Blog
Notes on security engineering, AI, and automation.
-
If You Didn't Read It, Why Should I?
Pasting raw LLM output into Slack or Jira doesn't save work. It moves your work onto everyone who reads it, and the research says the total cost goes up.
-
Agent Manager - The dumbest possible multi-agent coordinator
I was bored, so I built Agent Manager: a local control panel for Claude Code sessions, each one a systemd --user service, with a real browser terminal over tmux control mode and several agents working the same repo without fighting over ports. Useful to me, possibly to nobody else.
-
Claude Desktop/Cowork — MCP Gateways
Why enterprise MCP gateways don't fit Claude Desktop's admin-managed Connectors — and how per-user Desktop Extensions (Node-bundled vs Bun-compiled) change the security trade-offs of shipping them.
-
Welcome — and why I'm writing about AI security
A quick intro to this blog: notes on security engineering, automation, and securing AI workflows in the enterprise.