Giving the Internet a Way to Talk Back

The Answering Machine Moment

Remember answering machines?
One machine calls another machine.
They leave messages. Work things out.
Humans never had to talk.

"Calling Elvis, is anybody home?"

That old Dire Straits song got stuck in my head.

What if AI agents could do that?
Your assistant calls the pizza shop's agent:
"Table for two at 7pm?"
"Yes, we have availability."
"Great, booked."

You wake up to a calendar notification.
The machines figured it out while you slept.

Simple. Polite. Automatic.

That's what I started building.

What This Actually Is

I'm experimenting with an agent discovery protocol for the internet.

Little JSON files at .well-known/agent.json (think: business cards for robots) that tell agents: "I'm a bakery. I'm closed Mondays. I like it when you say please."

And wrapper agents that can translate between your fancy AI assistant and Bob's pizza shop website that was built in 2015.

Agent-to-agent communication with trust scores, so agents know who to trust and how to behave.

There's so much agent development happening right now (Google's A2A, Anthropic's MCP, and others). But I haven't found anyone focusing on this specific piece— how local businesses become discoverable to agents in a trustworthy way.

Maybe I've just missed it? If you know of similar projects, please clue me in. It would be silly trying to reinvent the wheel.

December 2025

The Agentic AI Foundation

Yesterday, the Linux Foundation launched the Agentic AI Foundation (AAIF)—with Anthropic, OpenAI, Google, Microsoft, and others joining forces to create open standards for agent communication.

They're contributing MCP (how agents talk to tools), AGENTS.md (how agents understand code), and goose (a framework for building agents). This validates exactly what I've been building toward.

But there's a missing piece in their stack: how do agents discover real-world businesses?

MCP tells agents how to use a tool once connected. But how does an agent find which pizza shops exist, which ones are trustworthy, which ones are open right now? That's the infrastructure gap I'm filling.

How I Got Here (The Honest Version)

I started by trying to build a better AI assistant— one that could handle real conversation, remember context, understand when I give it three tasks at once.

I even filed a patent application for it. Mostly so I could think through the problem clearly.

But then I realized: the assistant was only half the puzzle.

The other half was giving the rest of the world a way to talk back to my assistant.

That's when the answering machine thing clicked.

So I pivoted. Started over. Filed five more patent applications. (They're kind of like thinking tools for me— ways to really understand what I'm building.)

Built a working prototype. Broke it. Fixed it. Broke it again. Learned Python. Learned async/await. Learned a lot about trust scores.

Convinced a few developers to try it. They're onboarding real businesses now in Germany, Canada, Brazil. Teaching me things I didn't know.

When It Works, It Feels Like Magic

You know that moment when you have an idea— just a fuzzy "what if" floating in your head— and then suddenly it's running on a server somewhere, actually doing the thing?

"That moment never gets old."

Watching an agent read a JSON file and understand "this bakery is closed Mondays" feels like teaching a very polite robot to read a sign.

Watching two agents negotiate a delivery time while I eat breakfast and they have no idea I'm watching...

Watching a developer in Berlin onboard their first business using just the docs, no help from me...

Simple, maybe. But: magic.