Anthropic released partner On Monday, a new AI agent ability that boosts its wild success power Claude Code tool for non-technical users—and according to company insiders, the team built the entire feature in about a week and a half, largely using cloud code itself.
The launch marks a major inflection point in the race to deliver practical AI agents to mainstream users, and not just Entropic to compete with. Open Eye And Google In conversational AI, but with Microsoft’s Copilot In the growing market for AI-powered productivity tools.
"Companion lets you complete non-technical tasks like how developers use cloud code," The company announced Through your official Cloud account on X. The feature arrives exclusively as a preview of available research Claude Max Subscribers – via macOS desktop application – Anthropic’s Power User tier costs between $100 and $200 per month.
For the past year, the industry narrative has focused on large language models that can write poetry or debug code. with the partnerAnthropic is betting that the real enterprise value is in an AI that can open a folder, read a messy stack of receipts, and generate structured expense reports without the need for a human hand.
How Developers Using Coding Tool to Research Vacations Influenced Anthropic’s Latest Product
The birth of partner Anthropic has had recent success with the developer community. In late 2024, the company released Claude Codea terminal-based tool that allowed software engineers to automate root programming tasks. The tool was a hit, but Anthropic noticed a strange trend: users were forcing the coding tool to perform non-coding labor.
According to Boris Cherryan engineer at Anthropic, the company observed that users deploy developer tools for unexpectedly diverse tasks.
"Since we launched CloudCode, we’ve seen people use it for all sorts of non-coding tasks: researching vacations, creating slide decks, cleaning up your email, canceling subscriptions, retrieving wedding photos from a hard drive, monitoring plant growth, controlling your oven," Cherry wrote on X. "These use cases are varied and amazing – that’s because the basic cloud agent is the best agent, and Ops 4.5 is the best model."
Recognizing the use of this shadow, Anthropic effectively stripped the complexity of the command line from its developer tool to create a user-friendly interface. In his blog post announcing the feature, Anthropic explained Those developers "Quickly started using it for almost everything," which one "Prompted us to create a partner: an easy way for anyone – not just developers – to work with the same cloud."
Within a folder-based architecture that allows Cloud to read, edit, and create files on your computer
Unlike a standard chat interface where the user pastes text for analysis, partner A different level of trust and access is required. Users designate a specific folder on their local machine that the cloud can access. Within this sandbox, the AI ​​agent can read existing files, modify them, or create entirely new ones.
Anthropic offers several clear examples: sorting through each file and intelligently reorganizing a cluttered downloads folder, generating an expense spreadsheet from a collection of receipt screenshots, or generating a report from notes scattered across multiple documents.
"In Companion, you give Cloud access to folders on your computer. Claude can then read, modify, or create files in that folder," The company explained On X "Try creating a spreadsheet from a pile of screenshots, or create a first draft from scattered notes."
Architecture relies on what is known as "Agent Loop" When a user assigns a task, AI doesn’t simply generate a text response. Instead, it creates a plan, takes steps in parallel, checks its work, and asks for clarification if it hits a roadblock. Users can queue up multiple tasks and let the cloud process them simultaneously – a workflow Anthrop describes as a realization "Less like back-and-forth and more like leaving messages for a partner."
This system is built on Anthropic Cloud Agent SDKmeaning it shares the same basic architecture as cloud code. Entropic notices that fellows "Cloud can take on many tasks that handle code, but in a more accessible form for non-coding tasks."
The iterative loop where AI creates AI: Claude Code has reportedly written a lot of Claude colleagues
Perhaps the most notable detail surrounding the launch of the companion is the speed at which the tool was reportedly built – highlighting a recurring feedback loop where AI tools are being used to build better AI tools.
During a live stream hosted by Dan Schipper, an Anthropologist employee, Felix Reisberg, confirmed this tThat team Made a mate in about a week and a half.
Alex Volkoff, who covers AI developments, marveled at the timeline: "Holy shit anthropic built ‘mate’ in the last … week and a half?!"
This prompted immediate speculation as to how much Claude Codd himself had been complicit. Simon SmithEVP of Generative AI at ClickHealth, put it bluntly on X: "Claude Code wrote all Claude Companions. Can we all agree that we are at least somewhat in an iterative improvement loop here?"
The implication is profound: Anthropic’s AI coding agent has contributed substantially to the building of its non-technical sibling products. If true, this is one of the most visible examples yet of AI systems being used to accelerate their own development and expansion — a strategy that could widen the gap between AI labs that successfully deploy their agents internally and those that don’t.
Connectors, browser automation, and capabilities extend the companion’s reach beyond the local file system
A partner does not work in isolation. The feature integrates with Entropic’s existing ecosystem of connectors – tools that link Claude External information sources and services such as Asanafor , for , for , . ideafor , for , for , . PayPaland other supporting partners. Users who have created these connections in the standard cloud interface can leverage them in peer sessions.
Additionally, can pair with a partner Claude in ChromeAnthropic’s browser extension, to implement tasks that require web access. This combination allows an agent to navigate a website, click buttons, fill out forms and retrieve information from the Internet – all from a desktop application.
"Kokurt includes several novel UX and security features that we think make the product truly special," Cherry explainedto highlight "A built-in VM (virtual machine) for isolation, out-of-the-box support for browser automation, support for all your cloud.com data connectors, asking you for clarification when you’re not sure."
Anthropic An initial set of "skill" Designed specifically for colleagues, it extends the cloud’s ability to create documents, presentations and other files. They build on it Skills for Claude The framework, which the company announced in October, provides specialized instructions that the cloud can load for specific types of tasks.
Why is Anthropic warning users that its own AI agent can delete their files?
A transition from a chatbot that suggests editing to an agent that poses a significant risk to editing. An AI that can manage files can, theoretically, delete them.
In a remarkable display of transparency, Anthopak devoted considerable space to his announcement Alerting users to potential peer threats – An unusual approach to a product launch.
The company clearly recognizes this cloud "Can take potentially destructive actions (such as deleting local files) if directed." Because Claude may occasionally misinterpret instructions, Anthropic urges users to provide "Very clear guidance" About sensitive operations.
More immediate about this is the risk of injection attacks. A technique where malicious actors can embed hidden instructions in the cloud, potentially causing the agent to bypass security measures or take malicious actions.
"We have built sophisticated defenses against instant injection," Anthropic wrote, "But agent security—that is, the task of securing the real-world actions of the cloud—is still an active area of ​​development in the industry."
The company attributes these threats to the current state of AI agent technology rather than being unique from peers. "These risks aren’t new to colleagues, but it may be the first time you’re using a sophisticated tool that goes beyond a simple conversation," Announcement Note.
Anthropic’s desktop agent strategy sets a direct challenge to Microsoft Copilot
The beginning of partner Anthropic spaces in direct competition Microsoftwho has spent years trying to integrate it Copilot ai into the fabric of the Windows operating system with mixed adoption results.
However, Anthropic’s approach differs in its isolation. By limiting the agent to specific folders and requiring explicit connections, they are trying to strike a balance between the utility of an OS-level agent and the security of a sandboxed application.
What distinguishes the anthropic approach is its bottom-up evolution. Instead of designing AI assistant and retrofitting agent capabilities, Anthropic first built a powerful coding agent – Claude Code – and is now abstracting his talents to a wider audience. This technological lineage may give peers stronger agentic behavior from the outset.
CloudCode has generated significant excitement among developers since its initial launch A command line tool in late 2024. The company expanded its reach with a Web interface in October 2025, followed by the Slack integration in December. Cocork is the next logical step: bringing the same agentic architecture to users who may never touch a terminal.
Who can access Peers now, and what’s coming next for Windows and other platforms
For now, it remains exclusive to colleagues Claude Max Subscribers Using the Mac OS desktop application. Users of other subscription tiers – Free, Pro, Team, or Enterprise – may be waitlisted for future access.
Anthropic has indicated clear intentions to expand the feature’s reach. The blog post clearly mentions plans to add cross-device sync and bring colleagues to Windows as the company learns from the research preview.
Cherry set expectations appropriately, describing the product "Early and raw, Cloudcode feels like it did when it first launched."
For access partnermore and more users can download or update cloud macOS app and click "partner" In the sidebar
The real question facing enterprise AI adoption
For technology decision makers, the implications of peers go beyond any single product launch. The barrier to AI adoption is changing – no longer is model intelligence the limiting factor, but workflow integration and user trust.
As the company says, Anthropic’s goal is to make working with Cloud feel less like operating a tool and delegating to a colleague. Whether mainstream users are willing to hand over access to folders to an AI that can misinterpret their instructions is an open question.
But the pace of development of Colleagues — a key feature built in ten days, possibly powered by the company’s own AI — previews a future where these systems’ capabilities grow exponentially faster than organizations can.
The chatbot has learned to use the file manager. Anyone’s guess is anyone’s guess.