Software has a new user.

Agent 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.

Usetask success
Findretrieval + citation
Actauthority + recovery

Three connected outcomes. Three separate measurement boundaries.

A working definition

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

The field map

AX overlaps developer experience, information retrieval, agent-computer interaction, evaluation, and security. It should not collapse those disciplines into one score.

Usability

Can an agent understand the available capability, choose it correctly, complete the intended task, and recover from failure?

AX-U

Visibility

Can an agent or answer engine find, retrieve, cite, and accurately represent the relevant public information?

AX-V

Delegated action

Can an agent act for a person with explicit authority, bounded permissions, provenance, and a safe path back?

AX-D

Research under review

The 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.

Working synthesis

Definitions, history, and the boundaries of Agent Experience

Inspect →
Literature map

Tool discovery, documentation, routing, and agent tool use

Inspect →
Landscape

Benchmarks, datasets, harnesses, and validity threats

Inspect →
Risk analysis

Permissions, provenance, prompt injection, and recovery

Inspect →

Authority should be inspectable.

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 method
04Practical guidance
03Reviewed finding
02Falsifiable claim
01Public evidence

Building an agent-facing product, evaluation method, or research program? Contact Himadri Mishra for an independent conversation.

Start a conversation