I photographed a broken clip. Claude printed a new one
I don't do CAD. I took photos with a caliper in frame, and Claude traced them in Fusion 360, exported the 3MF and sliced it. Here is the photo recipe that made the second try fit.

Pedro Dias
Norwegian pragmatism · Portuguese stubbornness · Too much code
Acting Chief AI Officer at Tradesolution. Writing code since 1982. Too many strong opinions about software, AI and speaker placement.
I don't do CAD. I took photos with a caliper in frame, and Claude traced them in Fusion 360, exported the 3MF and sliced it. Here is the photo recipe that made the second try fit.
A model on my own PC sorted 725 news stories while you slept and found 90 about AI.
99 sources, in Norwegian Read today's briefing
Everybody knows someone who says "I'll just". Here is a self-test, two pictures of me, in clay and with a bow, and a floppy disk.
I asked Claude to grill me about writing for AI readers. We ended up keeping my voice and building a frame around it that agents can read.
I built a Claude Code mod that reads my first prompt, suggests a model, and scores the one I started on. It showed me how often I guess wrong.
I packed a local coding model for a flight with no Wi-Fi. It was slower, weaker and thirstier than I hoped. Here is the checklist and the ten-minute ground test I wish I had run first.
My PC reads about 60 news sources every night and keeps the few that matter for my job. You can build the same thing in an afternoon, and it costs next to nothing to run.
Chat models changed how we write. Jev is built to change how software decides. I couldn't wait for API access, so I had Claude Sonnet 5 build a local stand-in: real model, real bridge, real test harness.
A camera too far from the hub. A Raspberry Pi that refused to relay its signal. Three days of driver bugs and bad power later, I gave up and let its touchscreen do the one thing it was built for: show something useful on the wall.
The HLK-LD2410B detects people who aren't moving. Getting it wired to a D1 Mini and housed in a 3D-printed enclosure was exactly the kind of project where AI made the difference between a drawer full of components and a finished thing on my wall.
I run one of the beefiest consumer AI rigs you can build right now. Local models for image generation? Excellent. For coding against something like Claude Code? Not even close. Here's my honest breakdown.
I started on Windows For Workgroups 3.11. I spent seven years at Microsoft. And now I'm switching to Mac for work. Not because I'm unhappy with Windows. Because the gravity in AI development has shifted, and my job requires me to follow it.
Most people open Claude Code inside a single service. Open it at the parent folder instead: then give it the tools it needs to not go blind.