Introduction

This is the first post I've written in a long time - generally, I operate in my head, and while I do build things in Public, I rarely publicize what I'm doing.

This is a (meager) attempt to fix that...

As my Day-Job puts me into the Identity world, I've been doing a lot of work and thinking about AI Agents, and how they work in a world of Human-centric digital services.

I spend a lot of time working with Customers and their "humans" (internal and external), and I'm increasingly convinced we've gotten it all wrong. Identity Providers want to own the entirety of the Identities and their associated data, and while Federation is baked into the underlying protocols - other than Social Identity (Google \ Apple \ Facebook), it's rare that one IDP trusts another.

As Humans, we don't have a single Identity - we have hundreds. I'm not convinced we should be imposing these patterns onto purely digital entitles.

I've spend energy over the past few years working with Distributed IDs (Verifiable Credentials), and they feel like that's a better way to handle Agents. They can self-identify as an Agent, and a Service Provider can choose whether to trust them - this doesn't need a "1st Class Identity" at each IDP - just a consistent Identifier that the IDP can make a decision against.

Where this intersects with the ATmosphere, is that AT is also designed, at the protocol layer, to allow for distributed IDs, and data, and control.

In that vein, I've done a couple of things to explore AT:

1. I started with BSky using a Domain-validated Handle
2. I moved that to a self-hosted PDS (this post should go there, and readable via pckt's AppViewer)
3. I built a set of Agents, running on a Raspberry Pi, that follow a pattern I first read about in "A Half-Build Garden" - by Ruthann Emrys. They grab some data from Environmental APIs, run through a single-domain AI analysis, and post that to a custom lexicon record on a PDS running on the RPi. An external Agent calls into that PDS for the individual observation records, and runs a cross-domain AI analysis and posts a summary to BSky (with a full record underneath it).

I find AT quite compelling -- separating the Identity and Data components, using flexible schema for the data storage, with the Indexing and Firehoses allowing custom AppViews of it all.

This feels like the future...