Foundations and terminology
The public history of Agent Experience, competing definitions, conceptual boundaries, and a neutral definition for the AI-agent context.
Research, with the seams visible.
The initial corpus maps a fast-moving field without pretending it is settled. Reports remain working research until every material claim is normalized into an evidence record and reviewed.
Public sources
Tool discovery and use
Research on documentation, API retrieval, tool selection, planning, function calling, execution, verification, and recovery.
Discovery and visibility
SEO, GEO, AEO, retrieval, citation behavior, entity signals, and the limits of visibility metrics.
Evaluation and datasets
Tool, browser, computer-use, coding, reliability, and harness-sensitivity benchmarks—plus what their scores actually measure.
Surfaces and standards
Protocols, APIs, SDKs, CLIs, tool descriptions, skills, schemas, documentation, authentication, and runtime enforcement.
Trust and safety
Delegated authorization, least privilege, prompt injection, tool poisoning, sandboxing, provenance, auditability, and recovery.
Starting points
These public materials anchor parts of the current research map. Inclusion is not endorsement, and vendor-primary material establishes only what its publisher publicly claims.
- AgentExperience.ax — public category site and concepts.
- Introducing Agent Experience — category-origin essay.
- Generative Engine Optimization — foundational GEO research.
- AI Agents That Matter — benchmark-validity critique.
- Model Context Protocol specification — current public protocol surface.