Imagine building your first AI enabled feature into your app. Your demo worked, your pilot work, and you've shipped to production.

Then the reports come in: users are unhappy and confused because you've vibe engineered your solution without actually doing the most important part - making sure it works before you ship.

In this session, Spencer will share ten rules for building AI systems that actually hold up in production, based on his experience building real AI applications for his clients. We'll go over why "just prompt better" isn't a strategy, how to build evals you actually trust, and when to constrain the model instead of trusting it.