud can reorganize a Downloads folder by sorting and renaming files. It can read a directory of screenshots, extract amounts, and create an expense spreadsheet. It can sift through the scattered notes in that folder and produce a structured report draft.
The interface maintains interactions in the standard chat surface. You describe the task in natural language. The cloud creates an internal plan, executes file operations, and streams status messages as it progresses. You can continue sending follow-up instructions while the job is running.
Connection to Cloud Code and Cloud Agent SDK
Colleagues are not a separate model. Anthropic says Cowork is built on the same foundation as Cloud Code and uses the Cloud Agent SDK as the underlying agent stack.
Cloud Code began as a command line oriented environment that allowed developers to run shell commands and mutate project files using natural language, with later web and Slack interfaces on top. Many Max users pushed cloud computing into non-coding workflows, using it as a general purpose agent that works on arbitrary directories and devices. That usage pattern directly informed the peer.
TechCrunch calls Cowork a more accessible version of cloud codeImplemented as a sandboxed instance of the same agent stack. Users still specify a specific folder, but there is no need to work in the terminal or configure a virtual environment.
Connectors, skills and browser based workflows
Peers can move from local storage. Anthropic allows Cowork to reuse existing cloud connectors, integrating external services like Asana, Notion, and PayPal. Cowork also supports an initial set of skills that are optimized for creating documents, presentations, and similar artifacts. These skills package instructions and resources for specific job tasks.
When Cowork is paired with Chrome in the Cloud, the same agent plan can include browser steps. anthropic And The Verge Both articles highlight browser-related functions where the cloud can follow links, read pages, and perform tasks inside web apps under the supervision of the user.
This combination gives Cowork a three-layer tool surface:
- Local file system access is limited to the selected folder.
- Connectors for structured external systems.
- Browser actions via the cloud in Chrome.
From an implementation perspective, this is a standard agent tool configuration on top of the Cloud Agent SDK. The difference is that the Cowork tool hides the graph and exposes only an interactive work interface.
Agent behavior, planning, and parallel tasks
Anthropic emphasizes that cowork plays out with more agency than regular cloud conversations. Once you specify a task, the cloud creates a plan, executes a series of tool calls and file operations, and keeps you updated about the intermediate steps.
You don’t need to paste the reference again and again or convert the output manually. Reuses the peer folder as a persistent reference range. It writes intermediate artifacts directly to that directory, then consumes those artifacts in later stages.
Security models, access control, and prompt injection
Because Cowork works on real files, security constraints are evident in the product design. Anthropic says users choose which folders and connectors the cloud can see. The cloud cannot read or edit content outside those structures.
The coworker always asks for confirmation before taking important action. Of course, with all the benefits, there are some caveats: It is recommended to give precise instructions and warn that misinterpretation is possible.
Anthropic also cites early injection as a primary risk. If the cloud processes untrusted content on the Internet or in local documents, it may attempt to change the content plan and drive behavior away from the user’s intent. Anthropic says it has deployed security, but it describes agent security, defined as securing real-world activities, as an ongoing research problem.
Availability
Cowork is available today as a research preview for Cloud Max customers using the macOS desktop application. The Max plan costs between $100 and $200 dollars per month, depending on usage. Users of other plans can join the waiting list. Anthropic plans to add cross device sync and Windows support in future iterations.
key takeaways
- local folder scoped agent: Cowork Cloud runs as an agent inside a macOS app that can read, edit, and create files only inside user-selected folders, giving the cloud direct access to the file system with explicit scoping.
- Same stack of cloud code, different surface: Cowork is built on the same cloud agent SDK foundation as Cloud Code, but targets non-coding workflows through a GUI rather than a terminal-based developer interface.
- Tools: Files, Connectors, Browser: Cowork combines three tool layers in one plan, local file operations, a cloud connector for services like Notion or Asana, and browser actions through the cloud in Chrome.
- Agentic, multi-step execution: Once a task is given, Cowork plans and executes multi-step workflows, streams progress updates, and can queue up multiple tasks in parallel instead of working as a single rapid response loop.
- constrained but destructively capable: Access is limited to selected folders and configured connectors, and Cowork asks before major actions, but it can still perform destructive operations like file deletion, so it should be treated like a powerful automation tool, not a harmless chat bot.
check it out Technical details here. Also, feel free to follow us Twitter And don’t forget to join us 100k+ ml subreddit and subscribe our newsletter. wait! Are you on Telegram? Now you can also connect with us on Telegram.
Check out our latest releases ai2025.devA 2025-focused analytics platform that models launches, benchmarks and transforms ecosystem activity into a structured dataset that you can filter, compare and export.
Max is an AI analyst at Silicon Valley-based MarkTechPost, actively shaping the future of technology. He teaches robotics at Brainvine, fights spam with ComplyMail, and leverages AI daily to translate complex technological advancements into clear, understandable insights.