Agentic experience, or AX, is a design discipline focused on how humans and AI agents share a workflow, covering how people delegate tasks, receive outcomes, and maintain trust, while agents perceive and navigate interfaces and collaborate with users.  The term was introduced by Mathias Biilmann of Netlify.

Traditional UX is deterministic, while agentic experience introduces probabilistic behavior – an AI agent may take actions autonomously, and the design must account for transparency, override controls, and trust-building patterns that don’t exist in conventional interface design.

In addition to the core UX skills – interaction design, user research, information architecture, and UX writing – the new ones are a conceptual understanding of LLMs, designing for uncertain and probabilistic outcomes, and prompt design as a content discipline.

Earlier is better, but it’s rarely how it happens. Ideally, AX runs alongside agent development so the interaction model is designed against real system behavior. That said, AX is also the right response when an agent is already deployed but adoption is lower than expected; it’s as much a diagnostic discipline as a design one.

The most common failure is that users don’t adopt the agent or quietly work around it. They revert to manual tasks because they don’t have a clear mental model of what the agent will do, can’t tell when it’s acted, and don’t know how to recover if something goes wrong. AX addresses exactly that gap.

Usability testing, task analysis, and interviews all transfer, but the questions change. In conventional UX research, you’re testing whether users can complete a task. In AX, you’re testing whether users understand what the agent did, whether they trust it to do it again, and whether they know how to step in when it’s wrong.

The UX/UI designer owns what the user sees – the interface layer across screen, chat, and voice. The AX designer owns the conditions under which the agent acts and how trust is earned over time. In practice, they work on the same surfaces but from different starting points: one from the user’s mental model, one from the agent’s behavior.

Wireframes and prototypes still apply, but AX adds deliverables that don’t exist in conventional UX: trust architecture maps that define what the agent handles autonomously vs. what needs a confirmation step, and intervention design specs that describe exactly what users see when the agent acts, fails, or needs input.