Anthropic Tests 'Conway,' an Always-On Autonomous AI Agent Platform

Anthropic appears to be quietly developing its most ambitious product to date — an always-on autonomous AI agent platform codenamed "Conway" that would mark a fundamental departure from the company's existing chatbot and coding assistant offerings. Details of the project, which first surfaced in early April 2026 through leaked testing catalogs and were subsequently confirmed by multiple industry outlets, describe a standalone agent environment designed to operate continuously, respond to external triggers, and execute complex workflows without requiring constant human prompting.

Unlike Claude's current conversational interface or even the more hands-on Claude Code tool, Conway is architected around persistent, event-driven operation. The platform reportedly features a sidebar-based UI that runs independently from traditional chat windows, giving the agent its own dedicated workspace. Central to its design is a webhook activation system that allows external services to wake the agent through public URLs — effectively enabling Conway to listen for and respond to real-world events around the clock, from incoming emails to database changes to deployment alerts.

Perhaps the most striking aspect of the leaked details is Conway's extensibility framework. The platform introduces a proprietary .cnw.zip extension format that enables third-party developers to build and distribute add-ons with custom UI tabs, context handlers, and specialized tools. Users can reportedly install extensions simply by dropping the compressed files into the system, pointing to Anthropic's ambition to cultivate an ecosystem of plug-ins around its agent — a strategy reminiscent of early browser extension marketplaces or smartphone app stores, but applied to autonomous AI.

The platform also integrates directly with Chrome, granting the agent controlled internet access with configurable permissions. This browser integration, combined with a built-in notification system that alerts users when tasks are completed, paints a picture of an AI that operates more like a digital colleague than a tool. Conway can reportedly observe its environment, browse the web for information, execute multi-step workflows, and then report back — all without requiring a user to sit in front of a chat window issuing commands.

Sources familiar with the project indicate that Conway is being developed in conjunction with Anthropic's Cowork initiative, with a target audience that extends well beyond software engineers to encompass all non-technical professionals in the workplace. The vision, according to those briefed on the effort, is nothing less than a new "operating system for AI" capable of functioning autonomously across both local and cloud environments. If accurate, this positioning would place Anthropic in direct competition not only with OpenAI's own agentic product ambitions but also with enterprise workflow automation platforms like Microsoft's Copilot ecosystem.

Industry analysts have been quick to frame Conway as evidence that the era of passive AI chatbots is drawing to a close. The combination of always-on availability, native extensibility, browser control, and external event integration represents a capability set that no major AI company has publicly shipped in a unified product. While OpenAI and Google have each demonstrated elements of agentic AI behavior in research previews and limited product features, Conway — if it reaches general availability — would represent one of the first integrated platforms purpose-built for continuous autonomous operation.

Anthropic has not officially commented on Conway or confirmed any timeline for a public release. The company, which has raised billions in funding and positions itself as a safety-focused AI lab, would face significant scrutiny around the autonomy and security implications of a persistently running agent with web access and extensible capabilities. Questions around guardrails, data privacy, and the boundaries of autonomous action are likely to intensify as the product moves closer to launch. For now, Conway remains in testing — but the details that have emerged suggest Anthropic is betting that the future of AI is not a smarter chatbot, but an always-on digital worker that never clocks out. The Conway revelations came to light following an accidental leak of Claude Code's source code on March 31, which exposed hidden feature flags and internal project codenames.