Most agent demos skip the question: what does an agent actually cost to run? A REST call costs the same every time. An agent doesn't, since the same prompt varies with reasoning depth, tool calls, retries and context growth. We'll map what you really pay for in Microsoft Foundry beyond model tokens and then look at several architectural options for controlling spend.. plus when NOT to build an agent at all. Expect some live demos in AI Foundry portal, and you'll leave with a FinOps checklist.

Gartner expects 40%+ of agentic AI projects to be cancelled by 2027, and cost is a the biggest reason. A traditional app has a cost you can plan; an agent has a cost distribution you can only govern. Learn to design Azure AI agents for cost-per-outcome before the invoice, not after.