19 June 2026
Claude Code in my repo. Open weights at the door.
How I use Claude Code day to day — plus the week AI started borrowing billions.
Dear Reader,
Two posts this week: one practical field note from my own desk, and one rather large week in AI news. The first is about handing real work to an AI assistant. The second is about who actually owns the assistants, the chips, and the debt that paid for them. They turned out to be the same conversation from two ends.
How I Actually Use Claude Code: CLI, Desktop, Diffs, and Blog Workflows https://daehnhardt.com/blog/2026/06/15/claude-code/
I expected another AI coding assistant — autocomplete with opinions. What I got was something that walks into the repo, reads files, runs commands, fixes the failing tests, and shows me the diff. Think of it less like a clever calculator and more like a capable apprentice: you stop dictating individual keystrokes and start handing over whole jobs, then review the result.
The shift that mattered was not learning a command. It was changing the size of the task I was willing to delegate.
The piece that earns its keep is CLAUDE.md — a single file in your project root holding your standing instructions, rather like leaving a note for the apprentice pinned to the workbench so you never explain the house rules twice.
Pair that with the -p flag, which lets you pipe a git diff or a log file straight into Claude like any other Unix tool, and the glue work that used to eat an afternoon quietly disappears.
The most important habit in the whole post is knowing when to slow the apprentice down. Database migrations, auth logic, billing flows — anything you cannot easily undo — start in Plan mode, where Claude shows you its intended steps before any run.
Being the adult in the room is still part of the job: Claude handles the glue, you keep the judgment. There are also a few seven-day guest passes tucked inside the post, if you would like to lend someone a try before they commit.
There are also a few seven-day Claude Code guest passes tucked inside the post. They use my referral link and include a week of free usage, but please check the terms before signing up — you may need to cancel in time if you do not want to continue. No pressure at all.
Meanwhile, the question of who you are really renting all this capability from was playing out at a rather grander altitude.
Open Weights, Big Debt: AI Weekly Signals, 13–19 June 2026 https://daehnhardt.com/blog/2026/06/19/open-weights-big-debt-ai-signals/
A quieter week for model launches, a much louder one for the money and machinery underneath them.
Z.ai shipped GLM-5.2 with open weights, a one-million-token context window and an MIT licence — and on selected long-horizon coding benchmarks, it landed uncomfortably close to the closed frontier. For an agentic pipeline making thousands of calls a day, cheaper inference is not just a discount; it can be a different business model.
My old rule of thumb — open weights mean accepting a capability gap in exchange for control — suddenly looks rather threadbare.
At the same time, the AI build-out revealed its true scale on the balance sheet. SpaceX lined up a bond sale of at least $20bn to refinance xAI-related bridge debt, which is what frontier compute looks like once the venture cheques stop covering it: you fund it like a railway or a power station, not a startup.
Meanwhile, Google is turning its TPUs into a commercial infrastructure business for customers, including Anthropic — a tidy irony, given Anthropic competes with Gemini.
Frenemies with a purchase order, as I put it.
Then came the part that turned all of this from theory into a fire drill. After a US export-control directive, Anthropic suspended access to Claude Fable 5 and Mythos 5. Whatever the final policy outcome, the operational lesson is already clear: hosted model access is not the same thing as ownership.
Elastic, sensing where the pain now lives, bought an AI site-reliability startup to help operate the mountain of code these models produce.
The thread tying all four stories together is ownership: what do you actually own, and what have you merely been granted?
Take care, and do hit reply if anything here sparked a thought.
Elena
Enjoyed this? Get more like it.
Weekly notes on AI tools, Python, and what I'm actually building — plus two free gifts: the 15-page Fantastic AI: The 2026 Toolkit and a Git Commands & Contribution Workflow Cheatsheet.