Curated from writings on casey.berlin. Pieces that show how I think, build, and work, not the journal that lives next door.
The boring disciplines behind a data migration you can actually prove: irreplaceability order, checksum gates, offsite-restore tests, uid preservation, and verifying by function.
Continue on casey.berlin →A Raspberry Pi, an e-paper panel, and a battery that lasts because the device spends almost all its life switched off. The architecture, the design, and what the schedule taught me.
Continue on casey.berlin →Retiring a flaky, load-bearing box in an afternoon: DNS, public ingress, home automations and all, with nobody noticing.
Continue on casey.berlin →A field-tested runbook for moving Zigbee2MQTT from an ember (EFR32) coordinator to a zstack (CC2652P) one in place, including the undocumented channel_mask trap, a socket-contention root cause, and what an AI pair-operator actually contributed. With three postscripts: the beta-firmware auto-update ambush, the week-long hunt that ended in a wrong-board-variant firmware and a Wi-Fi AP squatting on the mesh's spectrum, and a 72-hour soak that finally turned the claim into proof.
Continue on casey.berlin →The infrastructure layer of MCP has already won. The mass-market moment has not arrived. That gap is exactly where the cheap experiments live.
Continue on casey.berlin →why mcp-readwise has fourteen tools, but the one that actually mattered ships markdown to your Reader Library by SMTP, and what that detour says about API design.
github.com/CaseyRo/mcp-readwiseContinue on casey.berlin →The gap between "I know exactly what this should do" and "I have working software that does it" has collapsed from months to days. You don't need to code. You need to specify.
Continue on casey.berlin →Five patterns that signal your business is ready for digital transformation, a practical self-assessment from 15+ years of consulting.
Continue on casey.berlin →How we built a proper state machine for our home automation, and why Home Assistant handles devices while Node-RED handles the logic.
Continue on casey.berlin →Mittelstand companies think they're behind on digital transformation. They're not, they're actually better positioned than enterprises. Here's why size, speed, and proximity to customers create an automation advantage.
Continue on casey.berlin →Rolling out AI with 'transform or die' energy breeds resistance, not adoption. How to help teams move from threat to agency in 2026.
Continue on casey.berlin →Nine practical rules for designing AI experiences that feel human, built on EY research into empathy, trust, and high-stress moments.
Continue on casey.berlin →Zero-click searches are replacing website visits. Here's how to stay visible when AI answers the question before anyone clicks.
Continue on casey.berlin →AI moves fast, but speed without direction is drift. Six bearings, from transparency to education, that keep your AI compass true.
Continue on casey.berlin →AI gets the headlines, but open source, low-code, serverless, and better hardware all bent the automation payoff curve. The xkcd math has changed.
Continue on casey.berlin →Fast, cheap, or good, pick two. AI bends the iron triangle by stretching speed and budget, but quality stays stubbornly human.
Continue on casey.berlin →