Start here

The archive runs newest-first, which is a reasonable default and a bad way to meet someone’s writing. This is the curated version: what I’d hand you first, grouped by what it’s actually about.

Leading engineers

#

The craft of the job, and the coaching half I keep finding useful in it.

• Your interview loop is measuring the interviewers. A loop without a rubric doesn’t measure the candidate. The four stages I’ve landed on, and the one piece I changed my mind about.
• “I don’t know yet” is a senior move. Deferring the answer is a status, not an exit. When to use it, and the script that keeps it honest.
• Disagree and commit, in code review. When to push back, when to merge with a question, and when to actually block.
• Five coaching primitives every engineering leader should learn. The coaching moves that transfer cleanly into engineering, without the vocabulary that gets you side-eyed.
• The 1:1 you’re not having with yourself. Everyone gets a manager’s attention except the manager.
• Nobody makes a principal engineer do the basic form. Beginner’s mind as a budget you spend deliberately, learned mostly on a mat.

AI in real teams

#

Less about the models, more about what happens to a team using them.

• When the dashboard goes green. Every delivery metric improved, and that’s the trap. What went up, what quietly didn’t, and why a leader watching only velocity banks hidden debt.
• Faster to build isn’t cheaper to own. Generated code has the same maintenance cost as code you wrote, and nobody budgets for it.
• The test AI didn’t write. I covered for a QA bottleneck with an AI-built test plan. It missed the exact thing a real QA would have caught.
• The reps you skip are the intuition you don’t build. The struggle AI removes is the struggle that built your judgment. When to deliberately do it the hard way.
• The hard part was never the wording. AI will draft the difficult message for you. The difficulty was never the draft.
• The agent ran for twenty minutes. What supervising an agent is actually like, and what it does to review.

AI, hands on

#

The practical end, with the trade-offs named.

• When is a local LLM actually the right tool?. Most of the time the local model loses. Here’s the narrow band where it wins.
• When the meter starts running. What token costs do to how you work once you can see them.
• The week my model went dark. An outage, and what it revealed about how much I’d come to depend on it.
• AI made starting free. Finishing is still yours.. Spinning up ten ideas got trivial. The switching cost didn’t move.

Tools and systems that stuck

#

Things still in use a year later, which is the only review that counts.

• Years of editor-hopping: what I actually kept. Zed, VSCodium, Void. What survived the hopping.
• Dotfiles demystified. Homebrew and GNU Stow, and a setup that survives a new laptop.
• Why I take daily notes (and how I do it in Obsidian). The note system that outlasted the enthusiasm.
• Customizing Blowfish without forking it. Four tiers of Hugo theme customization, and why the note you leave yourself matters more than the override.

The off-hours

#

Not filler. Some of the best thinking about running a team arrives from well outside work.

• What the GM’s chair taught me about running a team. Session zero as kickoff, and the players who hijack the plot.
• Nothing new on race day. Why you don’t try the new thing when it counts.
• Build an HTPieC. A Raspberry Pi doing more jobs than it should.
• The ultimate guide to Calibre. The ebook Swiss Army knife, documented properly.

Everything else is in all posts and the TILs, which are shorter and weirder. If you want the summary of who’s writing this, that’s about; what I’m doing at the moment is on now.