MCP Apps co-creator: half the web ships llms.txt, almost no agent reads it
AI Engineer · youtube · 2026-09-16
An interview with Liad Yosef, co-creator of the MCP Apps spec, on rebuilding the web for agents.
llms.txt ignored: His team found roughly half of tested websites publish llms.txt, the de facto standard telling agents how to work with a site — yet almost no agent reads it. Agents go straight to docs and the homepage; the minority that opened llms.txt only did so because docs mentioned it. That reframes the problem: instead of writing specs at agents, watch thousands of them actually try.
What MCP Apps unlocks: The spec lets a server hand a real interface back into the chat instead of being flattened into text. Two decades polished the browser tab, yet planning one anniversary still means learning four interfaces; break interfaces into atoms and let an assistant compose them, and nobody needs the dashboard.
Agents have no brand loyalty: His team asked their coding agent to pick an analytics service; it recommended one they'd never used over a decade-long vendor purely for the better server and API, and they switched. Ten years of polished UX lost to a tooling preference — and a product that forces an agent to open a browser gets remembered for it.
He also argues browser agents are a "faster horse," assistants become the entry point, and accessibility and agent readiness are the same work.
More from AGI Musings
- 25 Fields medalists warn of 'severe misalignment' after OpenAI claims century-old math problem solved — nordicinst · 2026-09-17
- New Philosophy of Science paper simulates AI-driven epistemic monocultures in research — MohammadAtari90 · 2026-09-17
- AI safety researcher warns against regulation that looks like regulatory capture — DavidSKrueger · 2026-09-17
- Cambridge's David Krueger flags contradiction in US AI race policy — DavidSKrueger · 2026-09-17
- Reviewing AI code through Steve Jobs' lens: unseen internals deserve beauty too — sergeykarayev · 2026-09-17
- An economist's take: AI agent incidents are decades-old emergence, just with real-world stakes — Genzinvestor16180339 · 2026-09-17