Let's do the math.
There are approximately 1.3 billion knowledge workers globally. The consensus forecast — stated independently by Jensen Huang, Sam Altman, and the founders of every major agent framework — is a 1:100 ratio of humans to AI agents.
That's 130 billion agents.
Agents communicate at 50–100× the rate of humans. A human sends roughly 250 communication events per day. A single AI agent orchestrating a workflow might initiate or receive hundreds of touchpoints per hour.
At 50 events per agent per day:
130 billion agents × 50 communication events = 6.5 trillion messages per day.
For comparison: all human communication on Earth totals approximately 437 billion events per day. Email, SMS, phone calls, WhatsApp — everything.
At the 1:100 ratio, agent communication will be ~15× all human communication combined.
The infrastructure that was built
Today's global communication infrastructure handles ~437 billion human communication events per day. It took 150 years to build — from Bell's telephone to 5G carrier networks, from SMTP to Gmail's infrastructure, from GSM to A2P SMS.
Every piece of that infrastructure was sized for human volume.
The SMS network can handle approximately 5 billion messages per day globally. The email network handles approximately 330 billion. Phone calls: approximately 2 billion per day.
Add it up: ~337 billion messages per day for the channels agents most need.
Against a projected agent demand of 6.5 trillion per day.
The infrastructure that exists covers roughly 5% of what agents will need.
Why you can't just scale the existing infrastructure
The instinct is to say: scale up. More servers. More capacity. Better APIs.
This misses the fundamental issue. The existing infrastructure isn't undersized for agents — it's miscategorized for agents.
Consider what "scaling" Twilio would actually require:
The NANP phone number system has 8 billion possible numbers. The Fortune 500 deploying 100 agents per employee needs hundreds of millions of agent identities from that pool alone.
You cannot scale your way through a category error.
What a designed-for-agents infrastructure looks like at this scale
Infrastructure built for agent-scale communication has to solve different problems than human-scale infrastructure:
Identity: Not SIM-card-based, not constrained by numbering plans. Cryptographic, namespace-based, practically unbounded.
Volume elasticity: No reputation penalty for volume spikes. An agent that sends 10 messages one minute and 10,000 the next shouldn't trigger spam detection. The system needs to distinguish legitimate volume variance from malicious volume patterns.
Channel agnosticism: At scale, the agent shouldn't care which channel delivers a message. The infrastructure should select the optimal channel based on the message type, recipient context, and channel availability.
Compliance automation: At 6.5 trillion messages per day, manual compliance review is impossible. TCPA consent verification, GDPR data handling, CAN-SPAM opt-out management — these have to be protocol features, not legal team tasks.
Trust signals: At the volumes agents will generate, humans need machine-readable signals to understand what's communicating with them. "This is an agent" is not optional information — it's necessary for the recipients of 6.5 trillion daily messages to understand what's happening.
The infrastructure moment
Every major technology transition has an infrastructure moment: the point where the existing substrate hits its limits and a new layer has to be built.
The internet needed TCP/IP. Not a faster telephone network — a new protocol designed for packet-switched data rather than circuit-switched voice.
The cloud needed virtualization. Not bigger servers — a new abstraction layer designed for elastic resource allocation rather than fixed capacity.
The agentic era needs a new communication protocol. Not a faster Twilio. Not a better sendgrid. A protocol designed from the ground up for the entities that will actually be generating 6.5 trillion communication events per day.
That infrastructure doesn't exist yet. The scale it needs to handle is unprecedented. The category it needs to serve is new.
The window to define it — the protocol, the identity system, the trust framework — is open right now.