Enterprise auth for AI agents is the unsolved problem

When you give an agent tools, you give it credentials, and that's where enterprise security starts breaking. I wrote about this because it's the question I get asked most by clients.

The short version: an agent acting on your behalf is a whole new identity surface. Least privilege, scoped tokens, audit trails — it all needs rethinking when the 'user' is a loop that can run unattended for hours.

Curious how teams are handling agent identity today — service accounts? Short-lived tokens? Something custom?

I wrote this up in more detail in the full blog post — The Convergence of AI Agents and Enterprise Authentication Security — with the deeper context, data, and code. Give it a read if you're curious.