Explainable Actions
CanonicalConfidence
Cognitive Load
Low
Evidence
production validated
Impact
feature
Ethical Guardrail
Never act silently or with vague explanations. Never hide reasoning to appear more confident. Always offer Show me more detail for deeper transparency.
Design Intent
Users cannot trust what they cannot understand. Explainable Actions makes every agent or system action transparent by showing exactly what was done, why it was done, and what data or logic was used.
Psychology Principle
Users cannot trust what they cannot understand.
Description
Make every agent action transparent by showing what was done, why, and what data was used -- eliminating the black box feeling.
When to use
Every agent or automated action -- especially at confirm_execution or autonomous levels, or when confidence is below 90%.
Example
Claude/Grok Agent Response: After every action -- I updated the section because [specific reason + data sources]. Here's exactly what changed. Undo or edit?
Autonomy Compatibility
Behavioral Objective
Users understand and trust every agent action because the reasoning is always visible and verifiable.
- Higher acceptance of agent suggestions
- More frequent corrections when needed
- Stronger long-term confidence in the AI
Target Actor
role
Everyday user
environment
Mixed human-AI decision workflows
emotional baseline
Needs to understand before trusting
ai familiarity
medium
risk tolerance
medium
Execution Model
pre_action_preview
Show the plan before executing at confirm or autonomous levels.
User is surprised by what the agent does.
post_action_explanation
Immediately after the action, surface the why and how.
User has to ask or hunt for the reason.
on_demand_depth
Allow users to expand for full transparency.
User wants more detail but cannot easily get it.
Failure Modes
Explanation is too vague or generic
Always include specific data sources and logic
Explanation overwhelms the user
Default to short version with optional expand
Explanation arrives too late
Show preview before autonomous actions
Agent hides low-confidence reasoning
Explicitly call out uncertainty
Explanations feel defensive or overly formal
Use warm human tone
Agent Decision Protocol
Triggers
- Any agent action is about to occur
- User shows hesitation or surprise
- Confidence is below 90% or stakes are high
Escalation Strategy
L1: Diagnose the failing element via behavioral_signals
L2: Nudge -- adjust copy, timing, or visual salience
L3: Restructure -- simplify flow, add progressive disclosure, restructure form
L4: Constrain -- lock Autonomy Dial to confirm_execution, add Strategic Friction
L5: Yield -- flag for human designer or domain expert review
Example
Agent auto-fills a section -> immediately shows I pulled this from your last three similar documents because the language matched 94% -- here are the sources.
Behavioral KPIs
Primary
- % of actions where user views or expands the explanation
- User trust score after explained actions
- Correction rate when explanation is shown
Risk
- Why did it do that? questions
- Distrust reports after agent actions
Trust
- User-reported understanding of agent behavior
- Autonomy Dial usage when explanations are provided
Decay Monitoring
Revalidate when
- Agent capabilities or data sources change significantly
- New interaction paradigms emerge
- User familiarity with AI explanations evolves
Decay signals
- Rising Why? questions
- Drop in explanation view rates
- Feedback that explanations feel stale or unhelpful
Pattern Relationships
Amplifies
Conflicts with