The Agentic Web: Present and Future.

Something I keep seeing in the AI-readiness conversation that is worth clearing up, because it is causing founders to optimise for the wrong thing.

llms.txt and WebMCP are not the same thing. They do not compete. They sit at completely different layers, and understanding the difference changes what you should actually be building or fixing right now.

llms.txt tells AI agents what your site is about. It is a content knowledge layer. Ship it. It takes ten minutes, and Lighthouse audits for it now. But it does not let anyone do anything on your site. It is a map, not a mechanism.

WebMCP is the mechanism.

It lets your site register callable tools in the browser so an AI agent can take actions directly without navigating your UI. Google and Microsoft are both behind it. John Mueller has called it the more meaningful direction. And it is currently in a Chrome 149 origin trial with production stability still months away for most sites.

The gap in the middle is where I have been building.

Right now, today, a visitor using an AI assistant to try to use a website can read its content but cannot complete any action without clicking through the UI themselves. llms.txt does not solve that. WebMCP will solve it eventually.

Neither solves it today.

Hunch is a single script tag that sits in that gap.

Visitor types what they want, the widget handles the action end-to-end in the browser, right now, on any stack. We built it because we were tired of watching AI agents stall out at the action layer while everyone argued about which static file format to adopt.

We also have a free public audit that checks your site across all three layers. If you want to know exactly where your site stands on content discovery, agent protocols, and actual action surfaces, it is at hunchbank.com/agent-readiness-audit. Takes about a minute.

No pitch beyond that.

The three-layer framework is genuinely useful regardless of whether Hunch is the right fit for you.