How we developed an Agent Score to improve Fern Docs

AI agents are increasingly reading API documentation, but many docs are not optimized for these non-human readers, often due to structural issues rather than content quality. To address this, a free, open-source benchmark called Agent Score was developed, evaluating documentation on agent-readiness with a score from 0 to 100. Partnering with expert Dachary Carey, the AFDocs standard was created, detailing 22 checks for ensuring documentation is accessible to AI agents, covering aspects like page rendering, discoverability, and structural integrity. The AFDocs spec aims to prevent issues like bloated HTML and JavaScript-rendered pages that hinder AI parsing. The Agent Score tool measures compliance with this standard and provides actionable insights for improvement. By enhancing docs to be agent-friendly, companies can better reach the growing number of developers using AI agents to access information, thereby increasing integration opportunities.