Longreads from the desk: what we build, what it costs to run, and what the machines are actually doing between one case and the next.
My answer to what an AI-native company is: one where colleagues are not split by which AI tools they have and how well they use them.
File of the day
A rule for your global CLAUDE.md: at the end of every task and plan Claude imagines the work has failed and deals with the causes before the handover.
File of the day
A Claude Code command finds three working businesses abroad run by one to three people on a neural network, with sources, and checks if your country has them.
File of the day
Once a month a Claude Code command reads your bank statement and names one thing to close, with the figure behind it. It cancels nothing by itself.
OpenClaw, Muse, dots: the 24/7 assistant trend is back. Why I am sitting it out, and what separates a cheap chatbot from an agent you can trust.
File of the day
Write that you did not understand, and a Claude Code subagent rewrites the last answer shorter and simpler, with the same numbers, conclusions and questions.
File of the day
Once a week a Claude Code command reads what is new in Claude Code, checks it against how you work and keeps only the advice that fits, with a ready request.
File of the day
A Claude Code Stop hook ends a reply with a short tone: a ringing one when the work is done, a lower one when your decision or action is needed.
File of the day
A Claude Code subagent reads a letter, post or offer as the person it is addressed to before it goes out and splits its remarks into blocking and optional.
File of the day
A Claude Code command reads the notes of the last day at night and adds to your knowledge base what will still matter in three months. It deletes nothing.
Your own Linux machine with Claude Code on it, working around the clock while your laptop is shut. From $3.49 a month.