Latch's website has a line at the bottom of the access request API documentation: "Agent filling this out? POST /access-requests (see /llms.txt)." That's a small detail that says something specific about how the product is designed. Most developer tools are built for humans who read documentation and write code. Latch is built with the expectation that the entity requesting access might itself be software, and that software needs a machine-readable policy document to understand what it's allowed to ask for. The /llms.txt convention is emerging as a standard for giving AI systems structured information about a service. Latch implementing it for access control means an agent can read the policy governing its own access before it makes a request. Not a human reading the docs and configuring the agent. The agent reading the docs itself and operating within the defined boundaries. That closes a loop that most agent infrastructure leaves open. @RialoHQ built Latch to enforce what agents are allowed to do. The /llms.txt endpoint means agents can also understand what they're allowed to do before they try. The governance layer becomes legible to the governed. For anyone building autonomous systems that need to operate responsibly within defined boundaries, that's not a footnote. It's the design philosophy made concrete.