The XKCD Math

Last updated: March 23, 2026

XKCD 1205: "Is It Worth the Time?" A table of how often you do a task, how much time you save, and whether the automation investment pays off over five years. Simple math. Except the inputs changed.

Quinn Slack pointed out the same idea applied to AI coding: the time you invest in your harness pays off across every future session. The xkcd table assumes you do the automating. What if the agent does?

The meta-skills math: about 15 minutes saved per skill. Across 38 skills, that's roughly 9.5 hours. Plus every future skill. The xkcd math says this is worth it if you create skills frequently enough. I do.

What changed: the cost of automation dropped. When the agent scaffolds new skills, when the daemon automates summarization, the "time to automate" column in the xkcd table shrinks dramatically. Automation that used to take an afternoon now takes a prompt.

Weekly cost efficiency improved about 30% over time, from $0.079 to $0.054 per tool call. The system gets cheaper to run as skills and caching compound. The xkcd table has a time axis. This system has one too, and it slopes down.

The honest caveat: I'm treating the total spend as tuition, not overhead. The economics page has the numbers. The xkcd math works for the mechanical parts: summarization, skill creation, deployment automation. For learning this way of working, the calculation is different. It's an investment in understanding, not a time-saving optimization.

The xkcd table asks "is it worth the time?" The loop asks "does it compound?" Time saved is linear. Knowledge that compounds into better future sessions is different. The math is harder but the payoff is bigger. The boring stuff thesis as arithmetic.