Leaked code of Anthropic (March 2026) reveals something remarkable. Conway, a previously unannounced, always-on AI agent. This leak reveals a comprehensive, multi-surface platform strategy that signals the start of intense AI platform wars.
Key Takeaways:
- The Conway Agent: Operating as a persistent, standalone environment within Claude, Conway can monitor emails, track Slack channels, and perform autonomous tasks overnight based on learned user patterns.
- The Platform Play: Anthropic is executing a five-surface strategy including Claude Code (developers), Claude Co-Work (enterprise), Conway (the always-on agent), a marketplace for procurement, and new enforcement mechanisms.
- The "Active Directory" Parallel: Much like Microsoft's strategy in the 90s, Anthropic is attempting to lock in enterprise users by controlling how businesses compute and interact with AI, with Conway serving as the central, sticky "Active Directory" layer.
- Proprietary Lock-In: While Anthropic champions the open Model Context Protocol (MCP), Conway uses a proprietary extension format (CNW.zip) that traps developers into the Anthropic ecosystem.
- Behavioral Lock-In: Conway learns your specific work habits and behavioral patterns over months, switching to a different provider would mean losing that accumulated context - effectively starting from scratch with a "brilliant stranger”. I am naming it as “intelligence portability issue”.
The competitive landscape is shifting from foundation models to the "persistence layer." Companies and individuals are now facing a crucial choice: stick with a convenient, proprietary platform like Conway (leading to extreme vendor lock-in) or prioritize open, portable memory layers.
Posts
No comments:
Post a Comment