I have been exploring the use of artificial-intelligence agents more and more in my work as a Product Designer. And the more I experiment, the more I see that the largest potential of AI is not only in doing things faster, but in changing how we think about and build products.
Today I do not use AI only to generate ideas, screens, or code. I also use it to question decisions, explore possibilities, and see problems from different perspectives.
AI as a second opinion
Before thinking about the solution, I want to be sure I am solving the right problem.
In practice, I work with agents that take different perspectives — product strategy, UX, AI architecture, and software engineering — and I put those views in dialogue to test assumptions and find problems before moving to execution.
The first idea is not always the best one. And having an AI that helps question that idea can be as valuable as having one that helps execute it.
More than trying tools, I have been looking for ways to use AI to improve decisions, explore solutions, and connect the different stages of product development. Three real examples:
1. A framework for thinking before building
One piece of work I developed was APTF (Agentic Product Transformation Framework).
It came from a discomfort: how do you use the potential of AI agents without simply turning any product into a chat-based experience?
Instead of going straight to the interface, APTF considers product strategy, UX, AI architecture, and software engineering from the start. The idea is to question the assumptions, understand what users need, and judge whether each experience should be conversational, visual, or hybrid.
That changes the order of things. Before deciding which agent to create or which tool to use, you need to understand the problem, the capabilities required, and the value you want to create.
For me, this is one of the most interesting uses of AI in design: not only finding answers, but improving the questions we ask.
2. A product with AI from the start
Precivox is an even more direct example of this shift. On it, I was the only human in the process — and artificial intelligence did not arrive at the end as an extra feature. It was there from the conception, shaping the value proposition, the experience, and the architecture of the solution itself.
The product turns market information into better decisions on both sides: the shopper compares their list across neighborhood markets, and small markets gain intelligence to compete — with LLMs as an important part of the experience.
One assumption that fell along the way: comparing only the shelf price seemed enough, but it created a false saving. Distance, time, and promotions change the math. The answer was to design around net savings, not the lowest price.
That also changed how I work as a designer. I had to think not only about screens and flows, but about how the user interacts with an intelligence, how that intelligence responds, which information it can use, and how to turn those responses into useful decisions.
In this kind of product, design stops being only the drawing of the interface. You also have to design the behavior of the system.
3. AI to review interfaces without reinventing the product
The third example comes from how I structured my design and implementation flow with Figma, Cursor, and MCP — the protocol that lets agents access real tools and files.
I developed an approach to guide the AI in analyzing existing interfaces, checking spacing, typography, buttons, icons, images, components, and visual consistency.
But there was an important rule: it was not there to create new screens, invent flows, or redesign everything from scratch. The goal was to compare what already existed with the standards defined in the design system, find the gaps, and guide the fixes using the existing components and variables.
That distinction sounds simple, and it makes a large difference. An AI can generate a visually convincing interface and still introduce inconsistencies, duplicate components, or ignore decisions the team has already made. Generating an interface does not mean building a good product.
In that setup, the agent stops being only an interface generator and starts to support analysis and product quality. And what it points out still comes through me: I question it, review it, test it, and adjust it.
Not everything needs AI
One thing I have been learning in this process is that it does not make sense to put artificial intelligence in everything.
What matters is understanding where it can actually create value, whether that is simplifying an operation, improving an experience, or helping make better-grounded decisions.
It is still a field in constant change, and that is exactly what interests me: experimenting, learning, and finding new ways to build. (I wrote more about this role in What an AI Builder is.)
In the end, I do not see AI as a substitute for design thinking. I see it as a way to extend our ability to explore, question, and turn ideas into real products.
I believe the difference will not be simply knowing how to use AI, but knowing what to ask, what to question, and what to do with the answers.
If you are thinking about where AI makes sense in your product, let's talk.