Today’s article is the 3rd (out of 4) in the EM-builder series!
You’ll read Florian’s story - about how he built a 3D no-code tool, while working full-time as a Director of Engineering in a startup.
It’s a bit different than Piotr’s and Taylor’s stories, and is about bittersweet success, difficult pivots, and the tough parts of the zero-to-one phase.
Thanks Unblocked for sponsoring today’s article!
Tools like Cursor and Claude Code are great for side projects. But when you’ve got microservices and a messy codebase, the results are underwhelming.
Unblocked fixes that. It builds an engineer-first brain by connecting to your code, tickets, docs, and discussions - and has an understanding of how it all fits together to answer questions like your most senior dev would.
The best part is, they have an MCP server that brings this context into Cursor and Claude if you use them on production codebases:
🎤 to Florian!
I’ve always wanted to get some hands-on experience with infrastructure - to actually understand how things work behind the scenes.
Back in 2022, there was a lot of hype in the no-code/low-code areas, but all apps felt and behaved the same. I thought - wouldn’t it be cool if you could design your application logic not in 2D, but in 3D? You’d place different artifacts on a kind of map, and then define data flow between them.
Sort of like building an RTS (Real Time Strategy) game, but instead of armies, you’re orchestrating your app:
I won’t lie, it took some effort. I was putting in around two hours every evening, plus all of Sunday, and a bit of Saturday too. So roughly 15 hours a week, consistently, on top of my full-time job.
That period was basically just work, side hustle, and time with my wife.
Getting the MVP out
By the end of the first year, I had built the whole thing - a full 3D app editor.
You could create simple CRUD applications: forms that send data to a backend, store it, and retrieve it later. One of our current customers is in the chemical industry. They use it as an order management system - built completely with our no-code tool.
Everything is fully handled by our platform - CI pipeline, Docker images, Kubernetes clusters, version control.
Going all in
We had an actual product in use, but I didn’t have the capacity while working full-time as a director to actually promote it, do outreach, talk to customers, and get feedback.
In my day job, I worked at a VC-funded scale-up that raised a ~$50M round B in 2021. We were building 3D scanners: you’d drive into a big sensor-filled box, and it would generate a 3D image to detect car damage.
At the end of 2023, the company didn’t secure its next VC round, and closed down. I thought: okay, this is the right moment.
My co-founder and I decided to jump in and go full-time on the project. We already had a working MVP. We knew it worked technically. Now the big question became - is there actually demand for this product?
Looking back, maybe that’s not the best way to build something: start with the tech, then look for the problem. But that’s how it happened…
Anton note: A common mistake in the EM-builder series :)
Getting first customers (a bittersweet moment)
Once we went all-in, the next step was getting customers. We started with the obvious step: reaching out to our existing network - family, friends, old work relationships. That’s how we found our first customer, which was a great fit.
They’re a small company, no technical people on the team, limited budget - exactly the kind of setup that could benefit from a no-code platform. They needed a tool to build internal apps, but couldn’t build it themselves.
Our plan was to create a self-service product - something that semi-technical users or engineers could use themselves, through a SaaS subscription.
But right at the beginning, we ran into a mismatch.
Our platform turned out to be too complex for completely non-technical users. When we showed it to people, the reactions were clear: engineers picked it up fast - they wanted more and more features like custom SQL queries and advanced logic. But non-technical people were overwhelmed. Variables, workflows, inputs, endpoints, it was just too much for them.
Everyone loved the 3D interface - but the underlying concepts were still technical. It just wasn’t usable by people without some kind of engineering background.
So in the end, we built the full application for the customer using our platform.
That was a bittersweet moment.
We had built a real product that solved a real customer problem, and it felt rewarding. The platform worked, it delivered value. The customer we built that solution for is still with us today!
But we had to face the reality - we hadn’t built the kind of self-serve experience we were aiming for. We’ve tried to iterate and improve the product, but in the end, we understood we need to try a different direction.
The Pivot: From No-Code 3D Builder to Test-driven AI for Engineers
After our first product, we realized we needed to get better at building the right thing from the start. My co-founder ran ~100 interviews with potential users to understand where there’s real demand - and more importantly, what kind of market is big enough to build a serious company in.
What we kept hearing is that people weren’t happy with the quality of AI-generated code, especially in industries where testing and compliance really matter. So we thought - instead of generating code first and testing later, why not flip it?
So we pivoted away from non-technical users and focused entirely on engineers, architects, and engineering managers.
Our idea now is to generate code from tests. We help developers write high-quality test cases - assisted by AI - and then produce code that satisfies them. It’s an approach that makes a lot more sense in regulated environments like healthcare, where software bugs can have huge consequences. This shift is already getting us interest from developers and even VCs!
If you want to hear more - you can DM Carsten and Florian on LinkedIn. They are currently in private beta and have a couple of spots left. Check out Vilosia to see for yourself (not sponsored).
What’s next
It’s been ~18 months since I took that leap.
So far, we’ve been scraping by with a combination of small revenue from the 3D no-code version and a nice government grant for highly innovative startups (they are more common in Europe). That grant gave us almost 2 years of personal runway.
Right now, we are starting to get some traction for our test-driven AI agents product, both with a customer from V1 and a new med-tech scaleup that handles critical infrastructure (our perfect ICP).
I won’t lie, the Zero-to-one phase of a startup is radically different from anything I've done before. Switching from technical IC to management was one of the best decisions I made, and I really loved it. Mentoring people who are driven and eager to go their next step is one of the most awesome things I have been a part of in this world.
On the other hand, starting a startup from scratch is teaching me so much about how products fit a market, financials, and more, that I would definitely do it over again.
Discovery weekly
My favorite: The Renaissance Worker. Why are people starting to maintain multiple careers (something I definitely experienced myself).
Stop Being Busy. Start Being Ruthless by
. On the power of anti-goals.What we’ve learned about building AI-powered features by
. How to cut the bullshit and create actually useful AI-powered features.Why everyone’s switching to AI credits by
.
Showing you can learn from failure is often more inspiring than easy wins.