We talk IoT · 2026-04-09 · 37 min
Edge AI promised intelligence everywhere. The reality? Most projects die between proof of concept and production. In this episode, Amir Sherman from DeepX and Michaël Uyttersprot from Avnet Silica reveal why moving AI from comfortable development kits to demanding industrial environments isn't just difficult. It's a maze of incompatible metrics, hidden power costs, and integration nightmares that catch companies off guard. We discuss why TOPS ratings mislead engineers, how ChatGPT triggered a wave of failed internal deployments, and what it takes to run vision AI in factories, delivery robots, and smart cities where five-watt power budgets matter more than marketing specifications. From Hyundai's factory robots to Baidu's Chinese character recognition systems, Amir and Michaël share real deployments that work, and explain the 50 years of embedded experience that AI code generators cannot replace. If you've wondered why edge AI keeps hitting walls nobody discusses in vendor presentations, this conversation delivers the answers.
From the cab to the cloud: How AGCO Fendt is making smart farming real - We Talk IoT #8669 / 100No app required: The NB-IoT pill dispenser changing medication management in care homes - We Talk IoT #85Change the OS, stop the train: Why industrial software stability is non-negotiable - We Talk IoT #84Don’t digitise the waste: How signal towers became IoT devices - We talk IoT #82Chips, Grit, and "Girl on Fire": Emily Yang on Leadership, Mentorship, and Moving Up - We Talk IoT #81