Affording your AI chatbot friends

Servers are expensive. Servers with GPUs are even more expensive. AI agents rely on servers with GPUs. If you don’t have control over what is happening at different parts of the stack, then things can change out from under you and your AI agent can change drastically without warning. Read: your AI chatbot friend can get massively depressed out of nowhere!

In this talk, I’ll cover all of the parts involved in a production-grade AI agent workload and how and where you can and should get control of them. This will cover the overall stack you’ll end up using, model management and the risks of models changing, cost-time tradeoffs and how to make educated decisions about them, as well as stories of my misadventures when things went wrong. The audience will leave this talk with practical strategies for maintaining control over their AI agent’s behavior and for controlling costs.