The agentic experience designer
What the role actually involves, what it demands, and how it differs from the design jobs it grew out of.
In one paragraph
An agentic experience designer designs how people delegate work to AI agents and remain in control of the outcome. Where a product designer shapes what someone does directly, this role shapes what happens when a model does it on their behalf: which actions require approval, how intent is shown before an action is taken, how a person interrupts or corrects work in flight, and how the system accounts for what it did. The output is less often a screen and more often a set of decisions about authority, visibility, and recovery.
Why the role appeared
For most of the history of software, the interface and the capability were the same thing. If there was no delete button, nothing got deleted. The affordances a designer put on screen were also the boundary of what could happen, which is why so much design practice is about making the next step obvious and the current state visible.
Agents break that identity. An agent’s capability is its tool set, which is almost always wider than anything rendered on screen. The visible interface no longer tells the user what is possible, and the person is no longer the one performing the steps. Feedback that used to arrive during an action now arrives after it. Errors that used to be prevented by a disabled button now have to be caught, explained, and undone.
Someone has to make those calls deliberately. Where that work is not owned, it tends to get made by default — usually by whoever wired up the tool calls — and the resulting product asks users to trust something they cannot see, predict, or stop.
What the work consists of
Deciding what the agent may do unsupervised
Classifying actions by consequence and reversibility, then deciding which ones proceed silently, which surface for approval, and which are never delegated at all. Reading a page and spending money should not sit at the same permission level, and the boundary between them is a design decision rather than an engineering one.
Making intent legible before action
Showing what the agent is about to do, in terms the person can evaluate, early enough that stopping it is still cheap. This is harder than it sounds: a plan summarised too abstractly cannot be judged, and one shown in full detail will not be read.
Designing interruption and correction
A stop control that works mid-task, a way to redirect without starting over, and a path back from a wrong action. Undo is a design problem here, not just an engineering one — some agent actions reach systems you do not control and cannot be reversed, and the interface has to be honest about which ones.
Calibrating trust
Helping a person build an accurate model of what the agent is reliable at. Both failure directions are real: over-trust leads to unreviewed consequential actions, under-trust means the person supervises everything and gets no benefit from delegating at all.
Designing the account afterwards
What the agent reports having done, at what granularity, and how a person audits it later. For out-of-the-loop work this is the entire user experience — the report is the product.
How it differs from adjacent roles
| Role | Core question | Primary artifact |
|---|---|---|
| Product designer | Can the user complete this task? | Flows and screens |
| Conversation designer | Did we understand each other? | Dialogue and repair paths |
| Agentic experience designer | Does the user understand what is about to happen, and can they stop it? | Permission models, escalation rules, accounts of work done |
A rough test: if you can enumerate every state a user reaches by clicking, you are doing product design. If you can enumerate every turn the system might take, you are doing conversation design. If you can enumerate neither — because the system composes its own sequence of tool calls to satisfy a partly-specified intent — you are doing agentic experience design.
Skills the work demands
- Interaction design fundamentals. The canon still applies; it just needs adapting to a case where the user is not the actor.
- A working model of how agents behave. Not model training, but enough understanding of tool-calling, context, and non-deterministic failure to design against how these systems actually break. You cannot design good recovery for failures you cannot anticipate.
- Risk reasoning. Classifying actions by consequence and reversibility is the load-bearing judgement in this job, and it is closer to security thinking than to visual design.
- Systems thinking. The deliverable is often a permission model or an escalation policy rather than a layout.
- Evaluation literacy. A surface that behaves differently on identical inputs cannot be QA’d by clicking through it. Designers in this role need to be comfortable reasoning about sampling, evals, and failure rates.
- Writing. Much of what calibrates trust is language: how an agent describes what it is about to do, admits uncertainty, and reports what happened.
Common questions
- What does an agentic experience designer do?
- An agentic experience designer designs how people delegate work to AI agents and stay in control of the result. The work covers deciding which actions an agent may take unsupervised, how it shows intent before acting, how a person interrupts or corrects it mid-task, and how the system accounts for what it did afterwards. It is closer to designing a working relationship than designing a screen.
- How is it different from a product designer?
- A product designer shapes what a person does directly. An agentic experience designer shapes what happens when someone else — a model — does it on their behalf. The interface stops being the place where work happens and becomes the place where work is authorised, monitored, and corrected. Most of the UX canon assumes the user is the actor; that assumption is what breaks.
- Is agentic experience design the same as conversation design?
- No, though they overlap. Conversation design concerns dialogue: turn-taking, prompts, repair, tone. Agentic experience design concerns delegated action, where the system commits to consequences in the world. A conversation designer asks whether the system understood; an agentic experience designer also asks whether the person understood what the system was about to do, and whether they could stop it.
- What skills does the role require?
- Interaction design fundamentals, a working understanding of how models and tool-calling behave (including that they fail non-deterministically), the judgement to reason about risk and reversibility, and enough systems thinking to design permission and escalation rather than screens. Familiarity with evaluation matters too, because you cannot QA a non-deterministic surface by clicking through it.
- Does the job title exist yet?
- The discipline is real and hiring is happening, but the title is not standardised. The work is currently posted under a range of names — product designer with an AI focus, AI interaction designer, and related variants — and is often absorbed into an existing design role rather than hired for separately. Expect the responsibilities before the title.
Where to go next
- What is Agentic Experience Design? — the discipline defined
- Interaction models — human in, on, and out of the loop
- Design patterns — reusable solutions for agentic flows
- Design principles — autonomy, honesty, safety, consent
Last reviewed . The title and its scope are still settling; this page is revised as the role takes clearer shape.
Stay Updated
Analysis of AI search, crawler policy and agent standards — sent when there is something worth reading, roughly twice a month. Unsubscribe anytime.