Usability
Can an agent understand the available capability, choose it correctly, complete the intended task, and recover from failure?
AX-UAgent Experience is the quality of software as experienced by AI agents acting on behalf of people. This field guide maps what they can find, use, verify, and safely do.
Three connected outcomes. Three separate measurement boundaries.
Agent Experience is the quality of a digital product or platform as experienced by AI agents acting on behalf of human users—including whether the agent can discover, access, understand, execute, recover, and be observed within the system.
Neutral synthesis from public definitions and adjacent research · Working draft, August 2026
AX overlaps developer experience, information retrieval, agent-computer interaction, evaluation, and security. It should not collapse those disciplines into one score.
Can an agent understand the available capability, choose it correctly, complete the intended task, and recover from failure?
AX-UCan an agent or answer engine find, retrieve, cite, and accurately represent the relevant public information?
AX-VCan an agent act for a person with explicit authority, bounded permissions, provenance, and a safe path back?
AX-DThe current corpus is public-source working research, not a finished standard. Each report exposes uncertainty and must pass claim-level review before publication as guidance.
Every durable insight should trace backward from practical guidance to a reviewed finding, a falsifiable claim, and public evidence or a reproducible public run.
Read the publication methodBuilding an agent-facing product, evaluation method, or research program? Contact Himadri Mishra for an independent conversation.
Start a conversation