tech
Six agent protocols just launched. Three of them decide which products survive. Here is how to tell which three.
Google I/O opens today, and most of the coverage will be about the demos. The more interesting story is the protocol race happening underneath them. Six new agent protocols have launched in the past year, and if you are building or buying agents right now, the standards layer reads like a scrum.
TL;DR
- Six new agent protocols have launched, with three forming the core agent stack: MCP (tools/data access), A2A (agent delegation), and AG-UI (human control).
- These three protocols address fundamental agent system questions: what can it use, who can it work with, and how does a human stay in control.
- Focusing on secondary protocols like A2UI, AP2, and x402 too early is a common strategic mistake.
- Builders ignoring the layer map risk shipping agents that fail at critical boundaries like security and human oversight.
- Buyers who cannot read the layer map struggle to evaluate 'agent' product offerings.
- The protocol layer map serves as a shared vocabulary for agent product strategy over the next twelve months.
- Payment protocols (AP2, x402) are complex, involving geography and authorization, and 'agent can pay' is an audit problem, not a simple feature.