When agents are the users: Rethinking interfaces beyond screens

At an Amazon fulfillment center, a robot named Sparrow selects an item from a bin and hands it off without a human in sight. Nearby, a fleet of 750,000 autonomous mobile robots quietly weave through aisles, scanning QR codes, lifting shelves, and routing packages.

There’s no dashboard. No button to push. No screen at all.

These machines aren’t responding to human clicks or keystrokes. They’re acting on structured data, machine-readable protocols, and direct system calls. What we’re witnessing is a different kind of user experience — one where agents, not people, are the primary users.

And it’s not just Amazon. Across industries, AI agents are taking the wheel: reading databases, making decisions, and triggering actions without human input. The interface still matters, but now it’s invisible, buried in schemas and API payloads.

This is the rise of agentic UX. And it’s forcing us to rethink what experience design even means when the user is no longer human.

This is the foundation of agentic UX: designing for systems where AI agents interpret, decide, and act with minimal human input.

It marks a shift, where the interface is no longer built for people, but for AI agents that interpret data, make decisions, and take action independently.

The disappearing interface

Traditionally, UX design has focused on people. We build products to be accessible, intuitive, and human-centered. But what happens when the user is no longer a person?

In agent-based systems, interfaces don’t vanish. They evolve. The screen fades into the background, replaced by APIs, event streams, and machine-readable formats. Agents don’t need pixel-perfect designs or microcopy that nudges. They need clarity, consistency, and logic.

Take UiPath Autopilot, announced in October 2023. The platform integrates generative AI directly into automation workflows, allowing users to describe tasks in natural language. AI agents then translate those descriptions into working automation sequences, executing across applications without step-by-step human instruction. UiPath frames this as “AI at work,” co-piloting every stage of enterprise automation. It’s a clear signal that the interface layer is shifting: from screens and clicks to prompts, APIs, and agent logic.

This shift creates a profound challenge for UX practitioners. We’ve built careers around visual systems. But in a world of agent-to-agent interactions, those systems must now be legible to machines, without sacrificing human oversight.

Designing for the machine’s mind

Agents don’t “see” buttons or hear sounds. They process JSON, parse XML, and follow event schemas. That’s the interface.

And that’s where invisible UX comes in.

In agentic UX, the priority shifts from aesthetics to architecture: schemas, APIs, and logical flows become the interface.

At Klarna, the AI assistant now handles two-thirds of customer service chats without human escalation. Behind the scenes, that assistant pulls data from dozens of internal systems, all structured for machine legibility. Every field, every tag, every timestamp is a potential interaction point — not for a person, but for the AI agent handling the query. These are the new touchpoints in an agentic UX environment

For designers, this means shifting attention from the visible layer to the semantic architecture of systems. Label consistency, data integrity, and API predictability are now usability concerns. A mislabeled database field might not confuse a human, but it can halt an AI mid-task.

That’s not theoretical. In early 2024, a joint impact study by Cognizant and Oxford Economics estimated that generative AI could inject up to $1 trillion into the U.S. economy over the next decade, affecting as many as 90 percent of jobs. The report emphasized that as AI systems become deeply embedded across workflows, the challenge shifts from technical integration to ensuring systems work in context. Machine-to-machine coordination, once invisible, is now central to business performance and risk.

When humans still matter

This doesn’t mean the human disappears from the equation. On the contrary, our role becomes more critical when agents are in charge.

Imagine a system where AI agents trigger actions automatically: scheduling meetings, provisioning software, placing orders. Now imagine something goes wrong. Who understands what the agent saw, decided, or did?

Without visibility, trust collapses. And that’s where interface design must evolve — not for direct interaction, but for explainability and control.

One of the biggest challenges in agentic UX is creating visibility into agent decisions, without reverting to human-first interfaces.

Google’s PAIR (People + AI Research) initiative has focused on building AI systems that prioritize interpretability and transparency, offering users insight into how decisions are made rather than hiding them behind black-box logic. Their tools, such as the What-If Tool and TCAV, are designed to help both developers and non-technical users better understand model behavior. That principle — making AI reasoning visible — is becoming central to agentic UX. Even if agents don’t need screens, humans still need lenses into what’s happening.

Invisible, not uncontrolled

The concept of invisible UX isn’t new. In 2011, Marc Andreessen famously said that ‘software is eating the world.’ Today, it’s doing so silently, behind the scenes, and often without human eyes.

Voice assistants were the first glimpse. You say “Play Taylor Swift,” and a chain of invisible events occurs: speech recognition, intent matching, API calls, authentication, streaming initiation. All in under a second.

But now, those invisible chains are extending across systems — into HR, finance, healthcare, and logistics. Agents act not as assistants, but as autonomous workers. And the UX job becomes one of boundary setting.

Take healthcare. At the Mayo Clinic, natural language processing agents assist radiologists by extracting insights from reports and highlighting relevant findings. These insights are presented through user-friendly interfaces — simple for the user, complex underneath.

The UX challenge wasn’t to beautify the panel. It was to ensure the agent made accurate decisions, that clinicians could verify its logic, and that overrides were easy when needed.

That’s agentic UX in action: designing systems for control, comprehension, and contingency, even when no interface is seen.

A new skillset for designers

As systems evolve, so must we.

Invisible UX requires fluency in structured data, backend schemas, and API behavior. It rewards designers who can collaborate with architects and think in flows, not just screens.

It also demands a shift in mental models. We’re not just guiding humans anymore. We’re enabling agents to act correctly, ethically, and transparently.

Some companies are responding. At Atlassian, the Confluence team integrates AI technologies to assist users by extracting insights from content, summarizing pages, suggesting action items, and surfacing relevant findings in real time. These capabilities are seamlessly embedded into the workflow through intuitive, user-friendly interfaces, designed to appear simple on the surface but powered by sophisticated models and backend logic underneath.

It’s early days, but the trend is clear: experience design now includes systems without faces.

Designers who understand backend logic and system behavior are increasingly valuable in an agentic UX world.

Where we go from here

The rise of agents doesn’t eliminate the interface. It redistributes it.

For agents, the interface is logic and data structure. For humans, it’s visibility and control. Our job as designers is to mediate between the two—ensuring that what’s hidden isn’t also unchecked.

This is the heart of AX: Agent-based experience design. It’s not about screens. It’s about systems. And it calls on us to rethink everything we know about users because, increasingly, they won’t be people at all.