Digital Product Talks #38: Why Product Discovery Matters Even More With AI

We’ve recently covered how agentic development has changed the way our own developers work, allowing them to move faster on certain kinds of work, while giving them more time to focus on parts of the product that actually need human creativity, such as architecture, UX, or product decisions.

But when the way things are done changes, that means you also confront new challenges along the way. With AI, one of those new challenges is building the wrong thing. The faster you build, the faster you can go in the wrong direction, which can get very expensive very fast.

That was the focus of my discussion with Wolf Brüning, Lead User Experience Designer B2B at the German retailer OTTO, on the latest episode of Digital Product Talks.

Wolf initiated OTTO’s design systems, and its three product teams eventually grew to around a hundred. He also runs an internal product discovery school that's put 600 to 700 people through a full discovery cycle over the last decade, so when he talks about what happens when organisations build the wrong thing, he really knows what he’s talking about.

The feature factory doesn't slow down for AI. It speeds up.

One problem a lot of product teams slide into is piling features upon features on the product – the so-called feature factory – a list of features where the job becomes adding green checkmarks onto a list.

It comes from an older, project-shaped way of working where a client wrote a long spec and the team worked through it in time, budget and scope. That world is comfortable for managers, because a checklist and a nice-growing percentage are easy to visualise and explain.

In the product world, however, you don't always know the exact path, so it's not strange that teams keep going back to that way of working. Uncertainty is uncomfortable, and solid structure is easier to hold onto, even when it doesn't feel right.

Now add AI into the mix. The instinct in a lot of organisations is: fantastic, now we can build even more features! But if you build ten times faster while being pointed the wrong way, you don't save time nor money, you go so far in the other direction that going back is ten times more expensive.

Product discovery is risk management, not a design luxury

So Wolf has a crazy idea here: before you build, you work out what the right thing to build is.

To drive the point home, Wolf used a racing analogy I liked. Say your goal is to win a race. Your team starts immediately building a Formula 1 car, only to discover the race is on a dirt track. So, you’ve got a great product, but it’s completely off the track (heh) for the intended purpose.

So, product discovery isn't a fancy design ritual, or as Wolf put it, a "Schischi" design process (something with no substance, for our readers outside Germany). It's risk management, because building and running real software is expensive, and getting it wrong costs you more than money. You can drive away your customers, or spend years building the wrong thing because you stopped checking whether it’s actually solving anyone's problem.

"Fall in love with the problem, not the solution." Wolf Brüning, Lead UX Designer B2B at OTTO

We brought up two examples during the discussion that perfectly illustrate the point.

The first is Sonos. I bring this one up a lot when talking about product quality. In May 2024, right before launching a new pair of headphones, Sonos released a fully redesigned app, which, in theory, should have been nice… except the redesign stripped out features people used every day and broke setups that had worked for years.

By January 2025, the failed launch had wiped close to $500 million from Sonos's market value, the CEO published a public apology about 11 weeks later, and product launches were pushed back. For a company whose entire promise is great experience, that's… quite the blunder. How do you even rebuild your reputation from that kind of damage?

Now, the second example… in April 2025, Duolingo announced it was going "AI-first" and would lean on AI over the contractors who built its content. Many users took it as a betrayal of the very human thing they'd signed up for. People deleted the app and ended streaks they'd kept for hundreds of days. One user even shared a video of themselves giving up a 1,500-day streak, and it got millions of views. (If only we all took language learning that seriously!)

Interestingly, Duolingo's revenue and subscriber numbers went up that quarter. So was this a failure? Financially, things were fine for Duolingo, but their brand, which spent years earning affection, burned a chunk of it in a few weeks.

Maybe the damage won’t show up immediately, which makes it easy to wave it away, but there may be no coming back from the long-term damage. It's exactly the kind of thing a bit of user research would have told you in advance. As Wolf said, after 15 years of user tests, he's still surprised how often his gut is wrong.

AI in a silo won't make your team faster

Wolf wants product teams using AI to stop doing it in a silo. Designers go off and work out how AI makes them faster in Figma, while the developers separately work out how AI helps them code faster. That’s all valuable knowledge, but the real gains sit in the spaces between the roles.

Design systems are a good example of this idea – but only if you treat them as a product with real ownership and dedicated people, rather than a project you finish once. 

"A solid design system is like a vocabulary book for the AI." Wolf Brüning, Lead UX Designer B2B at OTTO

Left to itself, AI writes middling code and reinvents every button. Give it a design system to work from, and it has fewer ways to go wrong and a much better shot at producing something usable.

To go back to the premise of the article, and whether AI has made teams faster – the honest answer is that roles are changing. Designers, concept people, product people are moving away from drawings in Figma toward being curators and advisors.

And the products that stand out will be the ones that have people like that to point them in the right direction first.

Not sure whether what you're about to build is what anyone wants? That's the conversation we like having before a line of code exists.

Want to hear the full conversation with Wolf Brüning? Listen to the episode on the Digital Product Talks podcast on Spotify or Apple Podcasts.

‍

TL;DR

In a rush? Worry not, we've prepared a quick summary for you.

What is product discovery? Working out what the right thing to build is before you build it. In other words, it’s risk management.

Does AI replace product discovery? No. AI makes building faster and cheaper, which raises the cost of building in the wrong direction.

What is a feature factory? A way of working where the job becomes simply fulfilling the list of tasks.

How do design systems help teams working with AI? Left to itself, AI writes middling code and reinvents every button. A design system gives it a vocabulary to work from, so it has fewer ways to go wrong.