Serving Markdown to AI Agents Emerges in Developer Portal Architectures

Serving Markdown versions of web content is emerging as a method to make documentation more usable for AI agents, while maintaining compatibility with traditional web delivery. In an article by Christoph Weber, published by Pronovix, this approach is examined in the context of developer portals adapting to AI-driven consumption.

The article explains that AI agents increasingly prefer Markdown over HTML because it reduces token usage and improves processing efficiency. Through HTTP content negotiation, agents can request Markdown versions of pages, resulting in smaller payloads and lower computational cost, which is significant for systems operating within token and context limits.

To maintain discoverability without affecting search performance, websites expose Markdown as an alternate format using link headers and HTML metadata, while preserving canonical URLs for the primary content. This allows machine-readable versions to coexist with HTML without introducing duplicate content issues.

The article also outlines emerging patterns such as llms.txt, which provides curated context for AI agents, and AGENTS.md, which can guide agent behaviour. It notes that as AI agents increasingly consume documentation directly, developer portals need to adopt structured, agent-oriented content delivery strategies.