Embedded AI Podcast E07: Embedd, and using AI safely, with Michael Lazarenko

Michael Lazarenko, co-founder of Embed, tackles embedded development's messiest problem: the gap between hardware docs and usable software. He shares how they use AI to parse datasheets - and crucially, when NOT to. Parsing STM32 or TI datasheets is painful: inconsistent formats, critical info buried in PDFs, different taxonomies. Michael explains creating standardized intermediate representations that can be deterministically validated, using AI only where it adds value. Ryan and Luca discuss AI adoption in conservative industries like automotive and aerospace, exploring why enterprise quality expectations clash with AI's probabilistic nature and how using AI forces better engineering practices - especially testing. Luca shares becoming "such a stickler for tests" since using AI. Michael shares Embed's vision: a holistic embedded IDE connecting hardware design tools, auto-generating abstraction layers, bringing modern CI/CD to firmware. They're launching an MCU/MPU configurator in early December. 0:00 Welcome to Embedded AI Podcast 1:05 Hardware Software Decoupling Problem 6:18 AI Parsing Data Sheets 7:31 When NOT to Use AI 11:24 Choosing Between Deterministic and AI 16:48 Journey Through Trial and Error 24:27 Edge AI in Enterprise Devices 32:00 Enterprise Quality Expectations and Challenges 38:26 AI Forces Better Testing Culture 42:41 Future of Embedded Development Workflows 48:00 Vision for Holistic Embedded IDE 51:53 Product Launch and Wrap Up