The task at hand is drawing the circuit schematics for a robot I'm working on. I had already written down the components and the connections, all that's left is to draw it in KiCad. I had already started doing that, but then... I got sidetracked forcing Gemini to create the circuit using KiCad. I would have made progress if I had continued doing it by hand. I spent yesterday trying to generate an
More rules should mean better output. That's the intuition. I spent weeks building a comprehensive CLAUDE.md — 200 lines covering naming conventions, security rules, error handling, architectural patterns, import ordering, type safety requirements, and more. I was proud of it. I'd thought through every scenario. Then I scored the output. 79.0 / 100. My carefully crafted documentation was actively
Comments
This is part three of a series on display consistency in embedded systems. The first two parts were technical. This one is about why the technical parts worked. The picture: ATtiny85 thermometer. Neural network inference. QUAD7SHIFT display. Built from datasheets. He had datasheets. No Stack Overflow. No libraries to install. No AI to generate boilerplate. No tutorials that abstracted away the in
Have you ever looked at code you wrote six months ago and thought: "Who wrote this monster?"? Relax, it happens to all of us. In software engineering, writing code that a machine understands is the easy part. The real challenge is writing code that other humans (including your future self) can understand, maintain, and scale. This is exactly where Software Design Principles come into play. In this
If you've ever used a bottleneck calculator, you've probably seen a simple percentage telling you whether your CPU or GPU is holding your system back. But here’s the truth most people don’t realize: Bottlenecks are not fixed numbers — they are dynamic, workload-dependent behaviors. In this post, we’ll go beyond basic tools and break down how CPU and GPU bottlenecks actually work in real-world scen
Part 1 of 5 in The New Engineering Contract — what it means to lead engineers when AI is doing more of the coding. SWE-CI tested 18 AI models across 71 consecutive commits. Most broke something on commit 47 they'd already broken on commit 1. That's not an intelligence problem. That's a learning system that isn't learning. A paper made me uncomfortable this month. Not because of what it found about