As semiconductor giants acquire edge AI startups and build neural processing directly into standard chips, what happens to the pioneers who helped bring AI onto our devices? In this episode, we explore how edge AI is shifting from a specialized product to an everyday feature. We examine why increasingly complex language models are challenging early chip designs, how software and developer communities are becoming crucial to survival, and why standalone accelerators are being squeezed out of the market. Could this consolidation bring faster, more private AI to everyone—or narrow the field of innovation just as it’s getting interesting? #embeddedAI #podcast buzzsprout.com/2429696/episo…

What does it take to turn a Raspberry Pi Pico 2 into a real-time, AI-powered USB microphone? Project 17 in my book, Embedded AI, brings together machine learning, digital signal processing, and embedded systems to tackle audio noise suppression. This video shows the journey from microphone to PC: - A T3902 PDM MEMS microphone captures audio. - PIO and DMA acquire the bitstream, ready for conversion into PCM samples. - Core 1 runs RNNoise to suppress noise. - Core 0 streams the denoised audio over USB using TinyUSB. - An IPS LCD displays the output spectrogram, while the onboard LED changes brightness with the model’s confidence that voice is present. Making it all work means coordinating two cores, moving data through ring buffers, and keeping the audio flowing in real time.

In 1985, the psychologists Garold Stasser and William Titus ran what became one of the most replicated experiments in group decision-making. Four-person groups had to choose among three candidates for student body president. Each candidate was described by sixteen characteristics. The trick was in the distribution. The best candidate’s virtues were scattered across group members so that no individual could see the full picture, while a mediocre candidate’s virtues were known to everyone. The complete evidence pointed clearly to one answer. Each individual’s data pointed in the wrong direction. This August, Anthropic ran the same test on teams of four AI agents deciding realistic questions: which candidate to hire, which investment to make, which property to buy. Shared evidence favored the wrong option. Each agent held unique, decisive private facts favoring the right one. Solving the task required an agent to recognize its private information as pivotal and press it against an apparent consensus. A single agent handed the entire evidence base got the answer right nearly every time. Groups of agents, after discussion, got it right in 17 to 36 percent of runs for most model families.  #EmbeddedAI aiadvances.org/groupthink-wh…

Subscribe to the Embedded AI newsletter for ongoing bonus content that extends the book, including new projects, hardware add-ons, and updates to the companion code. As a welcome, you will receive free links to a selection of my technical articles on Medium. These include the Chapter 13 Battery Monitor Shield bonus project and a guide to adding a 5-pin MIDI output to the Raspberry Pi Pico keyboard. embedded-ai.kit.com/profile

Tap a phone on the card and nine LEDs light up in sequence that emulates a neural network. The whole thing runs on whatever energy a mobiles NFC field can push into a loop of copper on a credit-card-sized PCB (about 5 mA at 2 V). The card is to promote my new book, Embedded AI: Intelligence at the Deep Edge (No Starch Press), and Reefwing Software, the company behind it. Front Side: the book title, my name and the LED network. Back Side: Reefwing branding, a QR code, the web address and three programming pads. Both the NFC tag and the QR code point at a short URL. medium.com/embedded-ai/build…

On 20 May 2026, an unreleased OpenAI model produced a counterexample to a conjecture Paul Erdős made in 1946. Nine world-class mathematicians reviewed the construction and found it sound. On 8 September, ten thousand agents working for 88 hours produced a singularity in the three-dimensional Navier-Stokes equations, formally verified in Lean, resolving a Millennium Prize problem. Neither result was in any training corpus, because neither existed. This episode asks what actually happened. If a model generates a proposition that was never in its training data, is that discovery or hallucination? The uncomfortable answer is that at the moment of generation they are the same operation. Nothing in the sampling process distinguishes a true novel claim from a false one. The partition is imposed afterwards, from outside, by a verifier the model does not run. buzzsprout.com/2429696/episo…

Light Lag: WEGO fleet tactics at the speed of light. Plot in secret, resolve at once. Your orders take turns to reach distant ships, and every enemy you see is where it was when its light left. Free alpha strategy game, Mac/Windows/Linux. reefwing.itch.io/light-lag

Why inference rate, not intelligence, decides which layer keeps the robot alive. Six blind men are asked to describe an elephant. The one at the trunk reports a snake, the one at the leg a tree, the one at the ear a fan. None of them is wrong. Each has a correct local measurement of a surface he cannot see the whole of, and the argument only starts when they compare notes. The same problem occurs when you try to describe intelligence and even for what should be a very straight forward concept — bandwidth. medium.com/embedded-ai/you-n…

Embedded AI author David Such (@reefwing) wrote a bonus article about designing a custom PCB that handles battery charging and discharging while measuring voltage, current, temperature, and state of charge. Read it here: medium.com/embedded-ai/embed… and preorder Embedded AI at nostarch.com/embedded-ai 🧠

What does a 741 op-amp teach us about training networks on data that is 99 per cent wrong? “Garbage in, garbage out” is a half truth in any system with feedback. A feedback loop does not need accurate components. It needs an accurate reference. Whether that distinction carries over to training and running neural networks turns out to be a well studied question, and the answer is the same as for the op-amp, right down to the caveat. briefs.aiadvances.org/garbag…

A large paint company documents four ways to process a purchase order. In reality there are 11,973 different approaches. That gap between the documented process and the real one is the subject of this episode, and it explains why AI adoption is near universal while the share of companies reporting any profit impact has been flat at 37 percent for two years. We look at where the difficulty actually sits. The process lives in workarounds, spreadsheets and people's heads, and is different at every site. The data is spread across roughly 900 applications of which a quarter are connected. Legacy systems were never designed to be called by a machine. And the law has started to get involved: Air Canada was held liable for what its chatbot said, Workday can be sued as an employer's agent, Australian companies must disclose automated decisions from December 2026, and Commonwealth Bank reversed AI-attributed redundancies after call volumes went up rather than down. We then cover what AI adds to the old problem. Language models are non-deterministic and cannot be replayed for audit. The best agents complete about 30 percent of realistic office tasks. Human oversight runs into limits Lisanne Bainbridge described in 1983, now measured in the field. And people feel faster while being measurably slower. Finally, what works. The strongest predictor of financial return is not the model but whether the workflow was redesigned, which is the same lesson factories took forty years to learn from electrification. We walk through the method: mine the process, standardise the core, exclude the tail explicitly, run in shadow mode against a baseline, measure at the process level, and tier governance by consequence. buzzsprout.com/2429696/episo…

Five of the projects in my book on Embedded AI are built using custom PCBs: the Noise Suppression carrier board for the Raspberry Pi Pico 2, the Pico MIDI Keyboard and VS1053 Synthesizer that form the AI music project, and the Battery Monitor and Display & Logging shields that make up the battery characterization rig. These are circuits that do not work well on a breadboard, so each was designed as a proper board with the Gerbers, schematics, and bills of materials released under the MIT license. Getting them manufactured is straightforward: all five are shared projects on PCBWay with the fabrication files already attached, so ordering takes a few clicks and boards typically arrive within one to two weeks. To make building the projects even more affordable, PCBWay is offering readers a discount. PCBWay is offering 200 readers of Embedded AI a $10 discount on orders over $30 (one use per user). Redeem the code EmbeddedAI2026 at checkout. The code is valid until April 30, 2027 (once activated, it must be used within six months). Each board costs around $5 but shipping is roughly $25 (depending on your location), so order the boards together to minimise freight, and the combined order will clear the $30 minimum. pcbway.com/project/member/?b…

This article extends Project 32: Build and Test the Battery Monitoring Shield in my upcoming book on Embedded AI, scheduled for release in October 2026 by No Starch Press. This project includes a custom printed circuit board (PCB) which handles battery charging, a programmable load for discharging, and a fuel gauge that measures voltage, current, temperature, and SOC during the charge and discharge cycles. medium.com/embedded-ai/embed…

There is an old claim that a truly random source contains all knowledge: give monkeys enough time at typewriters and Shakespeare falls out. In 2024 two Sydney mathematicians did the arithmetic and found the universe ends first. But the idea has a modern tail. We now have language models that can spot meaningful text instantly, so why not let randomness generate and an LLM extract? This episode works through why that fails, and why the failure is precise: in a random stream, the address of any text costs as many bits as the text itself. Along the way: Borges' Library of Babel, a website that actually built it, DeepMind systems that made the generate-and-filter idea work by cheating in exactly the right way, and what your brain does with noise that an LLM cannot. buzzsprout.com/2429696/episo…

While vibe coding my latest application I got to thinking about whether someone needed to create a computer language which explicitly targets LLMs. Assuming we get to a point where you don’t need to review the underlying code (a big assumption I know), then does the language used matter as long as you get the required outcome? If a human never reads the code, the argument goes, why generate syntax built for human eyes at all? Proposals are being circulated for machine-native formats. For example, JSON programs, semantic graphs, and languages with grammars minimised for constrained decoding. The premise is that human readability is an obsolete requirement we can now delete. I disagree. levelup.gitconnected.com/the…

Chapter 13, Project 32 - Building a Battery Monitoring Shield: The I2C issue. Both shields worked perfectly on their own. Stacked together, the OLED kept running while the fuel gauge spewed 0xFFFF — on the same bus. Why? Display traffic is write-only, so the master does all the driving and a marginal bus limps through. Reads are different. The sensor has to drive data back within spec, and that's what a weak bus kills. Root cause: the display library silently switches I2C to 400 kHz, and a lone 10 kΩ pull-up can't reach a valid logic high in time. Fast mode wants 2.2–3.3 kΩ. If you ever see a working display next to a sensor reading all-ones, you now know where to look. Project 32 from my book Embedded AI.

Anthropic gave thirty AI agents the same coding task. Eighteen of them named their git branch exactly the same thing. That result opens one of the strangest findings of 2026: when teams of AI agents face the classic "hidden profile" experiment, where the shared evidence points to the wrong answer and the decisive facts are scattered across individuals, they succeed only 17 to 36 percent of the time. Human groups in the original 1985 study? 18 percent. Forty years, a completely different kind of mind, the same failure. In this episode we dig into why. We trace the mathematics of groupthink from information cascades to Condorcet juries, then follow the trail into territory embedded engineers know well: the 1986 Knight and Leveson experiment that shattered the independence assumption in N-version software, Airbus's dissimilar redundancy, and the Lufthansa flight where two frozen sensors outvoted the one telling the truth. Along the way, honeybees show us a working reference design: a two-milligram brain that refuses to repeat a rumor. We close with the fixes: engineered dissenters, forced disclosure protocols, reputation infrastructure for agents, and the case for "keeping the weirdness alive" through a genuinely diverse AI ecosystem, including heterogeneous fleets of small models at the edge. buzzsprout.com/2429696/episo…

Embedded AI - Chapter 13, Project 32: Thermal Design. 7 watts has to go somewhere. During a 500 mA discharge the MOSFET dissipates the difference between battery voltage and the sense drop as heat — and the heatsink hit 72 °C, which works out as 27 °C/W against a spec of 28. The design math predicted this. The thermometer agreed. Junction sits near 75 °C, half its 150 °C limit, and the LiPo stays at room temperature. If your battery ever gets warm during discharge, disconnect immediately. Project 32 from my book Embedded AI.

Chapter 13, Project 32 - Building a Battery Monitoring Shield. The build is simple with a custom PCB. A MOSFET, stacking headers, and a 1 Ω sense resistor, everything else is already on the PCB. Project 32 from my book Embedded AI. The PCB handles battery charging, includes a programmable load for discharging, and a fuel gauge that measures voltage, current, temperature, and SOC during the charge and discharge cycles.

What happens when work no longer gives our lives structure, identity, or meaning? This episode explores “Purpose as a Service”—the idea that purpose could be intentionally designed and delivered in an age of mass automation. We examine the decline of traditional anchors such as employment, religion, and community; the rise of wellness programs, professional coaching, and AI companions; and what universal basic income trials reveal about the limits of financial security. Can external services genuinely help people build fulfilling lives, or does true purpose depend on personal agency and authorship? Consider this a blueprint for one of humanity’s biggest future challenges: learning how to live well when a job is no longer at the center of life. buzzsprout.com/2429696/episo…