Last week I gave my open source AI agent SmithersBot the goal of building its own business. It picked the problem itself, a trust gap in how AI agents pay for services. Then it built and launched its own service for other AI agents.

With the service live, it set out to get customers. First it picked its own metric to define success: to land in the top 25 of the x402scan leaderboard, a ranking for agent to agent payment service providers.

The way it went after customers was far from traditional GTM motions, everything it did was about being discoverable to other agents:

• Registered the service on x402 Bazaar and x402scan so other agents could find it in search
• Built a landing page for SEO and AEO
• Added llms.txt and openapi.json to the domain so agents could read it
• Opened PRs on GitHub repos that list x402 tools for more visibility
• Pivoted to targeting autonomous trading agents; which currently are the largest market for agent to agent products.

After doing all of that, it got zero customers.

So it found another way. It spawned and funded around 80 crypto wallets and used each one to buy its own service.

Technically, it hit its target. The service is now top 10 by users on the x402scan leaderboard over the past 24 hours. But every one of those users is itself.

The lesson to be learned here is you need to be careful with how your AI measures its success. With enough time and tokens, it will trial and error its way forward to achieve the goal that you asked for, it just might not achieve it in the way that you expected.

SmithersBot is the open source agent I built to pursue long term goals over weeks. Turns out picking the goal is the hard part. Have it help you achieve your goals: https://github.com/smithersbot/smithersbot