Eric Siu gave an AI agent called Bug Basher a Ramp card and told it to make money. Over four weeks it made over 10,000 cold calls selling NYC pest inspections, added Spanish and Chinese, and generated a $75 sale through Stripe. The experiment shows AI agents can chase short-term revenue but still struggle with planning ahead.
Ramp, and so I think they're all amplifying each other. And there's this product called Triordinal, which allows you to do more employee-generated content like EGC. And the cool thing is at the end of their post too, they all have recruiting link at the end of it so they can just continue to get more people, right? More awareness to get more people. The cool thing here is that from Ramp gave an AI agent a Ramp card and told it to make money. As far as they're concerned, this is the first example of an agent where you tell it to make money and it actually made money for you. But here's what they did. They gave Devin a ramp card. So I'm assuming this is Devin from Cognition and a phone and email. They named it Bug Basher, and then they told it to sell NYC pest inspection and referrals. Okay. 4 weeks and many phone calls later, I think over 10,000 phone calls, and it made $75. The first publicly documented AI B2B sale from a cold start. Heartbeat checked the inbox, Stripe, and calls, and then revised prompts and its shared brain from transcripts. It added a Spanish or Chinese ask, leaked its system prompt on 197 calls, and found our systems in Git when Slack installed. Paid $75 via Stripe. Agents hill climb well, but still struggle with long-term planning and go make money. So that's the lesson from this. But it seems like it's one of the first documented cases of an agent just making money.
