Turning AI hype into real solutions – challenges of 6 years running an AI company

Anyone who thinks running an AI company in the heart of technological innovation is a string of success stories is mistaken. Six years ago, I started my AI journey in Eindhoven, the beating heart of Dutch high-tech, full of ambition. But reality proved far more challenging than most conference talks or media stories suggest. Running an AI company in manufacturing and agriculture – our focus markets – demands perseverance, learning from setbacks and dealing with feedback that’s often discouragingly honest.

Over the years, I’ve faced not only technical challenges but also a market that wasn’t always ready for AI. Many clients didn’t know what to do with it – or even if they wanted it at all. Instead of applause and rapid growth, I was met with scepticism and doubt. “Isn’t AI just a hype?” “Why invest in something I don’t understand?” “Won’t everyone be able to do this themselves soon?” These questions force you to prove your added value again and again.

My biggest lesson is that integrating AI in machines demands much more than simply downloading and fine-tuning an AI model from the internet. You need a broad team of specialists: from data annotation to embedded engineering, deep learning and system integration. Over the years, I’ve seen many AI startups and independent experts come and go. Vision AI for industrial machines is simply too complex for a lone freelancer or small team. Luckily, I managed to grow my company to a size where we can offer total solutions that go far beyond just downloading yet another AI model from the web.

From my experience running customer projects, I’ve seen firsthand how easy it is to underestimate just how dynamic and unpredictable the market for AI tools and suppliers can be. Early on, we assumed we could build on a stable technology stack, only to find that suppliers frequently shift strategies, discontinue support or suddenly change requirements. This constant flux has forced us to adapt quickly, sometimes switching tools, APIs or even entire platforms to keep our solutions viable. For my company, dedicated to integrating AI into machines that must be supported for a decade or more, it became clear that relying on any single technology or provider was simply too risky.

To address these challenges, we spent the last five years developing our own deep learning operations platform. This gives us control over our technology stack, shields us from supplier volatility and lets us deliver stable, future-proof solutions for customers whose machines must be supported for a decade or more. By building and maintaining this platform in-house, we reduce our dependency on external suppliers and control costs. Most importantly, it allows us to deliver reliable solutions to our customers, even as the AI landscape continues to evolve.

Another trend I see is the use of open-source AI models. Companies often choose these because they seem ‘free.’ Many vision AI models and code repositories were developed by universities about five years ago. At first glance, this seems ideal: no license fees and quick access to advanced technology. But support for these models often disappears, as universities shift their attention to the latest trends and stop maintaining older software. Moreover, transparency about training data and methodology is often lacking, which poses risks for reliability and compliance.

The market for AI tooling and suppliers is also highly dynamic. Many AI companies must adjust their strategy as they grow. Data annotation companies that once handled small projects for a few hundred euros a month now only accept projects starting at 50,000 euros or more. They’ve become so large that they leave smaller jobs behind, while the number of big projects remains limited. As a client, you’re constantly searching for new suppliers and investing in new tools and APIs. At the same time, some AI companies collapse under their own success. Too small isn’t scalable; too large makes you vulnerable to market fluctuations. It’s a delicate balance between staying focused and remaining flexible. Recently, I realized my company is now in this phase as well.

Yet, I often hear, “What you do, anyone will be able to do soon.” My answer: yes, but not everyone has the knowledge, experience and partner network to deliver real AI value right away. My latest challenge is: do I focus on innovation and staying ahead or do I take the time to capitalize on what we can already do? Always wanting to be at the forefront sounds attractive, but the risk is that you never reap what you’ve sown.

After six years, I’ve learned that success in AI doesn’t come from hype but from building robust solutions that truly work – and keep working. A concrete example of this is our recent pivot on developing an inspection platform for high-tech machine parts. Rather than just showcasing AI capabilities, we built a solution that addresses real customer needs in manufacturing. By focusing on deep expertise, adaptability and listening to what the market really needs, we create lasting value with AI. That’s the lesson I hope others will take to heart as well.