AI agents are getting added to everything, but not every problem needs an agent. In many cases, a deterministic workflow, a retrieval-based assistant, or plain software is simpler, cheaper, safer, and easier to operate.

This session gives engineers, architects, and tech leads a practical framework for choosing the right approach. We’ll compare agentic systems, workflows, and conventional software across uncertainty, tool use, latency, cost, observability, security, and failure handling. Using an anonymized real-world style example, we’ll examine the same business problem implemented three ways and show the trade-offs clearly. A live demo will walk through how each approach behaves, where each one breaks down, and how to recognize when an agent is adding value versus just adding fragility.

Attendees will leave able to explain the differences between these patterns, compare them using concrete criteria, evaluate fit for real use cases, and avoid overengineering. This is not a hype talk about autonomous magic, it's a practical guide to choosing the simplest architecture that can do the job well.