From UX to AX to AIX: Designing for People, AI Agents and AI Products

Published by Agile Design School · 11 November 2026 · 11 min read

The short answer

UX asks: can a person understand and use this easily? AX asks: can an AI agent understand and use this correctly, on a person's behalf? AIX asks: when a person uses AI in this product, do they understand it, trust it the right amount and stay in control?

The terms are new, and people use them in slightly different ways. AX has a clear origin. AIX is less settled: you will also see it called AI UX, human-AI interaction design or AI product design. The ideas matter more than the labels.

UX: designing for people

User experience design is the work of understanding people and shaping products around their needs, habits and limits. A UX designer researches who the users are, maps what they are trying to do, designs and tests the flows and screens, and keeps improving them with evidence.

None of that goes away. Every agent works for a person, and every AI feature is used by a person. If you do not know what your users need, an agent cannot get it for them and an AI feature cannot help them. That is why UX is the foundation of the other two. New to the field? Start with what UX design is.

AX: designing for AI agents

An AI agent is software that can take actions for a person: search, compare, fill in forms, book, buy or report back. When someone asks an AI assistant to "find a weekend UI/UX course in Hyderabad and book a demo class", the assistant becomes a user of the school's website.

Mathias Biilmann, CEO and co-founder of Netlify, introduced the term agent experience in January 2025 and described it as the experience AI agents have as users of a product or platform. His point was simple: agents experience software very differently from people. A step a person spots at a glance can be invisible to an agent if it is not clearly exposed.

What agents need from a website or service:

• Clear structure. Real headings, lists, tables and labels, so meaning does not depend on the visual design.
• Facts in plain text. Prices, dates, eligibility and contact details written out, not hidden in images, carousels or PDFs.
• Predictable actions. Forms and checkouts with clear steps, labelled fields and no surprise pop-ups.
• Errors that explain themselves. "The phone number needs 10 digits" helps an agent recover; "Something went wrong" does not.
• Machine-readable entry points. Structured data, a summary file such as llms.txt, an API, or an MCP server. The Model Context Protocol (MCP), introduced by Anthropic in November 2024 and since donated to the Linux Foundation's Agentic AI Foundation, is an open standard for connecting AI applications to outside tools and data.
• Safe limits. Clear permissions, confirmation for anything costly, and a record of what the agent did.

Much of good AX is good accessibility and good content design. A page that works with a screen reader and reads clearly as plain text is already most of the way there.

AIX: designing how people work with AI

AIX is about the person on the other side of the AI. A chatbot, a writing assistant, a copilot inside a banking app or an AI that summarises medical reports all need design that helps people understand what the AI is doing and decide when to rely on it.

This is not brand-new. In 2019, Microsoft researchers published 18 Guidelines for Human-AI Interaction at the CHI conference, built from more than 150 design recommendations and tested with 49 design practitioners on 20 AI products. They are grouped by when they apply: when people first meet the system, during normal use, when the system is wrong, and over time. Google's People + AI Guidebook covers similar ground.

The patterns that matter most in AIX:

• Set expectations. Say what the AI can do, and how well, before people rely on it.
• Show your working. Show sources, reasons or confidence where a wrong answer would hurt.
• Keep people in control. Let them edit, undo, regenerate or switch the AI off.
• Ask before acting. Get approval before the AI spends money, sends a message or deletes anything.
• Design for being wrong. Plan what happens when the AI misunderstands, and make correcting it easy.
• Learn from feedback. Give people a way to flag bad answers, and use what they tell you.
• Respect privacy. Make clear what data the AI uses and what it remembers.

The goal is not maximum trust. It is the right amount of trust: people should rely on the AI where it is good and check it where it is not.

UX vs AX vs AIX, side by side

One task, three layers: an example

Take one task: booking a doctor's appointment.

• UX: a patient opens the clinic's app, picks a doctor, sees open slots and books. The designer makes this quick and clear, including for older patients.
• AX: the patient asks an AI assistant to "book me with a skin specialist near Ameerpet on Saturday morning". The assistant has to find doctors, read slots and complete the booking. If slots only appear in an image, or the form has unlabelled fields, the agent fails or books the wrong thing.
• AIX: the assistant comes back with two options and asks the patient to choose, shows the fee and the cancellation policy, and books only after the patient confirms. If it is not sure which clinic was meant, it asks instead of guessing.

The same product needs all three layers to work. That is why designers who understand all three are becoming more valuable.

Skills designers need now

• User research. Still the core skill. You cannot design for an agent or an AI feature without knowing what the person behind it needs.
• Information architecture and content design. Structure and clear text are what agents read and what AI features produce.
• Writing. Microcopy, explanations, error messages and the instructions that shape how an AI feature behaves.
• Designing for uncertainty. AI output varies. Designers plan for wrong, partial and slow answers, not just the happy path.
• Systems thinking. A working understanding of APIs, data and how models behave, so you can talk with engineers about what is possible.
• Accessibility. Accessible design and agent-friendly design overlap a lot.
• Ethics and trust. Consent, privacy, bias and when not to automate.
• Prototyping with AI tools. Tools such as Figma Make or Framer let you test AI interactions quickly.

Our guide to the AI tools UX/UI designers use covers the tools side in more detail.

What this means for design careers

AX and AIX are not separate job titles yet at most companies. In India, most openings still say product designer, UX designer or UI/UX designer, and more of them ask for experience designing AI features. Some companies hire for titles such as AI product designer or conversation designer.

The practical takeaway: these are skills to add to a UX foundation, not a different career. A portfolio with one strong AI case study, covering expectations, errors, approval and testing with real users, will stand out. Read more in Will AI replace UX designers?

How to start learning

1. Test a website with an agent. Ask an AI assistant to complete a real task on a site you know, such as finding a price or booking a slot. Note every point where it gets stuck or wrong.
2. Audit an AI feature. Pick an AI feature you use and check it against the AIX patterns above. Where does it set expectations? What happens when it is wrong?
3. Redesign one failure. Design the error, correction and approval states for that feature, then test them with three people.
4. Write it up as a case study. Show the problem, your research, the design decisions and what you learned.

Frequently asked questions

Sources

• Netlify: Agent Experience (AX), Mathias Biilmann's introduction of the term, January 2025
• Microsoft Research: Guidelines for Human-AI Interaction (Amershi et al., CHI 2019)
• Google PAIR: People + AI Guidebook
• Anthropic: Donating the Model Context Protocol and establishing the Agentic AI Foundation
• llmstxt.org: the llms.txt proposal