AI Agent Protocols.
Right now, we're living through the early internet all over again. Before TCP/IP standardised how computers talked to each other, every system had its own way of communicating.
Industry expert Laurie Voss from LlamaIndex put it bluntly at a recent conference: most of the 14+ agent protocols currently in development are "half-baked at best" - see above video.
Why This Matters for Your Projects
These protocols determine whether your AI agents can:
Work together
Access the tools they need
Scale beyond proof-of-concept demos
Agent protocols define how:
Agents discover each other
Authenticate
Exchange information, and
Collaborate on complex tasks
Get this wrong, and you're building isolated AI islands that can't share data or coordinate effectively.
The Two Main Categories
Context-Oriented Protocols: Help agents access external tools and data sources. Think connecting your AI assistant to your CRM or database.
Inter-Agent Protocols: Enable direct communication between different agents. Imagine your customer service agent collaborating with your inventory agent to resolve complex queries.
The Challenge
We're in a fragmented market where:
Every major tech company (Google, Anthropic, IBM) is pushing their own standard
Open-source communities are building internet-scale alternatives
Your protocol choice today will determine how flexible and future-proof your AI investments become
Choose wrong, and you might find yourself locked into a vendor's ecosystem or stuck with protocols that can't scale.
The good news? One protocol is emerging as the clear frontrunner for immediate business needs.
Next week: We'll dive into context-oriented protocols and why MCP (Model Context Protocol) is becoming the foundation most businesses are building on.