Agentic Experience Design (AXD) is the discipline for designing trust-governed relationships between humans and autonomous AI systems. Founded in September 2024 by Tony Wood in Manchester, United Kingdom, AXD addresses how humans delegate, calibrate, observe, interrupt, and recover trust in agentic AI.
| Dimension | Traditional UX | Agentic Experience Design (AXD) |
|---|---|---|
| Primary material | Attention and affordance | Trust and delegation |
| User state | Present, navigating | Absent, delegating |
| Design output | Screens and interfaces | Outcomes and constraints |
| Temporal model | Session-based | Relationship-based |
| Success metric | Task completion | Trust calibration |
The four phases are: (1) Formation - the initial phase where the human first encounters the agent and develops expectations. (2) Calibration - the ongoing process of adjusting trust based on actual performance. (3) Maintenance - the steady state where trust is established and must be preserved through consistent performance. (4) Recovery - the phase following a trust violation, requiring acknowledgment, remediation, scope reduction, and gradual re-expansion of authority.
Because the mechanisms that build trust in one phase may be counterproductive in another. Extensive explanation builds trust during formation but feels patronising during maintenance. Minimal communication works during maintenance but feels opaque during formation. The agent must be designed to recognise which phase the relationship is in and adapt its behaviour accordingly - this phase-awareness is a structural requirement of trust architecture.
The cold start problem is the paradox at the heart of trust formation: the human must trust the agent enough to delegate a task, but the agent has not yet had the opportunity to earn trust through performance. Every human-agent relationship must cross this gap. AXD addresses it through accurate self-representation, low-stakes first tasks, transparent reasoning, and explicit boundary communication.
Trust recovery follows a five-step sequence: acknowledgment (clear, unminimised admission of the failure), explanation (honest account of what happened and why), remediation (concrete action to address consequences), scope reduction (voluntary contraction of the agent's authority), and gradual re-expansion (earning back authority through consistent performance at the reduced level). An agent that recovers well can emerge with stronger trust than before the failure.
The most common mistake in designing trust for agentic systems is treating it as a binary - the user either trusts the agent or does not. In reality, trust is a A new user encountering an agent for the first time is in a fundamentally different trust state than a user who has delegated successfully for six months. And a user whose agent just made a significant error is in a different state again. The design patterns that are appropriate for one state may be actively harmful in another. Excessive explanation during the maintenance phase feels patronising. Minimal explanation during the formation phase feels opaque. The same behaviour produces opposite effects depending on where the user is in the trust lifecycle. This is why AXD treats trust as a lifecycle rather than a metric. A metric captures a snapshot. A lifecycle captures the trajectory - the direction, velocity, and phase of trust at any given moment. Designing for the lifecycle means designing different experiences for different phases, with smooth transitions between them. The formation phase is governed by a paradox: the human must trust the agent enough to delegate, but the agent has not yet had the opportunity to earn trust through performance. This is the The Onboarding and Capability Discovery Framework addresses this phase directly, providing design patterns for introducing an agent's capabilities without over-promising or under-representing them. Key design principles for the formation phase include: The agent must communicate what it can and cannot do with precision. Over-promising creates expectations that will be violated, destroying trust before it forms. Under-representing capabilities means the human never discovers the agent's value. The agent should be given opportunities to demonstrate competence on tasks where failure is inconsequential. A financial agent that begins by categorising expenses - rather than executing trades - gives the human evidence of competence without risk. During formatio