New agents.
Familiar hard problems.

I spent three decades working on Internet infrastructure, from an early Texas ISP to Rackspace and Cloudflare. I now build and study multi-agent systems, with the same interest in what happens when software meets real permissions and real consequences.

What I’m testing

A model can give an answer. An agent can act on it. Give several agents tools, shared memory, and authority, and coordination becomes a security question as much as an engineering question.

I’m interested in the boundaries: how one agent’s mistake enters another’s context, how permissions spread through a workflow, and whether a human can reconstruct the decisions afterward.

Who understands the system?

I also want to understand what happens to operators as agents take on more work. An audit trail is useful. A person who can read it, question it, and intervene is essential.

A working notebook

I use this lab to run experiments and publish what I learn. My Field Notes connect the technical work to the decisions builders, security leaders, and boards have to make.

For my full background, board work, and speaking experience, visit johnengates.com. For speaking and selective advisory conversations, email me at contact@exagentica.ai.