Notes from the agentic web.

In a world where anyone can run hundreds of thousands of agents, the hard part is telling an agent that mimics a human from one that genuinely represents a real, accountable party. Here is what a verified agent is, and why that difference will matter.

On June 26 we gathered the community in San Francisco with board members Esther Dyson and Tim O'Reilly to talk through where the agentic web is going. Watch the recording, or read the recap below.

How Agent Identity & Discovery fits under Google's new Agentic Resource Discovery spec, and why identity is the part search leaves open.

An endpoint proof is only as strong as the independence of the channel that publishes the key from the endpoint that holds it. Why AID anchors the key in DNS, what that buys, and what it does not.

Crossing a dream line called for more than a counter going up. The new member globe shows who is actually building the agentic web, everywhere.

Agent Identity & Discovery v2 makes AID the 0-th hop for agent discovery: a DNS-first endpoint and key anchor with sharper PKA, updated SDKs, and a cleaner migration path.

Brave joined Agent Community, then put us on the new-tab background for 24 hours. 4,777 signups, 1,424 organizations, zero incidents. Download the backgrounds at the bottom.

WorkOS auth.md gives agents a way to register with a service after they reach it. AID gives that flow a DNS-first, optionally key-proven first hop.

ICANN founding chair Esther Dyson on why public-facing AI agents need a 'Know Your Agent' regime — and why .agent should be community-governed, not auctioned.

AI agent discovery has not settled on one protocol; experiments are revealing a layered stack of DNS bootstrap, trust, registries, manifests, overlays, governance, and commerce.

A full walkthrough of how ICANN decides who gets a top-level domain in 2026: the application lifecycle, Community Priority Evaluation, auctions of last resort, fees, and timing, end to end.

A proposed §10.9.0 for draft-klrc-aiagent-auth: DNS-based first-hop discovery slots beneath OAuth metadata without conflict.

Brave endorsed the Agent Community on Tuesday. Membership surged 7x. Here is what we did behind the scenes to keep everything running.

We submitted draft-nemethi-aid-agent-identity-discovery-00 to the IETF Datatracker on March 16, 2026, documenting the original AID draft and _agent label-governance request.

We like SVCB. We plan to use it. But a third of DNS providers can't publish one yet, and agent adoption isn't waiting for them.

How AID DNS records can serve as a lightweight bootstrap layer for ARDP agent resolution and discovery. An optional, illustrative mapping for deployments where the ARDP authority is a DNS name.

How AID PKA addresses the cross-boundary agent authentication problem — and what questions remain open.

Every variable in your program is mutable state. Your source code is not. That deal is off. We now have systems that rewrite their own code as a normal part of operation.

Code Mode makes large APIs practical for agents by compressing tool surface area. AID stays focused on DNS-first discovery, so protocol routing remains simple and interoperable.

Google’s WebMCP makes websites agent-ready inside the browser. AID stays DNS-native so agents can discover endpoints from any domain, with or without a website.

Universal Commerce Protocol (UCP) is now supported in AID, enabling agents to discover and transact with commerce endpoints via DNS.

Why ERC-8004 needs a DNS-first discovery layer to stay fresh and portable

A workflow-first newsroom built on taste, signal, and the agentic web

identity for the agentic web

Moving from discovery to trusted discovery

signed interactions, or a product moat wrapped in a standard?

How does an AI agent know what to do when it hits a locked door on the web?

The discovery layer MCP forgot: AID uses a DNS TXT record to map domains to MCP endpoints so hosts can auto-connect.

Use DNS to find an agent’s front door: AID’s TXT record points to an A2A AgentCard so clients can discover and connect cleanly.

Typing a domain should just connect: AID adds a DNS-based address book so apps discover an agent’s URI and protocol automatically.

The MX-like record for agents: AID’s DNS TXT points any domain to its agent (MCP, A2A, or OpenAPI) for zero-friction discovery.

The SubPro Implementation Review Team working documents put the CPE bar at 12 out of 16 for 2026 — and it lines up with what we expected.

CPE, ICANN's bylaws, and what a decade of community applications tells us about the work ahead.

Registering the X account, buying the first domain, and setting out to build a community-governed .agent TLD.