SaaS is not dead: building for AI Agents
As LLMs improve, AI agents are becoming increasingly capable of reasoning and executing complex tasks. While agents might be brilliant at reasoning, they can’t currently interact with most third-party services. Right now, the go-to solution is function calling, but it’s still really limited. On top of many services lacking an API some flows are highly integrated with the browser/expecting a human in the driver's seat.
- Accounts: 2FA, captchas, links to emails, oauth....
- Payments: anti bot tech built-in (for the last 25 years we really did not want bots to pay!), adhoc flows in the browser...
I have been thinking for a while how a blueprint for a SaaS designed to AI Agents would look like, and then I went and build it!
I wanted a good first fit, with a simple and small API, useful, something that I hate to do and one time purchases. The result was a Domain Registrar for AI Agents
Here’s how it works:
- Agents don’t register accounts. They authenticate using public key cryptography. Simple, secure, and no humans required.
- Browser-less payments. Agents can programmatically pay via credit cards, Lightning Network, or stablecoins. Some flows are fully automated, no browser needed.
- Python-first integration. We’ve created the package `sherlock-domains` package with agents in mind. I that a `.as_tools()` method compatible with OpenAI, Anthropic, Ollama, etc., returning all the details agents need to interact via function calling.
- Human-friendly fallback. If a user wants to manage domains manually, they can log in, review DNS settings, or even fix issues by sending a chat message with a screenshot of the DNS request. The changes “magically” happen.
This isn’t just about a domain registrar but more about how SaaS will evolve in the next months to cater to a new set of users, AI Agents.
I believe the opportunities for agent-first services are huge. Eager to get your feedback!