For thirty years, designers shaped the screen. We built the flows, mapped the journeys, crafted the affordances through which human intention flowed into digital action. We became very good at designing for systems that wait.
We are now designing for systems that act.
Autonomous agents do not wait for instructions. They anticipate. They orchestrate. They operate in the world while the human is absent - booking, filing, deciding, transacting - on behalf of the person who gave them permission, once, some time ago. The tools we built for screen-based design - user flows, wireframes, interaction patterns, usability heuristics - were forged for a world that no longer exists as the primary design challenge.
Agentic Experience Design is not UX for AI. It is the design of trust-governed relationships between humans and systems that act autonomously in the world. It is a new discipline.
This manifesto is a founding claim. It states what AXD is, what it requires, and what it must not become. It is the living document at the centre of a new field. It will be revised. It will be contested. That is the condition of any discipline worth building.
What AXD Is
Agentic Experience Design is the design discipline concerned with how humans delegate, calibrate, observe, interrupt, and recover trust in autonomous systems.
It operates at the boundary between human intention and machine action - a boundary that UX has never been required to design because, until now, machines did not act without being told to.
AXD draws from human factors, cognitive psychology, behavioural science, ethics, systems design, and service design. But it synthesises these into a new practice because none of them, alone, is sufficient for the challenge of designing human-agent relationships at scale. The AXD Vocabulary defines the twenty canonical terms of this new discipline.
What AXD Is Not
AXD is not conversational UX. Designing a chatbot is not AXD - it is interface design for a conversational modality. AXD concerns systems that act, not systems that converse.
AXD is not prompt engineering. Crafting inputs to language models is a craft skill. AXD is the design of the systemic human experience of agentic capability - including the moments when no prompting occurs at all.
AXD is not AI ethics. Ethical frameworks are necessary but insufficient. AXD is the design practice through which ethical principles are made operational - built into the architecture of delegation, consent, and observability.
AXD is not the future of UX. It is a parallel discipline that UX will need to understand, collaborate with, and in some cases cede ground to.
The Five Founding Principles
Agency Requires Intentional Delegation
Agents operate on trust that was given, not earned in the moment. Every agentic system begins with an act of delegation — a human granting permission for autonomous action. AXD designers architect this delegation: its scope, its duration, its conditions for expansion, and its mechanisms for revocation. Delegation without design is vulnerability.
Read: Delegation Design →Trust is the Primary Material
Where UX works in attention and affordance, AXD works in trust. Trust in agentic systems is not binary — it is calibrated, contextual, and fragile. AXD designers are trust architects. They design the conditions under which trust is formed, the signals that sustain it, and the recovery pathways when it breaks. A system that cannot recover from a trust failure is not fit for purpose.
Read: Trust Architecture →Absence is the Primary Use State
In screen-based design, the user's presence is assumed. In agentic design, the user's absence is the dominant operating condition. The most consequential experiences happen when no one is watching. AXD must design for the unobserved: for agents that act with integrity when unsupervised, and for experiences that make the absent human feel present in the outcomes.
Read: The Invisible Layer →Relationships Have Temporality
Agentic experiences are not transactions - they are relationships. They accumulate history, demonstrate character, and require consistency across time. AXD designers think in terms of relational arcs, not task flows. The first interaction is not the important interaction. The hundredth is. Designing for temporal depth is a new skill that screen-based design did not require.
Outcomes Replace Outputs
Screen-based designers specify what appears. AXD designers specify what results. The distinction is fundamental. When the path is chosen by the agent, the designer cannot specify the journey — only the destination, the constraints, and the conditions under which the human must be re-engaged. Outcome specification is the primary design artifact of AXD.
Read: Composable Interfaces →What AXD Demands of Designers
AXD requires designers to develop fluency in domains they have not traditionally needed to master: the architecture of consent, the psychology of trust and risk perception, the grammar of agent observability, the ethics of autonomous decision-making, and the design of failure and recovery at system scale. The AXD Practice section provides the frameworks for developing these capabilities.
It also requires designers to give something up: the assumption that the interface is the experience. In agentic systems, the interface may be the exception - the interruption surface that appears only when something needs a human decision. The experience is everything else.
This is uncomfortable. It is also necessary. The discipline that rises to meet this challenge will shape how humanity navigates the most significant transition in human-computer interaction since the invention of the graphical user interface.