tech
April 8, 2026
512,000 Lines of Leaked Code Reveal the Lock-In Strategy Coming for Your AI Stack
Anthropic is building an always-on agent called Conway. It wasn’t announced. It was found, buried in the Claude Code source that Anthropic accidentally published last week when a packaging error pushed half a million lines of code to a public registry. While everyone focused on the source code itself (the takedown notices, the security flaws, the modified forks spreading across the internet), the more consequential discovery went largely unnoticed.
TL;DR
- Anthropic's internal project, Conway, an always-on agent, was discovered in accidentally leaked Claude Code.
- Conway is a standalone agent environment with its own extension format, capable of reacting to external events and connecting to other tools.
- The leak suggests Anthropic is developing a platform strategy with Conway, potentially positioning itself as an operating system.
- Conway's proprietary extension format and the concept of behavioral context are highlighted as mechanisms for significant vendor lock-in.
- The article aims to explore the implications of an always-on agent that learns user behavior for vendor lock-in.
Continue reading the original article