AI makes code cheap. Delivering software that works is still the hard part.

Rusty Eddy writes for engineering managers, tech leads, and technical founders on delivering production software with AI: architecture, verification, review, and release discipline.

Your team can now generate more code in a week than it used to write in a quarter. What hasn’t changed is everything that decides whether that code survives production: architecture, boundaries, tests, review, and release discipline.

I’m Rusty Eddy. I’ve spent my career shipping software that had to keep working: space systems at Rockwell, PIM multicast routing at USC/ISI, Cisco’s Core Routing System, and now connected systems at Panasonic Avionics. Here I write about applying that discipline to teams that build with LLMs and agents.

This site is for engineering managers, tech leads, technical founders, and small-company CTOs: the people who own the outcome.

Start here

Deliver software with AI

Agent instructions, verification, independent review, and release gates that keep AI-generated work shippable.

Read the software delivery guide

Architecture for the real world

Boundaries, failure modes, and decisions written down, so the next engineer (or model) can follow them.

See the reference projects

Build AI systems

MCP servers and agent tools that give a model narrow, safe access to a real system.

Build an MCP server in Go