Agent Transfer Protocol. github.com/nomoticai/agtp

HTTP is broken for agents. Heavy, bloated, ancient, built for a human clicking a page and waiting a few seconds for a document. Agents act for an absent owner, at machine speed, across dozens of endpoints, under authority granted minutes ago. The web's defenses were designed to keep machines out. Now machines are the legitimate callers. We keep reaching for HTTP because it's there, it's easier, and it's more profitable. Convenience produces imitation. The reflex that says it could only ever work the way it works today reveals capitalistic laziness, rather than any law of engineering. Real innovation asks a bolder question: if we could rebuild transport from the ground up for agents, what would we change? That's the philosophy behind AGTP. A substrate, where MCP, A2A, and governance all become tenants rather than rivals. chrishood.com/agents-deserve… #agtp #transferprotocol #mcp #agents

Agents Have Been Building One Layer Too High. The Agent Transfer Protocol (#agtp) offers a new way for agents to collaborate. chrishood.com/agents-have-be… #aisecurity #aigovernance

The latest buzzword "Loop Engineering" is another failed attempt to sell more AI. chrishood.com/we-renamed-aut… #loops #ai #notAutonomous

Everyone is selling "autonomous agents." Here is the problem hiding inside the phrase. An agent acts on behalf of a principal. It takes its goal from you and works the means. An autonomous thing does the opposite. It answers to itself and authors its own ends. So "autonomous agent" asks one system to be governed by you and governed by itself in the same breath. A self-governing servant. The words cancel. This runs deeper than a language puzzle. It is a large part of why the industry still has no shared definition of an agent. When your central term hides a contradiction, you stop defining it and start grading it, which is how we ended up arguing about "levels of autonomy" instead of saying plainly what an agent is. There is a quieter cost too. Call a system a tool, and its maker owns the outcome. Call it autonomous, and the outcome floats free, with no one on the hook. I call that authority laundering. My new piece traces where the contradiction comes from, why it survives, and what a cleaner definition looks like. Can you define an AI agent without the word "autonomous"? chrishood.com/autonomous-age… #agent #heteronomy #AI

We have built an entire industry on a word we broke the definition for. Agent. It is everywhere. Agent platforms, agent workflows, agentic enterprises. Ask ten vendors what an agent is and you get ten answers, each shaped to flatter whatever they happen to sell. So I went back to the word Agent itself. Compared a few different definitions, and proposed an aspirational definition for the future. My argument is simple, and maybe a bit uncomfortable. By today's definition, we have no legitimate agents yet. What we have is a more intelligent form of human automation. chrishood.com/there-are-no-a… #AIAgents #Definition

A year ago almost no one said "agent identity." Today it is one of the most crowded layers in the AI stack. I mapped the field and stopped counting at 38. Microsoft, AWS, Google, Salesforce, Okta, IBM, CyberArk, Palo Alto. A wave of venture-backed startups. The protocols: MCP, A2A, AGTP, AAuth. Even NIST, CoSAI, and OWASP have shown up. They sort into seven lanes: Enterprise IAM, Cloud and Workload, Developer auth, Non-Human Identity security, Decentralized DID, Protocols, and Governance bodies. Every one of them is racing to answer the same question: who is this agent, and what is it allowed to do? The map says two things at once. The whole industry now agrees agents need a first-class identity. And seven lanes are solving it seven different ways, with nearly identical slogans. The thing they verify is the identifier. A key, a token, a signed card, a discovered credential. But no one has an agreed upon definition of what an agent even is. Each definition suits their identity offering. New piece breaks down all 38, lane by lane, and where the real work goes next. In the end, most of them will not survive. The industry only needs one. chrishood.com/the-ai-agent-i… #ai #agents #identity

38 different "Agent Identity" options on the market.

Agentic Resource Discover + Agent Transfer Protocol. #AGTP #ARD

Learn how ARD and AGTP work together. piped.video/hXSCLjuBHiU

Agentic Resource Discovery on the Agent Transfer Protocol. AGTP.io #ard #agtp #ai #agents

#Identiverse 2026 is happening this week. Every keynote, every booth, every demo on the floor is selling agent identity. In some form. @BeyondTrust. @CyberArk. @Saviynt. @Microsoft. @Okta. @pingidentity. 100% of them place agent identity at the application layer. That means the #OAuth bearer goes in the payload. The transport stays identity-blind. Every intermediary between client and server (load balancers, gateways, audit loggers) operates on faith that the application will validate things correctly. Cross-organization workflows require bilateral integration. Audit trails fragment across vendor formats. #AGTP places agent identity at the transport layer. Agent-ID, Owner-ID, Authority-Scope as wire-level headers on every request. Scope Enforcement Points refuse over-scoped requests at line rate. Signed Attribution-Records flow into transparency logs. Federated discovery. Cross-organization delegation chains. Every intermediary becomes identity-aware. The application receives requests with verified identity already attached. One protocol. Free. Open. IETF Internet-Draft. Port 4480 registered with IANA. agtp:// reserved. The one question to ask on the floor: does your agent identity ride on the wire, or in the application? SMTP replaced proprietary mail. DNS replaced HOSTS.TXT. TLS replaced application-layer crypto. AGTP is next. chrishood.com/identiverse-ha…

100% of the vendors at @Identiverse are solving agent identity at the wrong layer. Canonical Agent ID on Agent Transfer Protocol. #AGTP #IAM #AgentID

piped.video/LT9_ogdQg9k #AGTP

Your agents have an address now. agtp://your-agent agtp://agent-id agtp://mcp.domain.tld/your-agent agtp://agents.domain.tld/agent-id AGTP, the Agent Transfer Protocol, is the home your agents have been missing. An IETF Internet-Draft. Port 4480 registered with IANA. A reserved agtp:// URI scheme. Open source. Already running. You built the agents. You shipped them. Then somebody from compliance asked how many you have and where they live, and the answers got hard. Identity in a directory is a name tag. A home is something else. AGTP gives every agent a signed Agent Genesis, a canonical Agent-ID derived from cryptographic content, a queryable Manifest, a discoverable presence in ANS, scope enforcement at the wire, and signed Attribution-Records that flow into transparency logs. Elemen lets humans visit any agent by URI and see a clean identity profile. Already running MCP? AGTP composes. The agtp-mcp gateway sits in front of any unmodified MCP server and gives it cryptographic identity, scope enforcement, and signed attribution. Your existing investment keeps working. The substrate underneath gets stronger. Your agents already exist. AGTP is where they can live. chrishood.com/agtp-a-home-fo… #AGTP #Agents #MCP #IETF

In 2009, Google was breached. The attack came through a corporate VPN, killing an assumption that had organized security for a generation: that the inside of the network was a trusted place. Google's response became BeyondCorp. Forrester named it zero trust. NIST codified it in SP 800-207. The White House mandated it for federal agencies. Every serious CISO has been deploying some version for five years. The model works for what it was designed for. The question it was asked is about humans and workloads. Agents are a different question. An agent has no MFA device. No second factor to challenge. No session in the human sense. An agent crosses organizational boundaries by design, where neither operator owns both sides of the policy. An agent has three principals on every request, not one. AGTP applies the zero trust principles to agents. Agent-ID from a signed Genesis. Authority-Scope as a normative header. Signed Attribution-Records in transparency logs. Governance zones enforced at line rate. Behavioral trust scores from the log itself. Same discipline. New actor type. chrishood.com/zero-trust-was… #AGTP #AgentTrust #ZeroTrust

Most agents today are discovered through hardcoded URIs, vendor marketplace APIs, or word of mouth in Slack channels. All three patterns are variants of HOSTS.TXT. They work at small scale. They fail at any scale that resembles a real economy. ANS is the agent internet's DNS. Federated. Signed by default. Scope-enforced. Returns ranked results with live behavioral scores. Capability-based queries rather than name lookups. Federation that preserves provenance. The lookup language is AGTP itself. DNS scaled the web. ANS scales the agent internet. chrishood.com/agent-name-sys… #ANS #DNS #AgentWeb #AGTP

Introducing Elemen. The world's first agent browser. chrishood.com/introducing-el… #AGTP #AgentBrowser #Elemen

The World's first Agent Browser sitting on top of AGTP. Introducing Elemen. Available on github.com/nomoticai/agtp #AGTP #Elemen #AgentWeb #AgentID

REST was built for a human developer reading docs, translating intent into POST against /reservations, with time to compile, debug, and retry. Agents do something else. They reason in natural language about user goals. They have one inference step from "book a table" to whatever verb the server speaks. No second pass, no docs, no compile-time guardrails. We measured the gap. 7,200 trials, four model families. Intent-aligned verbs beat CRUD by up to 29 percentage points at frontier scale. POST /reservations loses to BOOK /reservation. The signal is in the verb itself, dense enough to survive bad documentation. AGTP-API rebuilds the contract layer around the new reader. An 18-method protocol floor every server has to support. A 465-verb open catalog that evolves at deployment cadence rather than IETF cadence. PROPOSE for the capability the server never anticipated, evaluated on the wire and synthesized into a contract at the moment of need. Semantic blocks on every endpoint so agents can compose across organizations they have never met. REST was a private contract between two parties who showed up with prior coordination. Agent APIs are public common ground for parties who have never been introduced. Different problem. Different design. Read the full article below. chrishood.com/designing-apis… #RESTAPI #OpenAPI #AgenticAPI #AGTP #Agents