Anthropic is quietly developing Conway, an unreleased always-on AI agent designed to operate persistently in the background. Built on Claude models, this system responds to real-world triggers without requiring constant human oversight.
The project surfaced through a significant code leak earlier in 2026 involving over 500,000 lines of code. Anthropic has not publicly acknowledged the platform's existence.
Leaked documentation references a specialized interface with webhook listeners, GitHub integrations, and push notification capabilities. The system can monitor external tools and execute actions automatically based on specific events.
Core features include scheduled task execution and advanced browser automation. This allows Conway to navigate websites, complete forms, and interact with web applications similarly to a human user.
This represents a fundamental shift from Claude Code, Anthropic’s existing developer tool that runs on a one-off session basis. Conway is engineered for continuous operation rather than isolated tasks.
Anthropic introduced routines for Claude Code in April 2026, enabling event-triggered tasks on managed cloud infrastructure. This feature appears to be a lighter precursor to Conway’s full autonomous capabilities.
Research data indicates average Claude Code session durations increased from under 25 minutes to over 45 minutes within three months. This trend signals growing demand for persistent AI workflows.
Conway positions Anthropic against competitors building persistent agent platforms. While the GitHub integration suggests a developer-first strategy, browser automation opens the technology to a broader enterprise user base.
No confirmed launch timeline or public testing access has been established for the Conway platform.