Live AI Operationsaihabits.ai

About

Hi, I'm Myles Saulibio.

I write Live AI Operations. It's about one narrow question: who has the authority to stop an AI agent, and how fast does that stop really work?

Lots of people are working on getting agents to do more. I spend my time on the other end. When something goes sideways, can a person halt it, and what keeps happening after they try?

I run stop drills on my own setup and write up what actually halts and what doesn't. Usually it's the boring stuff that bites: a token that's still valid, a scheduled job nobody remembered, a helper agent that never got the message.

A few things I keep coming back to

  • Reads keep going. Writes and spend halt.
  • A rule the model has to remember can fade. A switch outside the model doesn't depend on its attention.
  • Getting an alert about a problem and having the power to stop it are two different things.
  • Give each agent its own credential, so you can pull one without breaking the rest.

Who I work with

Operators at small and mid-size companies who already run AI agents that can take actions: send messages, write to systems, spend money or change records. If your agent only answers questions, you probably don't need me yet.

Where to start

Start with the posts, or follow along on the newsletter and on LinkedIn.

If you want a quick read on your own setup, take the Brake Drill. It's free, it takes about two minutes, and your answers stay in your browser. If the score surprises you, book a 15-minute read-back and we'll look at it together.

Personal views. Not legal or compliance advice, and not a security certification.