Why the clock beats the model

Here is the single idea this course exists to teach: the cost of a routine isn't the price of one run. It's the price of one run times every time it wakes up for nothing.

Think about what most routines actually do. They check things. Did a competitor change their pricing page? Did the build break? Did anyone reply to that thread? Is a ticket stuck? For the overwhelming majority of runs, the answer is "no, nothing changed." The routine wakes up, looks, finds nothing, and goes back to sleep — and you paid for the looking.

So doubling how often a routine runs roughly doubles its cost, while the extra value is usually tiny. You rarely learn twice as much by checking twice as often; you just pay twice as often to hear "still nothing."

Let's put real numbers on it, using xAI's published rates as a yardstick. The current flagship model, grok-4.7, costs $2.00 per million input tokens and $6.00 per million output tokens (x.ai/api). Take a routine that reads about 20,000 tokens and writes about 1,000 each run:

input:  20,000 tokens x $2 / 1,000,000  = $0.040
output:  1,000 tokens x $6 / 1,000,000  = $0.006
                                   total = $0.046 per run

Four and a half cents. Harmless — until you multiply by how often it fires. A routine on the five-minute floor runs about 8,640 times in a 30-day month. That's roughly $397 of compute for one job. The same job hourly is about $33. Same Bot, same model, same prompt. The only thing that changed is the alarm clock.

Now the honest caveat, and it's important. Grok Bot runs on your plan's usage allowance, not on per-token API billing. You will not see a $397 invoice. What you'll see is your allowance draining about twelve times faster, mostly on runs where nothing happened. The dollars here are a way to compare how much compute each schedule burns, not a bill you'll receive. The real meter is the usage view in the app, and the token counts above are an example — a routine that opens a long thread or a heavy page reads far more.

But notice the caveat doesn't rescue you. Whether the cost shows up as dollars or as a burned allowance, a five-minute routine is still twelve times hungrier than an hourly one, for the same work. That's why the clock beats the model. Swapping models is a config change that nudges the per-run price. Choosing the schedule is the decision that multiplies it.