I build systems for moving quickly with AI agents without pretending they are safe by default.

My public AI lab is Sedna Labs, where the current work focuses on Rust infrastructure for autonomous-agent operations: Model Context Protocol tooling, operator-controlled automation, local credential control, observability, and verification paths that humans can audit.

The operating assumption is simple: agents are useful because they move quickly, and risky because they make mistakes. The work is about keeping the speed while adding boundaries, provenance, review points, and fail-closed behavior when evidence is missing.

• Sedna Labs - public AI lab and home for the MCP/tooling ecosystem.
• mcp-toolkit-rs - reusable Rust foundations for MCP servers and clients.
• cloudflare-mcp, postgres-mcp, and related servers - operator-facing MCP tools for real systems.
• android-computer-use-mcp and codex - computer-use and coding-agent workflows.