Agent experience is developer experience for machines. As AI agents become first-class consumers of APIs, the experience they have discovering, understanding, authenticating, and calling an API matters as much as the experience a human developer has. An API that an agent can reason about and use reliably will be reachable by a whole new class of automated consumers. I treat agent experience as the next frontier of the work I have always done around developer experience. It means machine-readable descriptions, llms.txt and agents.md files, MCP servers, and operations documented well enough that a model can call them without guessing. The providers who invest in agent experience now will be the ones whose APIs get used as AI reshapes how software is built.
Agent Experience
Policies
Agent Readiness with agents.md
I require that every API provider ships an agents.md file that tells an automated consumer exactly how to behave when working with the API, including the conventions, the guardrails, and the workfl...
Agent Readiness with llms.txt
I require that every API publishes an llms.txt file at a predictable, well-known location so that large language models and the agents built on them can find a concise, curated map of what the API ...
Agent-Scoped Authentication
I require that APIs support authentication credentials scoped specifically to agents, so that an autonomous consumer acts under its own narrowly-bound identity with least-privilege scopes rather th...
MCP Server Provided
I require that an API of consequence exposes a Model Context Protocol server so that agents can discover its capabilities and invoke them through a standard, well-understood interface rather than a...
Operations Are Agent-Legible
I require that every operation in an OpenAPI carries a description rich enough that an agent can decide, from the definition alone, what the operation does, when to call it, and what it will get ba...
Tool Definitions Provided
I require that APIs publish explicit tool definitions that describe each callable capability as a discrete tool an agent can select, with clear names, descriptions, and typed inputs and outputs. Th...
Strategies
APIs Are Legible to Automated Consumers
I want our APIs to be legible to the automated consumers reading them, not just to the humans who wrote them. That means operations that describe themselves clearly, tool definitions that spell out...
APIs Are Ready for AI Agents
I want every API we operate to be ready for the agents that are now showing up as first-class consumers alongside our human developers. That means our APIs publish the machine-readable signposts ag...
APIs Expose an MCP Server
I want our APIs to meet agents where they already are by exposing a Model Context Protocol server that turns our operations into tools an agent can pick up and use without a custom integration for ...
APIs Provide Orchestration Workflows
I want our APIs to ship the workflows, not just the endpoints, because most real jobs take more than a single call and I would rather hand consumers a proven path than make them assemble one from s...
