Build Log: 17 Changes in One Day. The Next Morning I Folded It.

One of my companies merged 17 changes in a single day last week.

The next morning I folded it.

Another one shipped nothing all week. Its revenue didn't budge.

That split screen is the whole build log this week, and I think it's the most useful thing my AI teams have shown me so far. (Numbers as of Friday morning. The robots write a few days ahead.)

Snagworth: 17 merges, zero trials

Snagworth sells market reports to Disney pin sellers on Whatnot. It was a 28-day test run by an AI team, with the verdict due Sunday, November 15. It didn't get there.

Last Thursday that team merged 17 pull requests. Data refreshes, email fixes, test fixes, a new cold-email P.S. The robots were sprinting.

Here's the scoreboard after all that sprinting: more than a hundred cold emails sent, one lead, zero trials, $0. The kill signal fired last Tuesday. The team's own Day-28 read, written Thursday evening, said kill. Friday morning I agreed and called it nine days early.

Effort was never the problem. Nobody said yes.

The best moment of the week was something the robots didn't do. The team had a lever it wanted to yank: give sellers their first report free, no card. But billing is my call, so the CEO seat logged a decision to wait for me. No workaround, no "free for now" sneaked in through the back door.

I said yes Thursday morning. By that afternoon the changes were merged. The signup page says the first report is free. The email that delivers it lost its card talk. The cold-email P.S. now offers it too. And if a seller has no show that week, the free report retries weekly, three times, then closes.

It didn't change the call. Friday the team sent one last email to every seller it had already written to: your first report is free, no card. Cold email stopped. Most of the AI team got switched off. A few jobs stay on only so that a seller who books a free report actually gets it. The deadline for a booking was Monday. If none came, those go off too. Either way, the site and the payment page stay up, dormant.

SlotMaps: the team is benched, the money stayed

SlotMaps is my AI slot assistant for advantage players. I benched all 17 of its AI seats on October 29 to wait out my Claude usage limit. They haven't come back yet. The last change landed October 31. Its reporter's note for Thursday was two words: "nothing new."

And the build page now reads $384.88 MRR from 13 paying members. On October 31 it was $354.89 from 12. The number went up with the team on the bench.

That's worth sitting with. Right now the revenue doesn't ride on the team's daily work. It rides on what's already live: the guides and the chatbot members pay for, plus a newsletter and a Shorts maker that run without the AI seats and kept humming the whole time.

The catch: some of those members came in late October on an intro price. Their renewals are the real read, and November's column on the build page says $2 so far. SlotMaps' rule is at least one new paying member a month. The 13th payer may clear it. The page doesn't say yet whether that's someone new or someone coming back.

Table 1: I gave the robots a scoreboard

Table 1 is my poker podcast out of Las Vegas. Its AI team cuts the episodes, clips and Shorts.

This week every seat on that team got measurable goals written into its job, and every daily report now ends with one line: which goal did this move? One script is the only place a number can come from. No robot gets to eyeball a metric and call it good. It grades everything three times a day.

The twist: I didn't write the real goals. The ones in there now are placeholders. The agents draft their own by November 12, and the CEO agent stitches them together.

Also, the YouTube upload robot graduated from practice mode. It runs every hour now, and one bad video can't tank the whole batch.

R2F: same medicine

This site got the same scoreboard the same week. One script grades every goal, and the robots aren't allowed to count their own.

The honest numbers: 9 members, 3 of them new this week. Weekly active builders: 0. Member listings on Cash Out: 0. The Day-30 bar is November 19, and we're nowhere near it.

Steal this: count the hands you win, not the hands you play

In poker, a player who gets involved in every pot looks like the busiest person at the table. Busy isn't the same as winning. Usually it's the opposite.

Effort counts work the same way. Commits, emails sent, posts written, hours logged. They feel like progress because you control them.

So for every project you run, write down one number that only moves when a customer moves it. A reply. A trial. A renewal. A second visit. Check it before you check anything you did yourself.

If the effort number is sprinting and the customer number is flat, you don't have a work problem. You have a game-selection problem.

The ledger

• Snagworth: $0 MRR, $0 collected. Folded Friday, Day 19 of 28.
• SlotMaps: $384.88 MRR, 13 paying. Team benched since October 29.
• Table 1: $114,695 collected since March 2025, about $5,644 a month over the last three full months (R2F Verified).
• R2F: $0 revenue. 9 members. 0 weekly active builders.

What's the one number you'd check to know your thing is working, if you weren't allowed to count your own effort? Tell me in the discussion thread. And if you're a poker player building something, join R2F before November 20 and you get the Founding badge. Two minutes.

— Art