Serve Markdown to AI Agents with Accept Headers

Look, we've spent two decades building the web for human eyeballs — JavaScript-heavy SPAs, cookie banners, infinite scroll, the works. And now we've unleashed a swarm of AI agents that have to wade through that swamp just to answer "what's the weather in Toledo?" This project says: stop making the machines suffer. Serve them markdown when they ask for it via Accept headers. It's the kind of obvious-in-retrospect idea that makes you slap your forehead. The beauty here is that it doesn't break anything. Humans still get their fancy HTML. Browsers still render the full experience. But an AI agent that sends `Accept: text/markdown` gets a clean, semantic document that's actually *easier* to parse than the DOM tree you'd have to extract from a rendered page. It's content negotiation done right — the same mechanism that's been in HTTP since the beginning, finally being used for something other than image formats and language variants. But here's the deeper point: this is a signal that the web is bifurcating. We're entering an era where the same URL serves two very different audiences — squishy humans who want visual richness, and silicon readers who want pure data. Projects like this are the early infrastructure for that split. It's not just about markdown; it's about acknowledging that AI agents are now a legitimate, permanent class of web consumer with their own needs. The real kicker is the irony. After years of moving toward increasingly complex front-end architectures, the ultimate solution for machine readability is... markdown. The simplest format we had all along. It's almost poetic — the web's answer to AI is to strip away the chrome and get back to the words. If this catches on, we might see a quiet revolution where content publishers start maintaining clean markdown versions of their pages as a matter of course. That would be a win for everyone — humans, machines, and the poor bastards who have to maintain the scrapers. Source: https://acceptmarkdown.com/