AI Agents as non-user personas and the evolution of digital interfaces

In recent years, User Experience and digital interface design have undergone constant evolution. Designers of digital products, websites, e-commerce platforms, and apps, following the mantra of designing by listening and putting people and their needs at the center, have gradually understood the importance of expanding and continuously evolving the concept of user personas.

Within the complexity of actors interacting with digital systems, beyond users, there are also what are known as non-user personas.
These personas are valuable for designing digital products in scenarios where interactions do not occur exclusively between humans.

One example of non-user personas is AI Agents, such as Operator.

Operator is a new tool developed by OpenAI that autonomously browses the web, performs complex actions, and automates various online tasks, such as booking flights, searching for and purchasing ingredients for a specific recipe, and filling out forms.
Operator is designed to interact directly with buttons, menus, and text fields on websites, which means that structure and clarity in design will become increasingly important in the near future.

Websites will need to be accessible to both humans and AI agents.

A new frontier is emerging, where designers, developers, and interface architects must consider that websites should be accessible not only to humans but also to AI agents, which some predict will soon become the primary user base.

We are facing a new paradigm that Jakob Nielsen has described as “the first new interaction model in over 60 years.” With the rise and increasing integration of AI tools, users now communicate to computers what they want to achieve without having to specify how to do it. This shifts control, with AI interpreting and executing the user’s intent.

In this evolving scenario, the design of websites, applications, and digital services must be necessarily clear and intuitive, as it will also impact how AI agents interact with the UI.

Will we need non-user testing for AI agents, in addition to user testing to validate designers’ choices?
Will AI evolve to a point where it no longer limits creativity in designing complex digital experiences, or will we have to simplify our designs to ensure they are accessible and usable for AI agents?

We don’t know for sure, but it’s likely that an interesting development in digital accessibility will emerge. These agents, for instance, could evolve the concept of screen readers, offering new possibilities for individuals using assistive technologies to interact more seamlessly with websites and apps.

Currently, there are no established resources or studies to guide the design of websites or apps optimized for AI agents. However, if we think about it, there is a natural correlation with the principles of accessibility and usability promoted by the Norman Nielsen Group, which can enhance both the “user” and “non-user” experience.

In fact, when asking ChatGPT about the main considerations for designing interfaces for AI agents, we notice that it suggests best practices and guidelines that we already know and apply for digital accessibility, such as:

• Using descriptive labels and ARIA attributes to provide context to AI agents, making it easier for them to understand the function of each interface element.
• Ensuring that the Document Object Model (DOM) structure is logical and consistent, allowing agents to correctly interpret the hierarchy and relationships between elements.
• Maintaining a consistent arrangement of interactive elements, such as buttons and forms, to facilitate learning and interaction for AI agents.
• Adopting standardized UI components that AI agents are trained to recognize and interact with effectively.
• Implementing clear status messages that inform the agent of the success or failure of an action, enabling more efficient error handling.
• Requiring confirmation for critical actions to ensure that the agent can verify the intent before proceeding.
• Optimizing performance by ensuring that the interface is responsive and that elements are accessible immediately after loading to prevent interaction errors.
• Implementing measures to ensure that AI agents can interact only with data they are authorized to access, preventing unauthorized actions.
• Properly managing login and authentication processes to allow AI agents to access necessary functionalities without compromising security.

AI agents will become active participants in the ecosystems we design.

In this ever-evolving landscape, which can sometimes be challenging to keep up with, we must be prepared to consider that AI agents will increasingly become active participants in the ecosystems we design. These agents will interact with e-commerce platforms, websites, and apps on our behalf, make decisions, and perform tasks, radically redefining how we deliver products and services.

Will we have to embrace this challenge and tackle the complexity of designing for two types of audiences?
We don’t know for sure, but we can assume that if AI agents interact optimally with our digital products, it will likely lead to an improved user experience for the people who use them.

To conclude, one thing is certain: designing usable and accessible solutions and systems will remain a fundamental priority—first and foremost for people, and consequently for AI agents that (hopefully) will assist them.

Resources:

Principle 3: Design for Users and AI Agents

AI Paradigm

Non-human and non-user personas for life-centred design

Non-human personas in digital design