After years in the field, Pebble watches started failing—resources wouldn’t load, watch faces went blank, and support requests piled up. The culprit? A subtle timing issue with NOR flash wear that only surfaced as devices aged.

Pebble wasn’t built by industry veterans, it was built by hackers figuring it out as they went. Memfault founders François Baldassari and Chris Coleman, alongside Brad Murray of Beeper, dive into how Pebble’s startup culture led to a unique approach in firmware development.

A single unprotected region holding fonts changed everything. Memfault founders François Baldassari and Chris Coleman, along with Brad Murray of Beeper, discuss one of the most clever security bugs discovered through Pebble’s WhiteHat program.

When Pebble launched its SDK in 2012, it started as a pile of Python scripts. That was just the beginning. Memfault founders, François Baldassari and Chris Coleman, along with Brad Murray of Beeper, discuss the evolution of Pebble’s app sandbox, the challenges of early firmware development, and how a passionate developer community helped shape the platform.

Join the Founders of Memfault as they dive into this trend alongside special guest Alexander Samuelsson, CTO and Co-Founder of Imagimob (an Infineon Technologies company). This conversation on The Future of Edge AI and What It Means for Device Makers will explore how advancements in Edge AI are reshaping the embedded landscape, from hardware design to edge AI model development.

IoT margins are already tight—why make it worse? Many companies are throwing away money on preventable costs like unnecessary RMAs, bloated customer support, and costly technician visits. But there’s a better way: Observability and OTA updates can help reduce churn, cut support costs, and eliminate waste. We just watched a customer slash support tickets by 30% and RMAs by 50% using Memfault’s observability data. These are real numbers, real savings, and real impact.

Big tech has the resources, but startups have the real advantage in AI wearables: speed, agility, and the freedom to take risks. Right now, the AI wearable market is in the wildcard phase—no dominant device, no set form factor, and no clear winner. That’s a massive opportunity for smaller teams that can move fast, test in the field, and refine in real time. Unlike big tech, startups don’t need a five-year roadmap. They can launch quickly, experiment aggressively, and pivot without worrying about shareholders.

Meta is making a massive push into AI wearables, with at least six new devices launching in 2025. But here’s the catch—this wasn’t originally about AI. Meta built its hardware for the metaverse, only to find itself at the center of the AI revolution. With over 1 million Ray-Ban smart glasses already sold (and a goal of 5 million in 2025), it’s clear there’s demand. But can Meta actually scale this initiative from within, or will they lean on brand partnerships like Oakley to expand?

How can engineering teams have a bigger impact on the bottom line? By thinking beyond code. Most engineers love to build and solve problems. But in a business, building for the sake of building isn’t enough. Even the cleanest code is just an expensive distraction if it doesn’t move the needle.

Some IoT companies are making money; others are leaking it. Margins in IoT are already tight, but many brands are losing cash in ways that are completely preventable. RMAs, bloated customer support costs, churn, and on-site technician visits all add up. Too many companies default to replacing hardware instead of fixing the code. Without OTA updates and remote diagnostics, budgets get drained by unnecessary shipping and support costs.

AI isn’t predictable, it adapts, making embedded engineering even more complex. A model that works in the lab might fail in the real world. So, how do successful teams deploy AI at the edge? A/B test models in the field—controlled environments aren't enough. Collect real-world performance data—observability tools are key. AI deployment isn’t a one-and-done process. It requires constant iteration and real-world validation.

Fitbit was just fined $12M after Ionic smartwatches overheated and burned users. The issue? Lithium-ion batteries—powerful, but risky without proper safeguards. The best teams know you can’t catch every failure before launch. That’s why real-time monitoring is critical: Over-temperature protection isn’t enough without tracking trends. Live monitoring helps catch small issues before they become safety risks. Think about it: What if an e-bike motor overheats mid-ride? Or a smart oven malfunctions and starts a fire? Without monitoring, you’re gambling with user safety.

AI isn’t predictable, it adapts, making embedded engineering even more complex. A model that works in the lab might fail in the real world. So, how do successful teams deploy AI at the edge? A/B test models in the field—controlled environments aren't enough. Collect real-world performance data—observability tools are key. AI deployment isn’t a one-and-done process. It requires constant iteration and real-world validation.

Pebble OS is now open-source. Former Pebble engineers François Baldassari, Chris Coleman, and Brad Murray break down its innovations and why this release matters.

Security researchers found a massive flaw in Subaru’s remote vehicle system—hackers could unlock and track cars easily. The culprit? Homemade authentication protocols. Lesson: Don’t DIY security. Use trusted, third-party solutions. What do you think Subaru should have done differently?

Compliance documentation is a tedious but necessary part of embedded engineering. AI can help by automating structured tasks, drafting reports, checking for regulatory gaps, and validating submissions so that engineers can focus on development work. Would you trust AI-generated compliance documentation? Let’s discuss.

AI at the edge is built for embedded systems. And no need for tons of compute power— most of the heavy lifting happens during training so the models run efficiently on minimal hardware. With microcontrollers like STM32N6 optimizing for AI workloads, the potential is growing fast. Is AI at the edge part of your embedded strategy this year?