IoT has solved the sensor and connectivity problem at scale. The next

bottleneck is no longer how to read a sensor or send a packet — it's what

happens between the data and the action. Most embedded-AI tooling today covers only inference; teams then stitch together separate pipelines for model deployment, telemetry, OTA updates and multi-device coordination, typically rebuilt from scratch for every device family.

This talk introduces a complete lifecycle stack for autonomous edge agents running on IoT devices: build visually in weeks, deploy with one workflow from MCU to Linux gateway, monitor with AgentOps-grade telemetry, coordinate multiple agents via a local discovery registry, and enforce a governance layer for permissions and audit — all 100% local when privacy or data sovereignty demand it.

The session walks through three concrete production scenarios:

(1) an agent running on a battery-powered field node that decides autonomously when to act and when to escalate to a gateway;

(2) coordinated multi-agent inference across distributed devices via a local discovery registry — no cloud roundtrip required;

(3) OTA agent updates with rollback under bandwidth-constrained and intermittently-connected conditions.

Attendees leave with a clear mental model of which lifecycle layers are typically missing in current IoT-AI deployments, why a registry-based multi-agent topology beats centralized cloud orchestration for embedded contexts, and a practical checklist for evaluating whether their own stack is production-ready.

Aimed at IoT architects, embedded engineers, and platform owners scaling past pilot deployments.