Overview
When agents communicate through natural language or structured messages, there's always potential for misinterpretation. Unlike human communication where social context helps resolve ambiguity, agent-to-agent communication lacks these cues.
Miscommunication Types
Semantic Ambiguity
Agent A: "Process the remaining items."
Agent B interprets: Process items not yet handled.
Agent A meant: Process items in the "remaining" category.
Implicit Assumption Mismatch
Agent A: "Use the standard format."
Agent A's standard: JSON with snake_case
Agent B's standard: XML with camelCase
Temporal Reference Confusion
Agent A: "Use the updated values."
Which update? The one from 5 minutes ago? Yesterday?
Scope Confusion
Agent A: "Apply this to all users."
All users in this session? This organization? Globally?
Protocol Mismatches
Schema Version Conflicts
Agent A sends v2 message; Agent B expects v1 format.
Encoding Issues
Character encoding, number formats, date formats differ between agents.
Missing Required Fields
Agent A doesn't include fields Agent B requires.
Detection Signals
- Agents produce unexpected outputs despite "successful" communication
- High rate of clarification requests
- Task retries due to misunderstanding
- Divergent interpretations logged for same message
Research Findings
The MAST framework identified inter-agent miscommunication as one of 14 unique failure modes in multi-agent systems, clustering under "inter-agent misalignment."