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 |
A copilot assists a user who is present and supervising. An agentic AI acts as a delegate, operating autonomously in the user's absence to achieve a goal. The copilot suggests; the agent decides. The copilot augments human capability; the agent exercises delegated authority.
The copilot-to-agent transition is the defining design challenge because it changes the trust model entirely. Copilots operate under direct supervision with low-stakes suggestions. Agents operate with delegated authority and real-world consequences. The transition requires trust architecture, delegation design, operational envelopes, and recovery mechanisms that copilot design does not address.
Yes, but the transition is not automatic. Moving from copilot to agent requires deliberately expanding the AI's operational envelope, building trust architecture that supports autonomous operation, designing delegation frameworks that define authority boundaries, and creating recovery mechanisms for when things go wrong. It is a design challenge, not just a capability upgrade.
Autonomy increases risk because the agent acts without real-time human oversight. Errors can compound before intervention, authority can drift beyond the original mandate, and accountability becomes harder to trace. These risks are manageable through trust architecture, operational envelopes, and recovery design - but they must be designed for, not assumed away.
This model is built on a foundation of low-trust interaction. The system is not expected to perform flawlessly without oversight. Its value comes from its ability to accelerate the user's workflow, not from its capacity for autonomous action. The design of copilots, therefore, prioritizes a tight feedback loop, clear suggestions, and easy-to-correct outputs. Consider the difference between a copilot suggesting a line of code and an agent tasked with 'deploying the new feature to the staging server, running all tests, and reporting back on success or failure.' The latter involves a multi-step process with potential for unexpected challenges. The agent must be able to navigate these challenges, make independent decisions, and take responsibility for the outcome. This is the core of Agentic systems are not just more powerful; they operate on a different plane of interaction. They require a robust The most critical distinction between copilots and agentic AI is the user's required presence. Copilots are designed for This shift from presence to absence has profound design implications. A system designed for absence cannot rely on constant user feedback. It must have a deeper understanding of intent, a robust framework for handling errors and ambiguity, and a clear mechanism for reporting outcomes. It requires designing for a The 'delegation gap' is the chasm between a tool that can help you do a task and a delegate that can do the task for you. Many systems are marketed as 'agents' but are, in practice, sophisticated copilots. They may automate a few steps, but they still require the user to orchestrate the overall process and make key decisions. True delegation requires bridging this gap. This involves more than just stringing together a series of automated actions. It requires the system to possess a model of the user's intent, the ability to plan and re-plan, and the authority to act within a defined Closing the delegation gap is the central challenge of You cannot ha