AI agents do not fail because they cannot reason. They fail because the cost of their behaviour is not designed upfront. What begins as a simple agentic workflow with a planner, tools, and retrieval quickly grows into a system of chained calls, recursive loops, repeated memory access, and cascading model invocations. Each step may appear inexpensive, but at scale the combined cost becomes significant. For enterprise systems, this unpredictability is a critical risk.

This session examines the unit economics of agentic AI systems in production. Drawing from real deployments, it identifies where costs accumulate, including orchestration overhead from uncontrolled reasoning loops, excessive tool calls due to poorly defined boundaries, context window growth from unmanaged memory, and escalation from multi-agent coordination. The session also presents practical approaches to address these challenges, including agent loop budgeting, tiered model routing for sub-tasks, caching strategies for tool outputs, and observability patterns that make spend predictable.

What You Will Learn

• Where cost accumulates in agentic AI systems, including loops, tool calls, and memory usage
• How to design agent workflows with budgeting, routing, and caching to control spend
• How to apply observability patterns to make AI agent costs predictable at scale

Who Should Attend

• Software developers building AI agent systems
• AI and machine learning engineers
• Platform and infrastructure engineers
• Software architects
• Engineering and technology leaders responsible for cost and scalability