A Bullet Journal video gave me an idea, and a few hours later (between diaper changes) it was a working feature in Kombine. A year ago that took a swarm of agents. Now it takes a design system, CI and the engineering basics nobody finds exciting.
In June, I learned how to change diapers and convinced myself that 5 hours of sleep is a luxury. Somewhere in between, I also shipped iOS for Kombine — which was absolutely not on the roadmap. The roadmap said Android → Backend → Webapp → iOS sometime later. It was a good plan, and that’s exactly the kind of plan a pet project doesn’t have to follow.
I built an AI-augmented development pipeline that starts with a Telegram message on the toilet and ends with a merged, tested PR — no keyboard required. This post walks through five stages of evolution, from copy-pasting into ChatGPT like an animal to running a self-hosted ‘scrum team’ of LLM agents on a NAS in my living room. It’s not the shiniest setup, but it’s mine — and it works.
I find myself with a fresh golden hammer in hand every now and then, only to realize a screw would’ve been better. Following principles blindly doesn’t automatically render ‘better code’—there are always tradeoffs, and managing them is a sign of seniority and actual craftsmanship.
Most software teams are split up by technical layers—backend, frontend, and mobile. This seems logical, but it often creates bottlenecks, delays, and frustration. Mission teams flip the script: cross-functional developers own features end to end, from idea to release. This approach speeds up delivery, builds real ownership, and cuts down on endless handoffs. Too perfect to be true? It demands more upfront effort in communication and shared knowledge, but the payoff is worth it.
My wife needed a candle calculator app. I only know Android. So I used Claude Code to build a native iOS app without writing Swift. 19 days, 3 Apple rejections, and 5 hours of certificate hell later—it’s live, it works, and she’s happy.
We’ve been thinking about AI development backwards. Instead of building “AI-powered” apps, we should be building tools that power AI. The Model Context Protocol lets you create small, focused tools that LLMs can use autonomously - just like Unix commands from 1978, but orchestrated by Claude instead of bash. I built a currency converter tool and watched an LLM decide when to use it, extract dates from user prompts, and combine multiple API calls without me writing any of that logic. It’s the same composable philosophy that made Unix great, applied to making AI actually useful in the real world.
A critical look at AI-augmented coding after one month of hands-on experience with Cursor IDE. Exploring the promises versus reality of AI assistance in Android development with Kotlin and Jetpack.
Building a split ergonomic keyboard from scratch: a journey through 3D printing, soldering, and firmware programming. Despite the learning curve and typing speed challenges, discover why creating your own mechanical keyboard becomes an addictive hobby.
A chance encounter at the airport leads to insights into the real nature of technical debt in software development. Through the lens of Stoic philosophy and Lord of the Rings metaphors, this post explores how better communication between engineers and management can prevent our own Mines of Moria situations. Practical tips included for both technical leads and managers on how to discuss, document, and decide which technical dragons are worth slaying.