Occasional data analysis and experiments from Dave Baiocchi. Each one is a self-contained page, built for the fun of the question.
Six months of my own Claude Code logs. Output per response jumped fifteen times over as new models arrived while my prompting rate held flat. Then the per-response number settled and my prompting rate quadrupled, which is a different story with a different cause.
I pay a flat monthly rate, so I had never seen what my AI habit actually costs. Pricing six months of real usage at published API rates put the gap at sixteen times what I pay, and widening every month.
A visual picker for starting a new page in the right look. Clone a whole template, or compose your own from a palette, a font pairing, and a layout, and watch a live mock redraw as you choose.
The newer the model, the stronger its own voice, and the harder it is to dislodge. Seven practices that fight the model’s defaults, described in enough detail that you can re-create any of them. Plus the one that needs no tooling at all.
Agent, chat, or automation? A simple way to tell three AI tools apart, on two questions: does the task need judgment, and does it need hands?
Three AI capabilities running in my one-person practice: my full working memory served to my phone, fully local models for data that can’t leave the room, and Claude driving a real browser.
Six months of Friday Brooks & Capehart segments on PBS NewsHour, one cartoon speech bubble per week: the moment Jonathan Capehart breaks out of pundit mode, in his own words.
All 194 national days on one calendar: 193 UN members plus Taiwan and Kosovo, from a question my daughter asked on the Fourth of July.