Today’s guest is Tony Beltramelli, Head of AI Strategy and Product at Miro, who built Uizard from a viral ML research project into a product with over 3 million users before its acquisition by Miro. He spent seven years scaling the company from pre-product R&D to acquisition while operating a distributed team across Europe.

🎧 Tune in now on Spotify, Apple, YouTube and share your thoughts! In the meantime: Follow the Gradient and stay tuned!

🫶🏼 Melanie & Christian

PS: Has this e-mail been forwarded to you? Sign up here.

You should listen to this if you are building deep tech without clear product-market fit and need to decide between prolonged R&D versus early customer feedback.

As the conversation unfolded, the tension between founder intuition and customer reality became clear, especially in how overbuilding delayed learning.

This is one of the hardest questions in entrepreneurship, and honestly, we asked ourselves this constantly. Here's how I think about it.

First, there's a difference between "nobody wants this" and "nobody understands this yet." When we started in 2018, there was no general understanding of what AI could do for design. We had the Henry Ford problem — if we'd asked people what they wanted, they would have said "faster horses." So we couldn't purely rely on customer interviews to validate the idea early on. We had to build based on our vision first.
But that doesn't mean you fly blind. The signal that kept us going was pull. When I open-sourced pix2code in 2017, it went viral. People were reaching out asking if they could use it for work. That's not a product, but it's a signal that the problem resonates. Later, when we put out demo videos of early prototypes, people would sign up in droves to get access. That's demand, even if the product isn't ready yet.
The trap is when you confuse "people think this is cool" with "people will pay for this." We made that mistake — we waited way too long to introduce pricing. Paying users give you a completely different quality of signal. Someone willing to put dollars on the table validates your product in a way that 100,000 free signups never will. If I had to do it again, I'd introduce pricing from day one.
So my framework is: look for pull signals (organic interest, waitlist growth, people asking to use it), but validate with money as early as possible. If you have pull but no one will pay, you might have entertainment, not a business. If you have neither, you should probably iterate on your core idea.
And one more thing: things always take way longer than you think. It took Figma four years from inception to launch. It took us three and a half. That's roughly the same timeline, and Figma is now one of the most important software companies in the world. Patience matters — but patient conviction, not blind faith.

95% of our growth was organic, here’s is what actually worked:

First, I'd make the product inherently shareable. The visual “wow effect” was a key element that made Uizard marketing-ready. Because Uizard was a product generating visual outputs — app designs, wireframes, prototypes — users would naturally record videos of the AI in action and share them on social media because they were genuinely amazed. We didn't ask them to. They did it because the output was visual and surprising. If you're building something today, ask yourself: does my product produce an artifact that people will want to show others? If yes, make that sharing path as frictionless as possible.

Second, I'd build viral mechanics directly into the product and the pre-launch experience. We did this multiple times with gamified waitlists — when people signed up, they'd see their position in the queue and could move up by inviting friends or colleagues. This simple trick grew our pre-launch list to 100K users. It's a simple mechanic but it works incredibly well for products with visual, shareable outputs.

Third, don’t hesitate to invest in content from day one — but not generic blog posts. When we started in 2018, we wrote a ton of content specifically about AI for design to educate the market about what AI could do in the context of product design and software development. Nobody was searching for this stuff yet. But when ChatGPT launched and suddenly everyone wanted AI solutions, we already had our entire content funnel ranking on Google. We captured demand we had been seeding for years. The lesson is: create content around the future your product is building toward, not just around today's search volume. When the wave comes, you're already there.
Last but not least, distribution should not be an afterthought. As a first-time founder with an engineering background, we made the classic mistake of focusing on product before distribution. Successful SaaS startups have distribution inherently built into their products from day one.

I think the honest answer is that this is one of the most personal decisions a founder will ever make, and there's no universal right answer.
What I can share is the framework I'd encourage founders to think through. First, ask yourself: what do I want the next five years of my life to look like? Starting a company is an emotional rollercoaster that most people don't fully appreciate until they're in it. There's a great interview where Jensen Huang is asked if he would start NVIDIA again knowing everything he knows today, and he says no — the hardship, the vulnerability, the emotional toll is not something anyone in their right mind would sign up for. I completely agree with that sentiment. So when an acquisition offer comes, you have to be honest with yourself about whether you have the fire and the energy for another multi-year chapter of independent growth.
Second, think about what the acquisition unlocks that you couldn't do alone. For us, Miro serves over 100 million users. The AI technology we built at Uizard could suddenly reach a completely different scale inside Miro's platform. Sometimes the best way to maximize the impact of what you've built is to combine it with someone else's distribution and resources.
Third, consider the team. Your co-founders and employees took a bet on you. What outcome serves them well? What gives them the best opportunity going forward?
And finally — and this might be the most practical point — look at the fundraising environment honestly. Raising a Series B is hard even in good times. If you have a strong acquisition offer on the table, you need to weigh the certainty of that outcome against the probability-weighted outcome of raising more capital and continuing independently.
There's no formula here. But I'd say: be honest with yourself, be honest with your investors, about what you actually want, not what you think you're supposed to want.

The fundamental shift is this: we're moving from a world where humans are the central router of information — manually context-switching between a dozen SaaS tools — to one where humans delegate to AI agents, and those agents interact with your software on their behalf. The human provides the high-level goal. The agent becomes the primary consumer of your product's interface.
So what does this mean practically for founders?
Your API is your new UI. As Karpathy recently warned, products built on opaque formats with no scripting support won't survive the era of human-AI collaboration. If an LLM can't read and manipulate the core objects of your product programmatically, you're building a legacy product on day one. Designing an AI-first API today is becoming as important, if not more, as designing a user-friendly UI/UX 10 years ago.
Rethink pricing. Seat-based models collapse when one human deploys 100 agents doing the work of 100 people. The February 2026 "SaaS-pocalypse" — $300 billion wiped from software stocks in days — was the market realizing this in real time. The future is consumption-based or value-metric pricing, which is actually healthier because it ties your revenue to the value you create.

First, the AI and the product can't be separate. At Uizard, the inception of the company was from applied AI research, and the first building block of the product was a set of AI models built in-house. The product was built around the model, not the other way around. When I see companies where AI is bolted on as an afterthought, you end up with a weird disconnect — the AI experience feels magical, but the rest of the product is complex, and you get an inconsistent user experience that confuses users.
A great AI-first product team needs people who deeply understand both the capabilities and the limitations of AI. You need AI engineers who build solutions around off-the-shelves models and APIs, but you also need product thinkers who understand that defining intent is non-trivial — that turning brain waves into English sentences is far from a lossless compression format. The best AI product people internalize that AI will always make mistakes, and they design the experience around that reality rather than pretending it away.
At Uizard, our first two hires were a product designer and a computer vision engineer. That pairing was deliberate. You need the craft of great user experience married to deep technical AI capability.
It was true yesterday and even more so today: the team needs to be comfortable killing their darlings. We had to scrap features we personally loved because they weren't solving the right problem for users. Less is more — a simple product that does a few things very well is always better than a complex product doing a million things poorly. The best AI product teams have the discipline to say “no”.

In the early days, success was survival. We closed our pre-seed funding in the last month where I could still pay rent. At that stage, success was literally keeping the lights on and proving that the technology could work.
Then success became traction — users signing up, using the product, paying for the product, retaining. When we crossed $3 million in ARR and were serving millions of users, that felt like validation that we'd built something real.
But what I've come to appreciate more over time is impact. The thing that excites me most is the potential to amplify human creativity and help people solve problems that matter to them. At the end of the day, we are people solving problems for people. If the technology we build enables more people to express themselves and bring their ideas to life, I am happy 🙂 The revenue, the accolades, the public recognition — those are byproducts of impact, not the goal itself.
And personally, I never wanted to start a company. I wanted to build technology that would solve a specific problem, and that technology led to a company. I guess success for me is still fundamentally about building things that matter.

"Growth Hacking" by Sean Ellis and Morgan Brown was transformative for us. But I'll caveat that: don't invest in growth until you have strong signals of early product-market fit. The growth tactics in that book are powerful, but they're accelerants — if you pour accelerant on a fire that doesn't exist, nothing happens.
On pricing specifically, Patrick Campbell from ProfitWell was a huge influence. His thinking on pricing as an ongoing discipline rather than a one-time decision shaped how we approached our pricing strategy.

I'm quite a nerd so I consume a lot of AI research papers on arXiv (although it's hard to keep up these days!)
But practically, I think the most impactful learning comes from talking to other founders and operators who are in the trenches. The best insights I've gotten came from candid conversations with other entrepreneurs.

There's a classic Y Combinator advice that says: "If you're not embarrassed by your launch, you've launched too late."
I think this doesn't fully apply to AI startups. It's true to some extent — you shouldn't over-polish everything. The rest of your product can be awkward. But if you're an AI startup, your core AI magic needs to be pretty good before you launch. I'd say at least 80% there.
Here's why: AI products create expectations. The more intelligent a software product appears, the more intelligence users will expect from it, and the less willing they are to learn how to work around its limitations. If your core AI feature is mediocre at launch, users won't give you the benefit of the doubt the way they might with a non-AI product. They'll try it once, be disappointed, and leave.
So yes, launch before you're comfortable with everything else. But make sure the magic works.

I'm a huge guitar nerd, and it has become my meditation practice. When you play music, your brain can only focus on the playing — it's one of the best ways I've found to truly unwind. It's meditation through rock and roll I guess 😅
Beyond that, I try to cultivate serendipity. Some of the best ideas and connections come from unexpected places, but you have to create the space for them to happen.

This is by far the one I’ve shared with founders the most: try a gamified pre-launch waitlist. This was probably the single most effective growth hack we ever deployed, and we used it multiple times.
The mechanic is simple: whenever we were a few months away from launching a new feature or product, we'd create a marketing video and a landing page where people could sign up for early access. When they signed up, we'd show them their position in the queue — and they could climb higher by inviting friends and colleagues. That's it.
This turned every new feature launch into a viral growth engine. It grew our pre-launch list for Uizard 1.0 to over 100,000 signups. When we applied the same technique to promote major new features, we were getting 10,000 or more new signups every other day.
It works because it taps into two basic human motivations: the desire for exclusive early access and the social dynamic of sharing something exciting with people you know with a feeling of scarcity.

You have full control over what we send to you. We sent a weekly rundown on Tuesdays, and our podcast brief on Thursdays. If you do not want to receive this Thursday one, you can change this in the email preferences below.