I used Opus 4.8 to create the refactoring plans. I converted token savings using Sonnet 5 pricing at the time of writing as I'd expect to use that for more 'straightforward' feature additions.
That's what I've shown in this article. The token consumption shown is for adding a new feature (the same feature each time -- in a sub-agent) after every refactoring is applied. I didn't actually capture the token cost of the refactoring. That was a miss, and I'm making sure I do that in the future.
Kind of you to engage! Ahh, ok, I misunderstood. Very cool! I think maybe it would have helped me if you had given a one-or-two-sentence description of the "representative change" in the main text, to make it a bit more concrete up-front? But possibly just a 'me' issue...
Yep, this was a real surprise to me, too. That’s why I included that top line in the figure. It also doesn’t line up with the agent’s predictions when it laid out the refactoring before proceeding.
There’s further work to do to understand exactly what’s going on here.