The Product Market Fit Treadmill: A Conversation with Wiktor Sobolak

Wiktor Sobolak is a product leader who has built and scaled products from zero to one and beyond, across early-stage startups and scale-ups including companies with over a hundred million users. He has worked at DeepL and at Text (formerly LiveChat), and has been deeply involved in shaping product growth and AI-native teams. He is also a co-organizer of Growth Meetup, and the only guest to come back for a second time on this show, which happens to be the event that inspired it.

What makes Wiktor worth listening to is that he has built products, scaled them, and fixed them when they break, and big products break a lot. This conversation challenges some of the received wisdom of the show itself, starting with "distribution beats product plus brand," then moves through the product market fit treadmill, what counts as a moat now that AI has erased the old ones, and what an AI-native product team actually looks like in practice.

Distribution beats product needs nuance

Wiktor came to challenge the popular line that distribution beats product plus brand, and his objection is that it lacks nuance. He agrees brand is becoming less effective as every new AI disruptor arrives, and he agrees distribution used to beat product. But two things complicate it. First, product-channel fit: in product-led growth and sales, the way you build the product, the friction to enter it, the business model, all enable or foreclose specific distribution channels. A fifty-dollar product cannot afford a big ABM campaign, because the CAC would kill it. And channels themselves are decaying, SEO is less effective, outbound is easier to scale and therefore less effective, so the product has to carry more weight.

Second, and this is the heart of it, we are on what Elena Verna calls the product market fit treadmill. PMF is a moving target now. AI lets competitors ship and disrupt faster, and users' expectations have risen so quickly that yesterday's innovation is today's baseline and tomorrow's weak differentiator. So "mediocre product with strong distribution beats good product with poor distribution" is no longer true. The product has to lead to stay relevant, because distributing an irrelevant product does not work, even in the midterm.

PMF is a moving target

Why can you lose product market fit so fast now? Wiktor gives two vivid examples. At DeepL, a translator with a hundred million users, the moat was a genuinely 10x-better experience versus Google Translate, which drove huge word of mouth. When ChatGPT arrived and LLM translation got good enough, DeepL stayed superior but the gap shrank from 10x to maybe 20 to 50 percent better, and that made the word-of-mouth channel far less effective. The company recognized it early and moved smartly toward enterprise and a broader portfolio.

The second example is sharper. At a company dealing with public tenders, he listened to a sales call with a mid-size construction firm run by two men in their fifties, exactly the sort you would not expect to innovate, who showed up saying they had built their own agents and skills to handle tenders. The customer had become the competitor. That, he says, is how fast and from where you can now lose product market fit, and it ties back to AI leveling the playing field: people learn so fast they can become your competitor without realizing it.

What is actually a moat now

Wiktor is precise about what is no longer a moat. A huge product portfolio is easy to replicate. Switching costs have collapsed as LLMs bring migration cost close to zero, from copying your ChatGPT memory to Anthropic showing it could migrate legacy COBOL that IBM had built a whole consulting business around. And brand, in his contrarian view, helps you stand out in a red ocean but will not hold a user for long, because convenience wins over sentiment.

So what is defensible? Speed can be a moat, speed of execution, of shipping, of expanding channels; he cites a company doing more than 70 releases in around 50 days, so fast that competitors cannot even keep up mentally. Then two connected things: process power, the sophisticated process that delivers the last missing 10 percent a quick hackathon cannot; and unique domain knowledge, especially knowledge not on the internet, which is why people who have worked an industry for 15 or 20 years carry a huge unfair advantage in their heads. Add unique data and its connections, and economies of scale, and you have the real list. But he would bet mostly on unique data and unique knowledge as the most lasting, while warning none of it is permanently safe.

What AI-native product teams actually do

"AI-native product team" sounds great and means nothing until you see it, so Wiktor describes what he learned working with a genuinely AI-native younger team. It starts with thinking AI-first: whatever you are doing, you reach for AI first, using multiple agents for planning, execution, and QA. His own lightbulb moment came when he tried to build a tool to teach product engineers to assess onboarding flows, and the technical founder simply asked, why not use the AI itself as the judge to assess everything and just give them the output? He realized he was still trying to put AI in a box to build things, instead of letting its non-deterministic output show its magic.

Two more patterns define it. Roles fade into each other: designers and PMs open pull requests, his last designer ditched Figma to design and code in the same tool, product engineers talk to customers. AI empowers everyone to go beyond their role. And you build processes with AI at the center, not bolted on, to the point where around 30 percent of customer tickets were fully solved by AI. He is also blunt that classic roadmaps longer than three to six months are becoming a waste of time, because the ecosystem changes too fast for a year-ahead roadmap to survive; plan strategy and milestones, not the whole roadmap. Where he was wrong three years ago: believing SQL and heavy documentation were durable PM advantages. Now Claude does both, and he lets AI handle 70 to 80 percent of the documentation.

Everyone can be a builder

Wiktor is refreshingly honest that he feels himself becoming irrelevant at least once a week, and that the honest answer to what product teams look like in two to three years is that he has no idea. What he does believe is that everyone can now be a builder, though not everyone should be, and that "product builder" may become a new role combining product, engineering, and design. His advice to anyone entering product is to go out and build something, because shipping and talking to customers is how you develop the product sense that is the strongest moat you can own.

His accusations cut both ways. Product people should be growth and business people too, because shipping a feature is not the end; if adoption is poor and you ignore monetization and user flow, you are releasing cute-looking work that brings limited value. And growth people are too often incentivized to close deals fast rather than build sustainable pipeline, or they over-optimize tiny surface areas and over-experiment. With PMF fading, he would shift the classic 80-20 split of small optimizations to big bets closer to 50-50, and reserve experimentation for genuinely sensitive, high-traffic areas, because time is the most valuable resource and a test you did not need is time you wasted.

Key takeaways

A few things worth keeping.

Distribution beats product, with nuance. Product-channel fit and decaying channels mean the product has to lead. Distributing an irrelevant product no longer works, even midterm.

PMF is a treadmill. Rising expectations and faster shipping mean you can lose product market fit quickly, sometimes to a customer who just built their own version.

The old moats are gone. Portfolio, switching cost, and brand no longer hold. Bet on speed, process power, unique domain knowledge, and unique data.

AI-native is a way of thinking. Reach for AI first, let it judge and act rather than boxing it in, let roles blur, and build processes with AI at the center.

Everyone can build. The strongest personal moat is product sense, built by shipping and talking to customers. Go build something.

Frameworks worth stealing

Product-channel fit

Before betting on a distribution channel, check what your product actually enables. Price point, friction, and business model determine which channels are viable, so a low-ticket product cannot carry an expensive ABM motion. Treat distribution and product as one system, not two separate bets.

The real moat checklist

When the old moats erode, audit for the durable ones: speed of execution and shipping, process power that delivers the last 10 percent, unique domain knowledge that is not on the internet, unique data and its connections, and economies of scale. Weight your bets toward unique knowledge and data, while accepting none of it is permanent.

AI at the center, not in a box

Design processes so AI does the judging and acting, not just assists a human doing it the old way. Let it assess whole flows and return the output, automate a real share of support tickets, and let roles blur so anyone can build. Reach for AI first when you start any task.

Plan strategy, not the whole roadmap

Stop writing detailed roadmaps longer than three to six months, since the ecosystem will outdate your assumptions. Plan company strategy and milestones, verify assumptions fast by shipping, and reserve experimentation for sensitive, high-traffic areas where the cost of being wrong is high.

Quotes worth keeping

The lines I wrote down.

Mediocre product with strong distribution beating a good product is not true anymore. Product needs to lead.

The customer became your competitor.

Speed can become your moat.

And the honest answer about the future of product teams.

My most honest answer is: I have no idea.

Rapid fire round

The conversation ran long on ideas rather than a formal rapid-fire, so here are the threads worth keeping.

If he started from scratch: Maybe go toward engineering in an industry that will matter, like energy or space. In product, just go out and build hard, because everyone can become a builder now.

What he was wrong about: That learning SQL and building heavy documentation were durable PM advantages. AI does both now.

What product wastes time on: Alignment over shipping, and roadmaps planned a year or more ahead. Plan milestones, not the whole roadmap.

Where to find him: LinkedIn for now, with a personal site and some writing on the way.

---

Wiktor Sobolak is a product leader and advisor who has built and scaled products across startups and scale-ups, including DeepL and Text, and co-organizes Growth Meetup. Find him on LinkedIn.