About

A practical field log
from someone who actually
runs agents.

Living with Agents exists because there is surprisingly little honest writing about what happens when AI agents stop being a demo and become part of your daily operating system.

The premise

Most writing about AI agents falls into two camps: breathless hype about the coming revolution, or academic papers that ignore the duct-tape-and-determination reality of production systems. This publication is neither.

Living with Agents is a working field log — the day-to-day notes, experiments, recoveries, and pattern extractions from someone who builds and runs autonomous AI agents in his own life and work. It covers:

• Agent architecture and tool-calling patterns that actually hold up
• Prompt engineering that works at scale, not just in a notebook
• The mistakes, regressions, and “why did I do it this way” moments
• Evaluation, observability, and the hard problem of knowing if your agent is working
• Tooling and infrastructure builds you can borrow and adapt
• Honest takes on what AI can and cannot do right now

About Peter Foti

I'm a builder and operator who has been designing, deploying, and — most importantly — living with autonomous AI agents since they became practically viable. My background spans product engineering, systems design, and operational infrastructure at the intersection of human workflows and AI automation.

I started Living with Agents because the most valuable writing about agents was happening in private Slack groups and hallway conversations. I wanted to pull that signal into the open — a public, honest, iterative record of what it really means to run agents outside the demo, over months and years, not hours.

The publication is named for the experience most builders don't talk about: the quiet morning when you realize your agent has been running unattended for a week, and you have no idea if it made any good decisions. That's the territory this publication maps.

Follow along

The newsletter comes out roughly weekly-ish. No spam, no AI-generated filler, no sponsored content. Just the field log, as it happens, with the mistakes left in.