As AI takes experiences beyond the traditional interface, designers must rethink how context, trust, and control shape every interaction.
Ben Mitchell
Head of Product Design at Zendesk
최종 업데이트 2026년 11월 6일
For most of my career, "the experience" and "the screen" were close enough to be the same thing. If you wanted to know what a product felt like, you opened it and looked. That shortcut has stopped working, and I think it's one of the more interesting things to happen to design in a long time.
More and more of what people get from Zendesk now reaches them through somewhere else: a messaging thread, a voice call, an AI assistant, another company's agent. Perhaps the most over-used term in the industry has described this as “headless”. A headless experience is when people access a product's capabilities without using its own interface. For Zendesk, that could mean interacting with an AI agent through ChatGPT, a Slack workflow, or another application. Zendesk AI Agents “now operate across messaging, email, voice, and AI platforms like ChatGPT and Gemini, maintaining shared context and continuity across interactions" (Zendesk). Often there is no Zendesk screen in the picture at all.
For designers, the challenge is making sure the experience works wherever Zendesk shows up.
It's tempting to call that an engineering problem. Expose the data, publish the API, and let the experience take care of itself. I'd gently push back. Experiences don't take care of themselves. Someone designs them, or they get designed by accident, and you can usually tell which.
For designers, the challenge is making sure the experience works wherever Zendesk shows up. That means understanding what someone is trying to do, giving them the right context and information, and helping them take action with confidence. These are the things that make a great experience. A standard interface that you control is just one way of delivering them.
Why context matters more than pixels
Here's the idea I keep coming back to. Great experiences are made of context, not pixels. The pixels were only ever how the work showed up. The real work was understanding what someone was trying to do, knowing what mattered right then, and helping them act with confidence.
Nielsen Norman Group says it plainly: "A core assumption needs updating: “user” is no longer synonymous with “human.”” Once that's true, the design problem moves. Jay Goradia puts it well in ACM Interactions: "Agentic systems move important decisions outside that visible surface. The design problem is therefore larger than interface usability: Designers also need to make delegated system behavior legible."
So my working definition is simple. Design is how intent becomes trustworthy action, on any surface. The surface can be ours, a partner's, or a chat window we'll never see. The job doesn't change.
How we design for omni-interface experiences
If the screen isn't the focus, what is? In practice it's the reasoning: how a capability reads a request, what evidence it brings, how it explains itself, where its limits are, and when it should ask before it acts. We still sketch and prototype. We just do more of it in logic, language, and behavior than in layout.
Two pieces of our work show what that looks like.
Agentic Analytics: the model is the experience
A dashboard gives you a number and wishes you luck. Most of the value sits in the next three questions: why did it change, what's driving it, and what should I do about it?
With agentic analytics, people ask questions in plain language and get "cited, grounded answers with no technical or business intelligence expertise required," with "narrative explanations, root cause diagnosis, and recommended next steps" (Zendesk Help). It also "applies your account's permissions and roles for all data access."
What I like most is the model it creates: signal, then explanation, then recommendation, then action. Each step earns the next. You notice something, you understand it, you get a sensible next move, and you decide whether to make it. That sequence is the designed experience, and it holds up whether the answer lands in a Zendesk view or in a conversation somewhere else. Tableau describes the same shift: "Data is the raw material; knowledge is data with context, definition, and intent."
Copilot: designed to travel
We designed Copilot (our Zendesk AI assistant) to be extensible on any surface. It can proactively engage people with suggestions that are vetted with data, traced, and observable. It can also hand work between AI agents and human agents without dropping the thread. None of that was added on later. It's built into how Copilot works, so wherever that logic and reasoning shows up becomes the canvas for the experience we designed.
You can see the pattern in Zendesk today. Auto assist suggestions "require agent approval before being sent to a customer," and "Actions that are performed automatically are logged in the ticket's events and in the conversation log so that the agent is aware" (Zendesk Help). Propose, approve, record. That's a trust contract, and it shouldn't disappear just because the request came from another host.
That doesn't mean every pixel of in-product Copilot travels. It means the parts that matter most do: evidence, traceability, clear boundaries, and a human who stays in charge when it counts.
Proactive AI needs a dial and a brake
Proactive AI is great right up until it's confidently wrong. So initiative should be a dial, not a switch. Surface a signal, explain it, recommend a next step, or take the action. Where to set the dial depends on confidence, reversibility, permissions, and risk, and customers should decide where it sits. Google Cloud AI's Sheta Chatterjee calls the confirmation step "a deliberate pause and moment of designed friction to ensure that the user is the ultimate authority." (Figma) I'd only add that governance isn't a settings page bolted onto the experience. It's part of the experience.
A bigger canvas for experience design
We'll keep designing beautiful Zendesk interfaces. I'm not hanging up the Figma license just yet. But the measure of our work is no longer only what happens on our screens. It's whether people and agents understand intent, trust the evidence, and act confidently wherever our capabilities show up.
The UI was always a means to an end. The experience is the end, and that's still ours to design.
Ben Mitchell
Head of Product Design at Zendesk
Ben Mitchell is Zendesk's Head of Product Design. He helps customers navigate the rapid shift to AI by transforming complex systems into simple, high-impact tools. Prioritizing low-ego, cross-functional partnerships, Ben focuses on bringing his 2+ decades of experience to delivering intuitive, end-to-end customer experiences that drive real business value.
A new canvas for experience design
Great experiences can happen anywhere, with or without a traditional interface. Discover how Zendesk is bringing intelligent service to more channels, applications, and interactions.
Great experiences can happen anywhere, with or without a traditional interface. Discover how Zendesk is bringing intelligent service to more channels, applications, and interactions.
