From UX to AI: How Product Design Is Evolving

For years, user experience design has centered on a single, deceptively simple question: how do we make a product easier to use? We study users, uncover their problems, map their workflows, design the interface, and refine it through testing. That discipline is as valuable today as it has ever been.

But artificial intelligence is changing how we think about products—and, with them, what it means to design one.

In the past, users did most of the work themselves. They searched for information, entered data, reviewed results, and decided what to do next. AI changes that dynamic. It can interpret a request, retrieve information, summarize data, offer recommendations, and—increasingly—carry out actions on the user's behalf.

This raises a question every product designer now has to confront:

Are we still only designing interfaces, or are we now designing how people and AI work together?

In my view, we are doing both.

AI is more than a chatbot

When people picture AI in a product, they usually imagine a chatbot. But conversation is only one of the many ways AI can show up.

Consider an ERP system used by a finance team. Today, a user might open a module, apply filters, search through records, cross-check customer details, and manually flag overdue invoices. With AI, that same user could simply ask the system to show the high-value overdue invoices and explain which ones need attention first. The system retrieves the relevant records, analyzes them, and surfaces what matters most. The user is no longer navigating the application—they are working with it. That is a meaningful shift in the meaning of user experience.

What happens behind the screen

My engineering background leads me to look at AI products from both sides: the user's and the system's. As a product designer, I'm not satisfied with the final interface alone; I want to understand what happens behind it.

A simple AI feature might involve the interface, an API, an AI service, enterprise data, and a response. A more advanced enterprise solution can involve authentication, multiple APIs, an orchestration layer, retrieval from company data, a large language model, tool or function calling, and several layers of validation and business rules.

Understanding this helps designers ask sharper questions. Where is the AI getting its information, and is that information current? Does it have access to sensitive company data? What happens when it can't find the right answer? Can it change something in the system, and if so, does the user need to approve that change? And what happens when the AI gets things wrong? These are not only technical questions—they are UX and product questions too.

RAG makes enterprise AI more useful

One concept worth understanding is Retrieval-Augmented Generation, or RAG.

A general-purpose AI model doesn't automatically know what's happening inside a specific company. An enterprise application, by contrast, may hold thousands of customer records, invoices, purchase orders, policies, and internal documents. RAG lets the system retrieve the relevant pieces and supply them to the model as context before it generates a response.

The difference shows in the quality of the answer. Instead of vaguely stating that an invoice "may be a duplicate," the product can offer supporting evidence—the vendor, the amount, the invoice number, and the matching transaction—which makes the response far easier to understand and verify. Good AI experiences don't just deliver an answer; when the decision matters, they help the user understand why that answer was given.

When AI starts taking action

Things become more interesting when AI can use APIs or tools to act on the user's behalf.

Imagine a user asks the system to find overdue invoices above ₹5 lakh and prepare reminder emails. The AI can locate the invoices, check the customer details, and draft the messages. But should it send them automatically? In most cases, no.

A better experience keeps the user in the loop. Rather than acting silently, the product reports back: eight overdue invoices identified, eight reminder emails drafted, ready for review before sending. The AI has done the heavy lifting, but the user stays in control. That balance—showing what the AI has done, what it intends to do, and what still needs approval—is where thoughtful UX becomes essential.

AI can be wrong

AI carries a challenge that traditional software does not: it doesn't always get the answer right. It can misread a request, work from incomplete data, or produce something plausible but incorrect.

For that reason, we have to design for failure as much as for success. Instead of confidently declaring that "the customer is eligible," the product may need to say that it couldn't verify the latest financial information and ask the user to review the customer's details before continuing. It's a small design decision, but it makes the product more trustworthy. When the AI is uncertain, the experience should make it easy for the user to step in.

Human control matters

I don't believe the goal of AI is to do everything on its own. The better goal is to let AI absorb unnecessary complexity while keeping people involved wherever their judgment counts.

AI can flag a possible duplicate invoice, prepare an adjustment, or draft a customer message. But a financial transaction, a compliance decision, or a sensitive customer action may still call for human approval. A useful principle is that the level of human involvement should scale with the impact of the action: the higher the stakes, the more important human control becomes.

A new lens for the product owner

AI makes it remarkably easy to generate new product ideas—which creates a risk of its own. Teams may start adding AI features simply because they can.

The better question for a product owner isn't "Where can we add AI?" but "What problem are we solving?" Adding an AI assistant to a dashboard might sound appealing, but if users are spending thirty minutes identifying important exceptions, the sharper question is whether AI can help them find those exceptions faster. That framing starts with user value. AI is the capability; user value is the goal.

What changes for the team

AI makes close collaboration between product, design, and engineering more important than ever, because each discipline brings something the others need. Product understands the problem, the business value, and the priority. UX understands the workflow, the interaction, and the question of trust. Engineering understands architecture, APIs, security, and performance. AI and data teams understand the model, retrieval, evaluation, and its limits. QA accounts for the fact that AI responses may not be identical every time. And users, ultimately, tell us whether the solution improves their work in practice.

This is where an agile approach helps. We can start with a smaller problem, build an initial version, test it with users, learn from the results, and improve—without having to solve the entire AI problem in a single release.

AI also changes testing

Traditional UI testing is largely predictable: click a button, expect a specific result. AI products don't behave that way, so testing them means asking a different set of questions. Is the answer relevant, and is the information correct? Did the AI use the right data? How does it handle an unexpected question, and how does it behave when it's uncertain? Did it call the correct API and respect the user's permissions? And can the user recover from a wrong action? In other words, AI quality isn't only about the model—it's about the entire product experience.

Why designers should understand the technology

Not every product designer needs to become an AI engineer. But those working on AI products should understand the core concepts: large language models, RAG, APIs, embeddings, tool calling, AI agents, context, permissions, guardrails, latency, evaluation, and observability. We don't need to build these systems ourselves; understanding them simply lets us have better conversations with engineering and AI teams. Instead of asking, "Can the AI do this?", we can ask whether the AI can reach a particular API, what permissions it needs, what happens if that call fails, and where the user should approve the action. That is a far stronger product conversation.

This reflects something my work across UX, UI, and engineering has taught me: A good interface cannot solve every product problem. The experience is shaped just as much by data, architecture, performance, validation, permissions, and error handling—and AI adds another layer on top, from models and retrieval to agents, tool calling, and guardrails. Understanding the technology doesn't turn a designer into a developer. It helps us design experiences that are both useful and technically realistic.

The role is getting broader

I see the evolution this way. A UX designer focuses on designing experiences for people. A product designer focuses on solving user and business problems through products. An AI product designer focuses on using AI to improve those experiences and outcomes. The next step is designing how people and AI work together. This is not about replacing UX with AI; it is about expanding what we consider a product experience.

The future of product design

I don't think the future is AI versus UX. It's UX, product, engineering, and AI working together.

The best AI products won't be the ones with the most AI features. They'll be the ones where AI removes unnecessary complexity, helps people make better decisions, and lets them do meaningful work more easily. The user may never think about the technology at all—they'll simply notice that the product makes their work easier.

Perhaps that is the real goal of AI in product design. Don't add AI because a product is supposed to have it. Use AI where it can remove complexity, improve decisions, and create real value for people. That is where UX, product, engineering, and agile come together.

--

Scrum Alliance's AI courses prepare you to navigate new technology with confidence so you can drive results no matter your role or industry. Explore the courses to find one that fits your schedule. Microcredential courses take just several hours to complete, and you'll gain new skills, knowledge, and a digital badge to demonstrate your expertise.

Explore AI courses