Building an AI Venture Studio
Disclaimer:
This is a lot of information, but it’s everything we wish we had when we started last year. It’s
honest, and we’ve put it together as a team over the past few months with care.
This is all free, No fluff, no course, no upsell, just the real story we wish we had
when we started. If you're building or thinking about it, this one's for you.
We've hosted startup-style hacker sprints in:
❖ 🇫🇷 France
❖ 🇳🇴 Norway
❖ 🇳🇱 The Netherlands
❖ 🇦🇪 Dubai
❖ 🇺🇸 The USA.
We accept applications so follow along, we might be building in your city next :)
—---------------------------------------------------------------------------------------------------------------
-
This is our full behind-the-scenes playbook.
In one year, we went from three grads with no dev experience to shipping 7 products,
landing real clients, and running an early-stage AI-powered product studio.
We built fast. Learned faster. Made mistakes. And now want to let people in by sharing
everything: mindset, tools, tactics, and lessons.
- Koen, Hugues and Marcus
Table of contents
Building a Venture Studio
Stage 1 - Our First Venture: Alumco
Stage 2 - The Summer Pivot
Stage 3 - 50 Calls in 10 Days
Stage 4 - Launched in 3 Weeks
Stage 5 - The AI-First Summer
Stage 6 - What’s Next?
Appendix - Resources
CTO AI Prompt Engineering Guides:
Link 1: AI prompt engineering:
Link 2: Courses that we took:
Link 3: Mem0 Walkthrough
Link 4: Prompt Formulas:
Stage 1 - Our First Venture: Alumco
We started in education, not because we were experts, but because we were students.
We thought we would be able to get our first clients thanks to this. We’d grown up around
international schools. It felt like the closest problem/market to “founder-market fit” we had.
So we built Alumco, an alumni intelligence platform for international schools, giving schools
insight into where their graduates are and what they’re doing in the world.
We launched with no software dev experience, online resources, videos, talking to software
engineers tutorials, and late nights. Most of our early traction came from cold outreach and
hustling at 2 conferences we got in for free. I did over 200 calls with schools in 6 months.
Surprisingly, the appetite was there. Our niche solution landed better than we expected.
When our first school signed a multi-year deal, we were stunned. We felt lucky, but also…
like something clicked.
Over the next six months, we closed 9 paying B2B clients, all medium-ticket contracts. We
were pumped.
But we learned quickly: distribution is everything. Even if the product is solid, cracking
B2B sales in a slow-moving industry is a beast of its own. Education has just two sales
windows a year. This is something we saw as a big bottleneck
We made many many many mistakes, our favorite one was spending 5 weeks building and
training an AI search feature we thought would blow minds. Turns out, the admins just
wanted simple filters on a map. Nobody used the fancy AI tool.
It stung. We wasted 5 weeks basically. But we learned: build what they need, not what’s
impressive.
Alumco was a win, but it also taught us how much speed and distribution matter. And that
lesson would shape everything we built next.
Stage 2 - The Summer Pivot
As exam season rolled in, everything slowed down.
Our backlog cleared. The product worked. But suddenly, it felt like we were filling our days
with secondary tasks things that didn’t move the needle.
No growth. No momentum. Just stillness.
We knew if we kept building, we’d start overbuilding. The industry was slow and we were
eager every day to build fast.
Taking a break wasn’t on the table. All three of us had bet big on this path.
❖ One of us had dropped out of university for a year from one of the best universities in
Scandinavia.
❖ Another turned down an offer from Morgan Stanley and two other top banks in
London and Paris.
❖ They had walked away from multiple job opportunities to keep going after facilitating
a company acquisition.
So yeah, it was stressful.
The idea we had been obsessed with was starting to feel… slower, less glorious than we
hoped. But we didn’t sit still.
We made a call: pour the stress into action. Because we were having trouble sleeping
anyway
Zooms. Calls. Emails. DMs. Whatever it took to open a new door.
And there was one key mindset that held us together:
We were never in this just for the idea, we were in it for the team.
From day one, we knew that the three of us had something rare: complementary skills,
shared hunger, and speed. That was our edge. And now it was time to use it.
Even if the opportunity wasn’t in our niche, if it let us build, ship, and learn fast.
Stage 3 - 50 Calls in 10 Days
When things slowed down, we didn’t.
We knew from experience: our biggest wins came from talking to people not theorizing.
So we each opened up a Google Sheet and mapped out every single person we knew.
It took a full day. Friends, classmates, colleagues, second-degree contacts, everyone went
in.
Then we color-coded it. Who might have a budget? Who’s in pain? Who can we send a
message to?
And then… we started calling.
Lawyers. Bankers. Founders. Traders. Influencers. Farmers. People in fishing tech, real
estate, edtech, you name it.
We spent most of the time asking questions and listening.
The first 70% of every call was just asking questions. Deep ones.
What’s your biggest bottleneck?
Where are you losing time or money?
If you had a magic wand for your business, what would you fix?
Turns out, people love talking about their work, especially when you ask the right
questions.
Half-hour calls turned into hour-long conversations.
We took so many messy notes.
Most problems were too big and too chronic. Cashflow issues. Plateaued growth. Stuff no AI
stack could fix in a sprint.
But some were ripe.
By the end of the sprint, we had 5 paid opportunities on the table:
❖ A law firm wanted automation for legal doc drafting.
❖ A new hedge fund needed internal research tooling.
❖ An influencer wanted to scale a coaching business.
❖ A company had lead-gen ops bottlenecks.
❖ A watch store wanted a smart inventory system
That week was chaos.
The record was 12 calls in one day.
We had to scramble just to keep up, track notes, and remember who was who.
But it somehow worked.
Stage 4 - Launched in 3 Weeks
From the five paid opportunities, we picked the one with the most upside:
An influencer who was growing fast but drowning in operations.
He was brilliant at coaching, building, and scaling his audience...
But tech? Not his thing.
So we stepped in. The mission: build him a client management system with a brain, a
tool that tracked every client’s progress, surfaced insights, and gave his team control as he
was scaling.
We still can’t quite believe how perfectly the opportunity landed.The wildest part? He signed
the contract before we built a thing, just off the Figma designs. We didn’t write a single line
of code until the contract was signed.
We divided and conquered:
❖ I handled the sales, client discovery, and product flow.
❖ I turned his scattered process into wireframes and flows in Figma.
❖ Then passed it to my co-founder, aka the execution engine, who used AI tools to
build fast.
We stitched together a lean, AI-powered stack.
Everything moved fast. Sometimes too fast.
There were late nights. Bug hunts.
Half-broken demos. Last-minute fixes.
But we were obsessed we wanted to blow him away.
And we did.
In just 3 weeks, the platform went live. His users were onboarded. His team was running the
entire operation through what we built. He thought it was fantastic
The platform is still live and still growing. His employees and clients use it daily. And now
we’ve got two referrals in the pipeline, with calls booked for next week.
We still cant believe how fast it can go. We have no formal dev experience.
Just speed, teamwork, and AI and suddenly, we were in the game.
Here is probably what you came for :) Our AI tech stack. These tools have been a game
changer for us. We’re also assembling a database of incredibly valuable open source
resources that we have used in the last months. But ill need a bit of time to assemble this.
Category Tools Purpose
Frontend UI Libraries shadcn We use shadcn/ui to build clean, modern interfaces without starting from scratch. It gives us ready-
made building blocks so we can design our dashboards or tools quickly and consistently no need to
reinvent buttons, menus, or layouts every time.
Frontend UI Libraries v0.dev With v0.dev, we just type what we want the interface to look like, and it gives us working code
instantly. It’s like having a designer and developer rolled into one. It helps us test new ideas super
fast.
Frontend UI Libraries 21st.dev 21st.dev is our go-to place when we need high-quality interface components others have already built.
It saves us time and helps us keep our product design consistent without building everything
ourselves.
Frontend UI Libraries ui-verse UIverse is a huge collection of simple design elements like buttons and cards. We use it when we
need something quick or want visual inspiration. It’s lightweight and handy for adding small design
touches.
Backend & Infra Supabase Supabase is what powers most of our backend. It handles our user accounts, saves files, and even
helps with live updates. We rely on it to store data like user info, AI model results, or saved
conversations it’s simple but powerful.
Backend & Infra Clerk Clerk handles all our login and user management stuff. Instead of building sign-up and login pages
from scratch, we plug Clerk in and it just works. That way, we focus on building our AI features, not
dealing with passwords.
Backend & Infra Vercel Vercel is where we host our website and app. When we make a change, it’s live on the internet in
seconds. It keeps our site fast and running smoothly for users all over the world.
Backend & Infra Sentry Sentry tells us when something breaks or slows down. If our app crashes or loads slowly, we get
alerts and can fix it before users notice. It’s how we stay on top of bugs and keep things stable.
Backend & Infra PostHog PostHog helps us understand how people use our product. We can watch user sessions, test new
features, and track what’s working or not. It helps us improve based on real behavior, not just
guesses.
AI/LLM Layer OpenAI We use OpenAI’s models like ChatGPT to power the brains behind our product. Whether it’s writing,
answering questions, or generating content, this is what makes our product feel smart and
responsive.
AI/LLM Layer Gemini, Gemini, from Google, is like a super researcher that works across text, images, and even code. We
use it when we need concise, helpful answers or when building AI that works across multiple formats.
AI/LLM Layer Claude Claude is another AI we use that’s especially good at writing clearly and being thoughtful. It’s our go-
to when we need safer or more careful responses, like in educational tools or support assistants.
AI/LLM Layer Mem0 Mem0 gives our AI memory. It remembers what a user said before, so our product can feel more
personal and natural over time. Without it, every session would start from zero.
AI/LLM Layer Jina AI Jina AI helps us turn text and other content into searchable formats. If we’re building a feature that
finds the right answer or suggests content, Jina powers that behind the scenes.
AI/LLM Layer pgVector + pgVector lets us store and search through complex data like finding similar messages or answers
HNSW inside our database. It’s fast, reliable, and means we don’t need a separate system just for smart
search.
AI/LLM Layer Face & These tools Face & Ticktock break big documents into smaller parts that our AI can understand. We
Ticktoken
use them when users upload long files so the AI doesn’t get overwhelmed and can answer more
accurately.
Mobile Expo Expo helps us build our mobile apps once and run them everywhere on iPhones, Android, and even
the web. It saves us time and means every update reaches all platforms at once.
Stage 5 - The AI-First Summer
This summer, we stopped thinking like a startup and started operating like a product studio.
The rule was simple:
Build with AI, Kill what doesn’t work and Double down on traction
In just 4 months, we shipped 7 projects:
Project Description
Video to Ideas Turns long YouTube videos into bite-sized insights, not fluff.
BuildThisNow.com Trends, Ideas & 3,000+ tools we’ve collected that we wish we had at the
start.
Prompt-Fox.com A sharp prompt refiner trained on docs from OpenAI, Anthropic, Google
& more.
The Influencer App Custom mobile app we built from scratch for a coaching business.
Client Management Dashboard for coaches with built-in AI feedback and workflow tools.
Platform
Alumco Our alumni platform. Still running. Still supported. Still useful.
Lead Gen Enrichment Internal system for enriching and qualifying leads at scale.
Chat YC An AI Co-founder trained on the knowledge of all the Y-Combinator
resources
We didn’t kill any projects outright but some naturally drifted to the background.
We built things we actually use ourselves. If the traction wasn’t there yet, we watched and
waited.
Our workflow? Pure sprints.
We mapped out our weeks like builders on tour trying to find teams around the world who
shared our energy.
Speed was the only metric.
Sometimes we built features nobody used. But we learned fast, adjusted faster.
We didn’t waste time debating what the feature should be. We just shipped it.
Turns out, that’s what real learning feels like.
Yeah, it was hard.
Juggling side projects. Managing chaos. Keeping priorities straight.
But after a year of working on just one slow-moving idea, we wanted the chaos.
It gave us range, pattern recognition and momentum.
So what are we now?
In the short term: a product studio with real revenue from multiple ventures.
But long-term, we’re still searching for the one the product we’ll scale for real.
Until then, we build. We test. We kill.
And we don’t stop moving.
Stage 6 - What’s Next?
We don’t see ourselves as experts.
We’re not AI researchers or serial entrepreneurs.
We’re just three recent grads soaking up everything we can from the people pushing the
frontier forward.
But we’ve seen something firsthand:
Small, fast teams can now build real products.
And if that’s true for us, it’s true for everyone.
Our thesis?
Soon, anyone will be able to build anything.
Distribution, storytelling, and founder market fit will matter more than ever.
We think, if you know the user, and you can get it into their hands you're already ahead.
We’ve talked to hundreds of builders.
Some still doubt AI’s role in serious software development.
And yes there are real limitations.
But here’s the reality:
❖ We’ve shipped 7 products in 4 months.
❖ We’ve landed paying clients with no formal dev experience.
❖ And our learning curve is on 2x speed because we lean heavily into AI tools instead
of fighting them.
Where we’re headed
Right now, we’re operating like a product studio:
❖ Build it fast
❖ Test with real users
❖ Scale if it clicks
❖ Shelf it if it doesn’t
It’s not forever.
But in the short term, it’s the best way to learn fast, earn trust, and stay lean.
Will we still be a studio in a year? Maybe.
Maybe not.
You’re seeing more builders launch portfolios of micro-startups and we’re keeping our
options open for now.
One thing’s clear:
We think we’re on the right track.
We know more, can ship faster, and think of more ideas than we did a year ago.
People in the space we looked up to are now reaching out, impressed with what we’ve
pulled off.
What we won’t do
We’ve learned not to over-commit too early.
No more pouring our soul into slow-moving industries for now.
No more romanticizing a single idea at the expense of learning and momentum.
And if we had one critique of the builder space right now?
Too many are ignoring the tools:
❖ Not using AI deeply enough
❖ Not leveraging design + dev accelerators
❖ Not automating their research or outreach
❖ Not staying close enough to what’s being released every day
We used to do the same.
Now we force ourselves to stay shipping and stay learning.
What’s next
Right now, we’re doubling down on the project with our coaching client.
He’s growing fast, he believes in us, and we think the long-term upside is massive.
It’s getting most of our attention.
On the side, we’re still shipping micro-startups often in a day.
Some will fade.
Some might surprise us (hopefully :) ).
But either way we’re not slowing down.
We’re just getting better at knowing what to bet on.
We’re not experts. We’re just learning out loud and building as we go.
If anything here helps you move faster or think sharper, then it’s already worth it. Follow our
journey since we’re only getting started :)
Happy to see you made it to the end
Send me a message saying:
“Invite”
I'll also add you to a free, invite-only community where we'll be sharing more details about
what we're currently building. We’re planning to open-source a lot of our work, since that’s
how we learned everything ourselves just a year ago. Let me know if you'd be interested in
joining!
⬇️Appendix below ⬇️
Appendix
CTO AI Prompt Engineering Guides and Links:
Link 1: AI prompt engineering:
AI Prompt Engineering:
Link 2: Courses that we took:
Document sans titre :
Link 3: Mem0 Walkthrough
mem0 File
Link 4: Prompt Formulas:
Prompt Engineering Formula Recap
More to come!
