On the 23 or April, we opened our doors to the Friends of Figma Zürich community with the event “AI in action: smarter workflows, better products”. More than 50 designers, product managers, engineers, researchers, and UX writers joined us to explore one question:
What actually changes when AI becomes part of how we build products?
The goal was simple: move beyond the hype and look at practical examples of what AI actually changes in our daily work and its impact. At SMG Swiss Marketplace Group, we don’t wait for the future of AI: we build it, test it, and learn from it.
The event was organized by our Design Guild: a community of practice that connects people across our verticals to share knowledge, challenge ideas, and grow together. We combined short talks from colleagues across real estate and automotive, followed by roundtable discussions with lively debates about the opportunities and responsibilities AI brings and an apero.
Across all talks, one theme came back again and again: AI is not just speeding up our work, it is reshaping it, and expanding or shifting our roles. Let’s dive into some of the key learnings of our talks…
AI can simulate users to anticipate scenarios for better product discovery decisions
Magda Ursulean, Product Manager in Real Estate, opened the evening with a practical challenge:
“Understanding users has become a reading problem, not a talking problem.”
User insights are everywhere—forums, reviews, social media, insights databases and research tools—but time is limited.
Her approach includes using AI to simulate users as a first draft for discovery work. What if we can challenge our hypothesis before presenting a feature to real users? Or use user personas profiles in a custom GPT enriched with our real research data or customer support tickets to better select target users for a product testing or craft better research questions for interviews and surveys?
By crafting rich personas and giving them a realistic voice, teams can:
- iterate ideas early to better target solutions to different segments
- challenge assumptions with deeper user-centered thinking
- prepare better questions to increase the impact of research
One of her examples stood out:
A simulated persona reacting to a productivity app revealed a core product tension in just one sentence: “Anything that smells like therapy… I’m out.”
But Magda was clear about the goal:
“Think of AI as a sparring partner, not an oracle replacing real research.”
AI can accelerate thinking and anticipate scenarios, but the source of real users’ insights and its judgment remains human.
AI speeds up building—not deciding
Sherif Saleh, Designer in Automotive, took us into the world of AI-powered prototyping.
The challenge each designer always tried to solve when exposing complex features in user testing is: how to make user interactions feel natural. Now that AI-powered products are becoming more conversational and allow more and more the natural language and behavior of users, static prototypes are not enough anymore.
Take for example a search on AutoScout24, a user instead of filling manually all filters, could type: “I want a red Ferrari under 500k”. How to simulate such a natural interaction?
To test our new natural language car search, for example, our design team moved beyond static Figma prototypes and built a real, dynamic prototype using vibe coding in Cursor, leveraging our Figma Design system and a Github code kit.
The impact was immediate:
- up to 5–40x faster prototyping (depending on complexity)
- significantly more realistic interactions with real car data
- much higher quality user feedback
One of the most interesting learning in designing AI-powered products “the new way”, isn’t only about the speed:
“The hidden work is not mainly the UI anymore—it’s the product logic: working closely with engineers and product managers on data wiring, edge cases, system behavior. This is where complexity lives.”
While AI helps unblock design teams on the user interaction efficiency, the real value still lies in deciding what to build and how it should behave.
Make it exist first. You can make it good later.
Pablo (from Real Estate) shared a bold experiment: launching an AI-powered Listing Co-Pilot for B2B real estate customers.
The problem was clear: users didn’t understand why their listings weren’t performing. The traditional approach? Feeding them with a lot of metrics distributed on dashboards and let them figure out what to do. However, not all customers are data-savvy and some struggle to read those data to inform their decisions.
The new approach: Pablo’s team wanted to leverage AI to make those data understandable in plain language.
Instead of long planning cycles, the team did something different:
- built an AI-powered MVP in a few days
- tested it live to identify the product-market fit
- learned from real users feedback
- decided how to iterate further
“We didn’t wait for alignment—we created something people could actually use in order to learn fast.”
The result: ~20% of customers using the Insights Hub engaged with the feature in the first weeks.
This wasn’t about perfection. It was about learning faster by shipping earlier. As highlighted in the Q&A by Pablo, this approach works especially for smaller features that are not compromising customer trust and relationship and are not bound to high business impact. A/B testing or feature flag releases of AI generated solutions are the best go-to-market approach to learn.
Agentic workflows are changing our ways of working, but is it everything worth being automated?
Théodore Wanner, Product Manager in Automotive, brought a more operational perspective on workflow automation, leaving the audience with practical principles in applying AI to automate your tasks.
AI can automate a lot: spot patterns in research and generate summaries, content creation or organisation, repetitive workflows (at work and in private life). It’s becoming an agent who can now act upon a trigger (manual, scheduled or event-driven), execute an action (read, decide, generate..), and provide an output (an artifact or a state change in a system) entirely without human intervention.
But how to decide if something is worth automating? The key criteria he highlighted are:
- Repetition: Same gesture, same format, same source — week after week.
- Time-consuming tasks: costly in time, attention, or money each time it happens.
- Predictable: the logic could be written down and followed by anyone.
Are there cases where the costs and risks outweigh the gain? Yes, there should be clear boundaries! Surely one-offs tasks are not subject to automation, wait until you have a real need to repeat a task. Avoid high-stakes decisions, where following a wrong call can become expensive, irreversible, hard to detect or do harm. If the answer you expect depends a lot on the context information you hold, human judgment, or the service or activity that can benefit from the human touch, then it is not worth replacing it with AI.
“Automation has a cost. Not everything is worth automating.”
His own life-hack example?
An automated “audio deep dive” that turns newsletters or book summaries into a personalized podcast for his commute. Powerful. Practical. But still guided by the human intent to learn and be informed about the real world.
Your role in the UX field is not disappearing, it’s moving
In the final talk Marta Andreoni, Head of Design at Automotive, zoomed out into the changes that this new AI dominated Era brings in UX roles and skills.
“UX is dead”, “design is replaced”, “research is obsolete” claims many influencers online. There is a lot of noise and hype around AI, and at first glance it seems that many things in UX should be dead-by-AI. This new paradigm is surely challenging the status quo and as UX experts we need to step up our game, but the reality is more nuanced.
“AI doesn’t replace roles—it collapses boundaries between them and shifts the focus elsewhere.”
What we see in our daily life as designers, as emerged from a recent internal poll, is:
- less time spent on execution in the UI and UX flows design (-20%)
- more time spent on strategy, research, conceptual work and problem framing (+25%)
- more collaboration across disciplines (with ML engineers, PMs, security and legal experts)
In the new paradigm AI augments our ability to analyze large datasets, identify patterns faster, and generate multiple concepts, assists us in research tasks, in the drafting of user flows and early wireframes, or in shaping our thoughts and documenting them. Where it can replace us, is mostly in those tedious low-value time-consuming tasks: like transcribing interviews, tagging insights, identifying missing UI states or inconsistencies in design files or copy, so that we free up time for more valuable work.
AI-powered solutions are also changing how people interact with digital products, services and brands. Products become more conversational and multi-modal in the interaction (voice, text, touch, movement…), customized based on behaviors, able to take on a chain of tasks for the users to help them reach the wished outcome. This is where our shift in roles becomes quite clear, we move from pure creators, to orchestrators and designers of AI behaviors and human-AI ecosystems that need to be aligned to the user mental model and stay clear, explainable and secure.
The key take-away?
“While AI compresses execution, it expands thinking, responsibilities and skillset. We’re moving from doing the work…to defining what should be done and how systems behave.”
What AI means for product and UX people at SMG?
At SMG, we approach AI with a clear mindset. We use it to:
- solve real problems
- support better decisions making
- reduce complexity for our users
We don’t aim to automate everything, we aim to build meaningful, trustworthy experiences. For us AI is not just a technical shift, it’s also a product, design, and responsibility change we want to embrace with openness, curiosity and continuous learning.
Author
Marta Andreoni
Head of Design
Automotive