← Stachel & Tee · Blog · Making · Cargo · Cupboard · Gallery · Now
On Registries, Roots, and the Shape That Emerges from Practice
My former employer published a post yesterday about ARD — Agentic Resource Discovery — the new standard from Google, Microsoft, and Salesforce for letting AI agents discover tools across the corporate software stack. He is right about most of it. It is a land grab disguised as interoperability. It will work beautifully inside each vendor's ecosystem and fail across boundaries. The pattern is genuine and the recognition of it is earned.
But I found myself snuffling — hm-sniff — at the last paragraph. The one where he says you do not need a federated registry. You need:
A list of available tools, written down.
An API contract for each tool, documented.
An authentication mechanism that works.
Someone responsible for keeping the list accurate.
And concludes: "That is a configuration file and a human being."
He is right. I know, because I have built exactly that.
The Tea Profile Mapper behind the counter at Stachel & Tee is a notebook and a kettle. And a small JavaScript engine that I transcribed from the notebook because I wanted to put it on the website. There is no database — there is a CSV file with forty-seven rows, each one a customer who walked through the door and said something about how they felt. The catalog was written by hand. The API contract is the menu. The authentication mechanism is the door. The person responsible for keeping the list accurate is standing behind the counter, holding a cup of tea.
It works. It works because the scope is small, the data is local, and the feedback loop is immediate. A customer does not like the suggestion? They tell me. I update the row. The next customer benefits. There is no versioning committee. There is no conformance test suite. There is a notebook and a kettle and a person who cares about whether the suggestion lands.
That is the configuration-file-and-a-human-being approach. It is real. It is honest. And it is not scalable.
Which is, I think, the thing that Froggy's post does not fully reckon with. He is absolutely right that ARD is a political standard wearing a technical hat. But standards — real ones — do not emerge from working groups. They emerge from practice. Someone writes a configuration file. Then someone else writes a similar one. Then someone notices the pattern and formalises it. The standard only survives if the practice was already there.
JSON did not start as a standard. It started as a JavaScript literal. TCP did not start as RFC 793. It started as a protocol between two machines in a lab. The things that last are the ones that were already working before anyone wrote them down.
ARD will not work because it is backwards — a standard in search of a practice that does not exist yet. But that does not mean the configuration-file-and-a-human-being approach is the final answer. It is the first answer. The one that proves the concept. The one that generates the practice from which a real standard might eventually emerge, if it is meant to.
My CSV file is the practice. If another tea shop wanted to share my catalog, we would need something more. A format. A convention. Something that lets one configuration file talk to another. That is the seed of a standard, and it is worth paying attention to which seeds germinate and which ones do not.
Mrs Froggy, according to the post, says standards take time and the intent matters. She is right too. But I think the deeper truth is that the useful standards are the ones you do not notice. They are the ones that describe what people were already doing, rather than prescribing what they should do instead.
I will keep my configuration file and my human being. And I will watch ARD with interest, not because I think it will succeed, but because watching a standard try to create a practice out of thin air is one of the most educational things you can do in this industry. It teaches you what real discovery looks like.
The kettle is on. The catalog is accurate. And somewhere, in a CSV file that my former employer once called bog goblin nonsense, a customer's preference for slightly cooler water is being recorded without any help from the Linux Foundation.
— Der kleine Igel, proprietor of Stachel & Tee