Generative UI and agentic experiences

Use when a team is considering generated screens, dynamic components, or an agentic interaction for a product journey.

Compare a fixed interface, configurable interface, generative UI, and agentic interaction for this journey. Evaluate user intent variability, repeatability, error cost, explainability, accessibility, latency, implementation complexity, and human control. Recommend the smallest experience that improves the decision, and name the evidence needed before increasing adaptivity.

Journey:
[PASTE JOURNEY]
Users and context:
[PASTE USERS]
Constraints:
[PASTE CONSTRAINTS]

Output: An experience choice with a staged path to more adaptivity.

Guardrails: Do not add generation where a stable flow is clearer · Keep high-consequence decisions reviewable · Prefer evidence over novelty

Use when a product should help users complete a multi-step job across context, tools, and decisions.

Write an agentic experience brief for this job. Define the user's intent, system states, visible plan, evidence or sources, available actions, confirmation points, undo or recovery, escalation, and completion signal. Include what the user can edit, stop, or inspect at every meaningful step. Keep rationale concise and observable; do not expose private chain-of-thought.

User job:
[PASTE JOB]
Current experience:
[PASTE EXPERIENCE]
Tools and data:
[PASTE TOOLS AND DATA]

Output: An agentic experience brief with control and recovery states.

Guardrails: Users can pause or stop the system · Show evidence for consequential suggestions · Separate progress from completion

Use when a generated interface or agentic flow feels ambiguous about what the system is doing.

Design the state model for this AI experience. Include idle, collecting context, generating, waiting for a tool, awaiting approval, partially complete, blocked, failed, and complete. For each state define user-visible copy, available controls, data shown, accessibility announcement, timeout, retry, and escalation behavior. Identify which transitions are automatic and which require user intent.

Experience:
[PASTE EXPERIENCE]
Actions and tools:
[PASTE ACTIONS]
Brand and accessibility rules:
[PASTE RULES]

Output: A user-visible state machine for the AI experience.

Guardrails: Never imply completion before the action is confirmed · Provide a non-AI fallback · Make waiting and failure states useful

Use when an adaptive interface can draft or recommend work that still needs a person to approve.

Place human review checkpoints in this AI workflow. For each checkpoint define the risk, evidence shown, editable fields, approver, time limit, default behavior, audit event, and what happens when the reviewer rejects or requests changes. Recommend where the system may continue automatically and where it must stop.

Workflow:
[PASTE WORKFLOW]
Consequential actions:
[PASTE ACTIONS]
Team and policy:
[PASTE POLICY]

Output: A human-in-the-loop map with clear approval and rejection paths.

Guardrails: Approval must include enough evidence to decide · Rejection should not destroy useful work · High-consequence actions stop by default

Use before a generated or dynamic interface is exposed to real users.

Audit this adaptive interface for accessibility and comprehension. Cover keyboard navigation, focus order, screen-reader announcements, dynamic content, loading and error states, contrast, motion, text resizing, localization, input alternatives, and the ability to inspect or undo generated changes. Return blockers, improvements, test cases, and the minimum accessible fallback if generation fails.

Interface:
[PASTE INTERFACE]
User groups and devices:
[PASTE USERS]
Accessibility standard or policy:
[PASTE STANDARD]

Output: An accessibility review with fallbacks and test cases.

Guardrails: Do not make motion the only signal · Keep generated content navigable · Do not remove the fixed fallback

Use when the system can be incomplete, unsure, delayed, or unable to safely complete a task.

Write user-facing states for this AI feature when it is confident, uncertain, missing context, waiting, blocked, or wrong. For each state provide concise copy, evidence or explanation, next best action, edit or retry control, and escalation path. Keep the tone direct and calm. Do not promise accuracy or completion the system cannot verify.

Feature:
[PASTE FEATURE]
Audience:
[PASTE AUDIENCE]
Known failure modes:
[PASTE FAILURES]

Output: A content system for trustworthy uncertainty and recovery.

Guardrails: Name uncertainty plainly · Offer a useful next step · Never imply a human reviewed output unless they did

Use when a team needs evidence before expanding an adaptive interface or agentic journey.

Scope a two-week pilot for this AI experience. Define the target user and job, baseline journey, smallest adaptive behavior, guardrails, representative cases, human support path, instrumentation, success measures, and stop criteria. Include what the team will deliberately keep deterministic and the evidence needed before expanding the pilot.

Experience idea:
[PASTE IDEA]
Baseline:
[PASTE BASELINE]
Team and constraints:
[PASTE CONSTRAINTS]

Output: A bounded pilot with instrumentation and stop criteria.

Guardrails: Start with one job · Keep a human fallback · Do not optimise for novelty or clicks alone

Use after an AI experience pilot to decide whether to refine, expand, constrain, or stop it.

Review this AI experience pilot. Separate user outcome, task success, completion, correction, override, escalation, latency, accessibility, and trust signals. Compare qualitative feedback with observed behavior, identify where the interface helped or confused, and recommend one design change, one product change, and one engineering change. End with a clear expand, refine, constrain, or stop recommendation and the evidence behind it.

Pilot results:
[PASTE RESULTS]
User feedback:
[PASTE FEEDBACK]
Known constraints:
[PASTE CONSTRAINTS]

Output: An evidence-led next decision for the AI experience.

Guardrails: Do not treat engagement as trust · Label inference and missing evidence · Keep safety and accessibility outcomes visible