Insights, tutorials, and lessons learned from building software. Dive into our thoughts on development practices, tools, and technology.
When you build a feature with an AI, it inherits all your blind spots. Here's how I use Claude Code subagents to get a fresh perspective on my code before committing.
GitHub issue filing on a phone is miserable. Here is the OpenClaw setup I use instead: dedicated Signal number, sender allowlist, narrow GitHub access, and a custom issue-writing skill.
I don't learn well from a pile of tabs. So I built a workflow: Claude finds and sorts sources, notebooklm-py pushes them into NotebookLM, and Obsidian gets the final distilled notes.
I kept copy-pasting the same instructions into Claude Code every single session. Then I discovered skills - and realized they do way more than just save you from repetition. Plus: skills aren't locked to Claude. They're an open format that works across AI coding tools.
Million-token context windows sounded like the dream. Throw everything in, let the AI figure it out. After losing weeks to mysterious failures, I finally understood why more context often makes things worse.
Six months ago, my AI coding workflow was a mess. Endless iterations, code that was 'close but not quite right', and a lot of wasted time. Here's how writing specs first changed everything - and the two tools I actually use now.
We share insights from building software for clients worldwide. Have a project in mind? Let's talk.
Get in Touch