Agentic Experience Design
By Kurtis Kemple. Framework for designing the interaction layer of AI agents, arguing experience design is the single biggest factor in adoption or abandonment.
Core Thesis
"The single biggest factor in whether an agent gets adopted or abandoned is experience design." Agents represent a fundamentally new interaction paradigm -- partial autonomy -- requiring a distinct design discipline.
The Responsibility Model (Four Layers)
- Human Layer: Owns intent and judgment. Defines what needs to happen, evaluates outcomes. Judgment cannot be delegated.
- Agent Layer: Owns planning and outcomes. Determines what to do given a goal, accountable for results.
- Workflow Layer: Owns automation. Executes structured, repeatable sequences with predictability.
- Tool Layer: Owns execution. Performs single functions on demand without memory or judgment.
The Autonomy Model
- Humans: Full autonomy to intervene at every level
- Agents: Bounded autonomy within defined scopes
- Workflows: Deterministic autonomy (no deviation)
- Tools: Zero autonomy (respond only to invocation)
Three Enabling Conditions
Planning Requires Context
Context is "the informational environment that makes coherent planning possible over time." Includes goals, constraints, decisions, current state, history, and open questions.
Authority Enables Outcomes
Authority must remain aligned with human judgment through: transparency, steerability, resilience under failure, reversibility, guardrails.
Progressive Authority
Authority expands as agents demonstrate competence -- both for individual users earning trust and for admins rolling out capabilities gradually.
Design Posture
"Ship with narrow, inspectable, reversible defaults. Make the authority structure visible."
Critical Failure Mode
Drift: when context management lapses and the agent's understanding of user intent becomes stale and incoherent over longer interactions.