Relentless pace, we’re making progress!

The company

• Home sweet home: We’ve set up base at Wat.bxl, and we love it. The vibe is great, the people are great, and there’s something genuinely energizing about being surrounded by other entrepreneurs going through the same grind!
• First services: We’ve started delivering consulting and workshops to our first clients. It’s a good way to gain market insights, bring in some early revenue, and collect a few logos for the website. The admin side of things took me completely by surprise–I did not expect it to eat up so much time. But the clients are happy, the feedback is strong, and honestly, we’re having fun doing it.
• Officially registered: We’re now a real company, registered in Belgium. More admin than I anticipated, but it’s done. It’s official.

The product

• We found our topic: After three months of interviews and iterating on ideas, we landed on something we’re genuinely excited about: making PR review more efficient in the era of vibe coding. We think the current tools are simply not up to the task, and we’re building a new one. The exact shape is still evolving, but we’re moving fast, and our understanding of the problem is deepening with every sprint.
• Things move fast: We work in one-week sprints and try to keep our assumptions and hypotheses documented. It’s not uncommon for those hypotheses to be completely overturned by Friday. Monday’s conviction becomes Friday’s rethink. It can feel unsettling–we both like to properly understand a problem before acting–but the lesson keeps being the same: build fast, trust your instincts, and let the market correct you.

The mood

Life in Brussels is good. I’m meeting a lot of people and reconnecting with old friends. Definitely less lonely than I was in Zurich.

The challenge right now is sleep. Between the startup, the services, the admin, and a social life that somehow still exists, downtime is hard to find. I know I’ll have to stabilize eventually. But right now I feel genuinely energized–more than I have in a long time. Some friends told me I look five years younger. I believe it. It feels like being a student again: classes during the day, socializing at night, sleeping when possible. That’s what going all in does to you–you stop looking for energy and just find it. It’s a good feeling.

The rollercoaster is on, we are entrepreneurs!

The company

• Finding a home: We’ve visited a lot of co-working spaces in Brussels. We feel it is important to have a place where we can exchange ideas and also avoid staying alone at home for our own sanity.
• I say wat?: We got into wat.bxl, a new incubator in Brussels. It’s exactly like what you would imagine for an incubator; an old factory building converted into a workspace with “quick-and-dirty” desks and a crowd of people grinding on their own projects. We felt the energy of the place and the people. You exchange ideas, you receive feedback, you get to know people, you realize you’re not alone doing this crazy journey. After visiting, we both said it was the place to be for startups in Brussels.
• The red tape: The admin parts take longer than expected. There are a lot of things to be aware of when starting a company in Belgium.

The environment

• The local scene: The startup scene in Belgium is growing. There are more and more events, more and more people are interested in startups. Yet, the french speaking part of the country is still lagging behind (like, a lot). I believe it’s up to us to change that.
• The Belgian melting pot: I kind of enjoy the vibe in Brussels. The city is not too big, not too small, and the people are friendly. I like the mix of people from different backgrounds, the different cultures, the different languages. I can talk to people working in the music industry, in cinema, in science, in startups, in big corps, in government, in academia in just one day–I feel like I’m lucky and learning a lot.

The product

• The AI shifts: These are wild times to build a software product. Companies have woken up to the power of AI for coding and are using it to build their own internal tools. The market sentiment has shifted: instead of looking for new SaaS subscriptions, many are looking for ways to cancel them. Startups have already made the leap, SMEs are testing the waters, and big corps are considering but it’s going to take some time. It feels like the cards are being reshuffled, and we’re right in the middle of it (and quite well positioned to benefit from it).
• The marathon strategy: We’re going to take on some consulting and service work. The goal is three-fold: keep the lights on, gain market insights, and get some initial logos on our website. Plus, we’ve had some very cool projects proposed to us lately. More on that once things are official.
• The iterative approach: We do niche by niche, doing interviews, doing proof of concepts, testing our hypotheses, checking if there is traction. The speed at which we currently iterate is quite high. Honestly, I’m sometimes reluctant to say exactly what we’re building because I know it’ll probably be different by next week. It will just take time until we find the right idea–hence the consulting part.

The mood

I’m not going to lie: this is one of the hardest things I’ve ever done. You are constantly surrounded by doubt. People tell you you’re crazy, or that you should just settle down and chill.

Personally, I’m back to learning a lot of things (sometimes even the basics) and I feel like a junior again. It’s a humbling experience, but I actually enjoy that part. I like the “burn” of learning something new, whether it’s about myself or the world. The funny thing is, whenever I feel too insecure, I just start coding. I know I’m good at it; it soothes me and brings me back to center (autism confirmed 😆).

The journey is tough when you don’t have many positive signals. It feels like walking through a dark tunnel. Every now and then, you find a window with a bit of light and fresh air; you take a deep breath and realize you can keep going. It’s all about resilience. Having friends and family around is a huge help. I’m eager to see what’s next.

I said I would do updates on my entrepreneurship journey, so here it is for the past year and January 2026.

Why we pivoted

Background: We’re both engineers. In early 2025, I was still in my postdoc, and Arnaud (my co-founder) was working in industry, so we were building this on the side.

We built FeedbAct to make employee feedback more human and insightful.

The idea was simple: replace rigid questionnaires with open, conversational feedback, then use AI to summarize and extract key themes for HR and management.

We did dozens of interviews, built an MVP, and started testing willingness to pay.

Then reality hit.

HR is a tough market

It’s crowded, and acquisition cycles are painfully slow. To give you an idea, some companies only run feedback surveys every 2 years—simply because they can’t afford to go through the procurement process more often.

If you target medium to large companies, there’s little chance you’ll do business with them unless you already have a big name or proven results. Companies turned out to be more risk-averse than I expected for this matter (probably my researcher bias). They prefer something that’s proven and safe rather than innovative.

And on top of that, it’s hard to quantify the ROI of feedback/HR tools. The benefits are indirect and long-term, which makes the product feel like a “nice to have,” not an “I need this now.”

Sales friction everywhere

Our sales conversations went well until we reached the “paid demo” stage. That’s when the enthusiasm dropped (classic, but it’s good we tested the willingness to pay early). Even though we could probably have closed a few deals through personal connections, we feared we’d hit a plateau within six months, once we’d exhausted our network.

We also underestimated the friction to entry. Our product required the entire company to adopt a new process. That meant convincing not just HR, but also the CEO and sometimes the board. A few people told us openly: “We’d rather stick with our current tool, even if it’s not great, than add one more platform to our stack.” The phenomenon of “SaaS fatigue” is real.

We weren’t our personas

We built something we, as employees, would have loved to use. But our clients were HR and management—and they had different priorities. We noticed the misalignment in interviews: what excited us didn’t excite them. Their “key features” weren’t ours.

The market moved faster than expected

When we started, it looked like no one was doing what we envisioned. We knew we had to move fast—but five months later, some companies had launched almost identical features and even made them part of their core pitch.

I still believe we weren’t wrong; we were just too little too late. The differentiator we had wouldn’t have lasted another year.

When to stop

There’s always a balance between pushing harder and knowing when to stop. For us, the key signal was when the excitement started to fade. That’s when we decided it was time to move on.

Lessons Learned

• Do your homework continuously. Some companies were already heading in the same direction—we noticed too late. Part validation bias, part of not knowing our market well enough.
• Talk to people early and often. You’re probably not your persona, and your user might not be your buyer. The only way to find out is to talk, listen, and ask good questions (Mom Test style).
• Go-to-market is important. We built a solid product, but we underestimated how hard it is to get it in front of the right people. In B2B, credibility, access, and timing often matter more than product quality in the early stages.
• Integration beats innovation. Even if your idea is great, customers often prefer something that fits into their existing workflow over something entirely new. Next time, we’ll think about integrations and partnerships much earlier.
• Ego is a trap. Stopping was hard for me because I had publicly announced I was working on this. But I managed to convince myself that it’s braver to recognize early when something isn’t working than to double down and dig deeper into the rabbit hole.

What we’re doing now

So I moved to Brussels in January and am now working full time on new ideas. It honestly feels great to have some time to think about new things instead of rushing to the obvious next step. We’ll see what happens.

Anyways, let’s go to it. In this one, I am going to talk about some limits you might encounter (or not notice) when you apply Gaussian distribution to quantify, say a risk or an uncertainty.

Using Gaussian to predict people’s height

The usage of the Gaussian distribution is largely motivated by the Central Limit Theorem, which basically says that if you take enough samples in the real world, you will end up with a curve looking like a bell. For example, say I want to estimate the height of people entering a room, I start by measuring people as they enter, and update my model everytime. The process would look like the following gif.

In this graph, the x-axis is the height (cms) of people and y-axis represents the number of people with that height in what I sampled. As you can see, the more I collect people’s height, the more the curve looks like a bell. But what is all the fuss about this bell? Well, it turns out that once you get to this familiar shape, you can predict the height of the next person with some kind of confidence. For example, after sampling thousands of people, I can say with 95% confidence that the next person’s height is going to be between 155 and 194 cms, which seems pretty logical, isn’t it? This is called statistical inference (or prediction using machine learning in nowadays’ buzzwords) by the way.

Using Gaussian to predict people’s income

Well, now if we do the same experimentation but collect people’s monthly income (oddly enough, I always get more attention from people when talking about money). After sampling a few thousands people, I see the mighty bell shape coming back:

Now, I can make the same kind of prediction as for their height; with 95% confidence, I can say that the next person’s salary is going to be between 1,625 and 2,800. I can use this process pretty much everywhere thanks to the Theorem (some people even say if you have more than 30 samples you’re good to go), brilliant!

Well it turns out that if you start doing that, and making people believe you can put actual numbers on the uncertainty of everything you want to predict, you are going to be burned hard.

Imagine the next person coming to the room, his name is Swarren Blackett… and this guy makes 10 MILLIONS per months. Well, this is far off of the Gaussian distribution we have been constructing. But we said 95% confidence, not 100%, didn’t we? Yes, we did but anyways, the problem here is the magnitude of the difference between the average salary you have been sampling and this guy’s salary. Imagine the error you make when predicting the next dude has a salary of 2,200… If an unexpected event like this happens, it might ruin everything you built until then. This is called a Black Swan by Nassim Taleb, a brilliant dude.

To give you an idea of the effect of such an event on the bell curve and the 95% confidence interval, here is the same data, with my guy Swarren:

Now you are able to say with 95% confidence that the salary of the next person is between -60,000 and 65,000. Boy this does not look good anymore, heh? Take a moment to realize how it changed from the last confidence interval (1,625 to 2,800). All of this with one single person.

Gaussian usage in real life

Now a lot of people (me included until I read Taleb’s book) use and believe in these confidence intervals and uncertainty measures every day (from medicine to finance). For example, predicting with 95% confidence that a large bank is not going bankrupt 🙃. People blindly relying on Gaussian measures are not protected from unexpected events, even if they believe they do (well, technically in finance it is not even their money). And in some cases, this reasoning can cause a lot of damages (like actual lives).

Summary

We use Gaussian distributions to quantify uncertainty in a lot of cases because it is familiar and easy to manipulate;

• 1st problem: Not all kinds of data fit a Gaussian model. For example, people’s height fits such model since it seems very unlikely to encounter a person that is 4 meters tall (as the model would predict). On the other hand, we saw that people’s income does not follow such model since there are people earning way above the average.
• 2nd problem: When you think you quantify the uncertainty, you actually quantify how much the data you have seen varies. Not how much all the data can vary. There is a difference between seen uncertainty and unseen uncertainty, and the problem is that with Gaussian distribution, we only quantify the former, while the latter might have big consequences.

So what can we do?

• Question yourself if a Gaussian fits, and do not put too much trust in the model.
• There are also other distribution models, so called power-law that could fit better some kind of data.

I am no expert in statistics, thus I am going to stop here before I give wrong information. Cheers!

Life is made up of compromises.

Edith Warton

Now that I look cultivated with my citation, I might have one or two things to develop on the matter. As an optimization fan, I am constantly identifying situations where we could be more efficient in our lives, this post is about the efficiency of decisions.

Decisions in life are very often subject to compromises. For example, in our society, one has to trade its time for money - we usually refer to that as “working”.

To optimize such choices, what we usually do is observe multiple life scenarios and their outcomes. Here, I will use data generated by my imagination for the sake of the example, but be assured that it is close to what we do in AI or optimization. Anyways, my data are average salaries per month, based on the working hours per week of a person. Here is what it would look like in a table:

In a graph, the data would look like this: The work-life balance graph.

Naturally, you can see in the graph that the more time you spend working, the more money you can expect¹. The data seem logical and everything makes sense (at least I hope so). Notice that I have drawn two zones also in the graph. The red one is the inefficient zone; it means that if you are in that zone, well friend you work too much for the salary you get. The green zone is the unusually good zone, it means that you are doing better than the average for the hours you do. You can think of a billionaire for the latter.

In the middle, we can observe a curve. That curve is the efficiency zone: whenever you are on a point on that zone, it is in general impossible to increase one criteria, without sacrifying in the other. For example, you cannot earn more without working more, and you cannot work less without losing salary. This concept is called Pareto efficiency by the economists and guys like me.

Well now the news is that as long as you stay on that curve, no point is objectively better than the other. That means a person working 20 hours a week for 1000€ is not “better” or “worse” than a person working 40 for 2000€. This precise thing illustrates one of the many compromises we have to make in life.

Take-home - Finally, here is what science says: the point which is optimal depends on your preferences. So, yes, earning less money than your neighbour can be optimal for you, if you prefer more free time. And no, working full-time just “because everyone does” is not necessarily the best strategy for you (and that’s OK!). In the end, in the quest for happiness, looking at what choices the other make is irrelevant for you. What is relevant is determining what’s best for you i.e. your preferences.

Of course, this point does not restrict to work-life balance only. Here are other compromises examples: transportation price vs eco-friendliness, buying a house vs a flat, meat vs vegetarian diet.

PS: the optimal point for people is rarely lying on one end of the curve. Which is also why I think extremism in any form is a bad way to simplify decisions. The best trade-off often lies somewhere in the middle.