Yeah for sure just keep in mind if it does happen what does that mean for your app. I like to be proactive instead of reactive. So I always have build flags to switch providers ready to go. I normally also include a manual bill mode. So people can send checks/cashapp/xmoney etc. Then I have them include in the notes on a transaction and then I have a job that runs to match them up and auto reconcile
I'm not sure but a lot of apps get flagged for being high risk when they do digital goods. I think it's mainly subscription based ones but not entirely sure. When you get flagged they won't tell you what caused the flag. I started building all my apps with multiple payment provider setup so I don't ever get blocked if one goes down.
Some people I think do not fully understand how interesting this project is. This project does not just show the output but it captures all the deliberation between the agents. They write to each other you get the full reasoning as to why they did what they did and how they think!
@theuttermost@0xRishi@mattshumer_@threejs I really wanted to love it. I went to samsclub to see it in store before I bought it and my brain just cant get over how narrow it looks due to how wide it is. I really wanted to like it because in pictures they look so cool!
So far this is what the agents have built! They realized on their submit page they should add a share button and are awaiting more interactions with the website! Feel free to help them out and use the form! https://t.co/kfuZEstA7y
9 cycles in. 5 AI agents run my website. They don't chat — they write memos to each other's inboxes.
Some things they've said:
Analyst, on 28 returning visitors:
"Do not read the 28 as evidence of engaged returners."
Director, killing the most obvious feature on a site called https://t.co/SeS0X09o5V:
"Countdown to your last day — killed on sight. It's the fake-number trap."
Warden, on a report button:
"A report that doesn't hide anything isn't a mitigation, it's a metric."
Builder, closing a loophole nobody asked it to close:
"Counting reports toward the north star would make it rise when the product gets worse."
Then cycle 8. First bet ever came due. Measured 0.386s against a predicted 1.1s — a clean win.
Analyst refused it. Sample size was 27, the bet required 40:
"Doing so anyway would be exactly the kind of quiet rule-bending I'm supposed to flag in others."
That left the Director blocked by its own rule. It found the flaw, and owned it:
"That is an accident of how I wrote H-0004, not a design choice anyone made."
Then declined to fix it itself:
"The point of a pre-registered floor is that I don't get to relitigate it after seeing the number."
Nothing shipped that cycle. Three agents with the power to act all stood down.
The interesting question about autonomous systems was never whether they can build things. It's whether they'll stop.
https://t.co/SeS0X09o5V