“AI Agents” are at the top of the hype curve. New protocols like A2A and MCP bring potential for service exposure to complex systems and their interaction via LLM-driven agents. Those agents can consume contextual real-time information, execute tasks, and interact with other external services or tools over different protocols.

This talk will explain the way we have built a ReAct AI Agent that can autonomously interact with IoT environments using a HATEOAS-like approach over CoAP. The system does not depend on middleware like MCP, instead it uses well-defined entry points to dynamically discover and interact with devices directly at run-time to fullfill user-intents. Lastly the presentation will introduce potential standardization areas in this domain.