Building AI Systems that Ship
Resources for my talk on why AI agent systems break outside controlled environments, and how to make them ship.
Building AI Systems that Ship
Resources from my talk on what it takes to get AI agents working outside a controlled demo. Let’s connect on LinkedIn and BlueSky.
Presented at AI Engineer Europe 2026
Abstract
Most agent demos break the moment they leave a controlled environment. The model works fine. The systems around it don’t. In this talk, I’ll walk through what “agent-ready” means from two sides: a codebase agents can safely change, and a product agents can reliably use. I’ll use two real projects as examples: Case, a harness for orchestrating coding agents, and the WorkOS CLI. Together, they show how to give agents real freedom while keeping quality high through clear constraints, machine-friendly interfaces, and feedback loops that compound over time.
Resources
- Case Statement: Building a Harness — the deep dive on Case
- AIE Europe recap