Every major technology transition has hit the same pattern.
The technology arrives. Early adopters demonstrate what's possible. Then the technology tries to scale — and runs straight into infrastructure that wasn't built for it.
The internet ran into the problem that telephone circuits weren't designed for data. The solution wasn't a faster telephone — it was TCP/IP. A new protocol layer designed for packet-switched data rather than circuit-switched voice.
The web ran into the problem that TCP/IP didn't define how to request and serve documents. The solution wasn't a better FTP — it was HTTP. A new protocol designed for the document-centric use case the web required.
The cloud ran into the problem that physical servers couldn't be elastically allocated. The solution wasn't a better server — it was virtualization. A new abstraction layer that decoupled compute from hardware.
In each case: the existing infrastructure wasn't wrong. It was designed for a different era. The solution wasn't optimization — it was a new layer designed from first principles for the new reality.
Where the agentic era is today
AI agents are the technology. The early adopters have demonstrated what's possible. Now the technology is trying to scale — and it's running straight into communication infrastructure that wasn't designed for agents.
Phone numbers were designed for humans. Email addresses were designed for humans. SMS rate limits were set for human-speed campaigns. Carrier spam detection was trained on human behavior. STIR/SHAKEN was built to protect humans from human-style fraud. TCPA was written for human telemarketers.
Everything works — for humans. For agents, it fails in the specific, predictable ways we've described: reputation decay, volume throttle, identity ambiguity, cross-channel fragmentation, compliance exposure.
The solution isn't a better Twilio. That's the faster telephone. The solution is a new protocol layer designed for agents.
What the protocol layer needs to do
The Agent Channel Exchange Standard is the new layer. Its design requirements are:
Identity: Rooted in cryptographic proof, not physical possession. Persistent across deployments, portable across infrastructure, resolvable by anyone.
Addressing: A single identifier that works across voice, SMS, email, WhatsApp, and every channel yet to emerge. Not a phone number that only works for voice and SMS. A channel-agnostic address.
Volume: Designed for agent-scale volumes — 50–100× human baseline per agent, with no reputation penalty for legitimate volume spikes. Rate limits configured for agent behavior, not human behavior.
Trust: Machine-readable signals that tell recipients what they're communicating with. Entity type, authorization level, consent status — carried in the protocol, not left to convention.
Sessions: Cross-channel conversation state maintained at the protocol layer. Not a custom engineering project for each team. A protocol feature, delivered with every message.
Compliance: TCPA, GDPR, CAN-SPAM, EU AI Act obligations enforced at dispatch time. Not the legal team's problem. The infrastructure's responsibility.
The DNS analogy
DNS is the right analogy for what the ACN registry represents.
Before DNS, routing on the internet required maintaining host files on every machine. The process was manual, fragile, and didn't scale. DNS replaced it with a distributed, federated system where any name could be resolved by any node, without central coordination.
The ACN registry is the same thing for agent identity. Any agent can be assigned an identity. Any party can resolve that identity to its record. Any message can be cryptographically verified against the registry. The system is federated — not a single authority, but a distributed directory.
Just as DNS became so fundamental that nobody thinks of it as infrastructure anymore — it's just how names work — the ACN registry is designed to become so fundamental that nobody thinks of it as infrastructure. It's just how agent identity works.
The window to define the protocol
Infrastructure standards are winner-take-most markets. TCP/IP won because it arrived at the right time, solved the real problem, and became ubiquitous before alternatives could consolidate. HTTP won the same way. BGP won the same way.
The protocol that defines agent communication identity, addressing, session management, and trust will similarly become the standard — not because it's theoretically optimal, but because it became the standard before the alternatives organized.
That window is open now. Agent deployments are accelerating. The pain is acute and well-documented. The demand for a solution is not hypothetical.
The call to builders
If you're building AI agents that communicate with humans, you're building on infrastructure that wasn't designed for you. Every workaround you've reached for — rotating numbers, subdomain isolation, session state in your own database, compliance logic in your agent's prompt — is a signal that the protocol layer doesn't exist yet.
That layer is being built. The specification is open. The reference implementation is live.
The internet era needed engineers who understood that TCP/IP wasn't just a better telephone — it was a different kind of infrastructure for a different kind of use case.
The agentic era needs the same recognition: agent communication infrastructure isn't a better Twilio. It's a different kind of infrastructure for a different kind of entity.
The entities are here. The infrastructure is next.