# Dennis Jansen · interactiondesign.nl > I find the real problem before you build the wrong solution. I'm like AI. I always sound reasonable and trustworthy. But it is all about the execution of the insights. This file carries the visible content of https://interactiondesign.nl/ for AI agents. ## Observations ### 1. DO YOU TRUST AI AS MUCH AS YOUR LAWYER? You use AI for reading legal documents. The output sounds reasonable, but it would never stand before court. An architect checked an AI summary of a 60-page document against the referenced paragraphs: they were off. You can't tell AI to be an expert in something it is not. Source: Legal, architecture. Move: only an expert can audit it ### 2. THE TEACHER LET AI BUILD THE EXAM. EVERY STUDENT FAILED. When even a teacher believes AI is better than him, or is too lazy to create his own material as the expert that he is, he will create the wrong teachings and the wrong exams, and not recognise that the AI can't mark the exams. Source: Education. Move: the expertise was there, unused ### 3. THE WHOLE DEPARTMENT. USELESS DURING THE FLOOD. The government created policies to help water management collect data. Work needed a permit, systems were filled, everything in place, but all for the wrong reasons. The whole purpose of a water authority isn't administration. It's avoiding the risk of floods, maintaining the rivers, ditches, dikes. That is the starting point that went bankrupt, with over-regulation for the wrong reasons. Source: Government, water authority. Move: what was underneath ### 4. THEY ALREADY DECIDED THEY NEEDED AN APP. The funding was to help long-term unemployed people work in the care of the elderly. I asked to interview the elderly still living at home. They were not good at mobiles and apps. They wanted to be self-sufficient. My concept: no app for the elderly, only an app for the employee working in the flat. It was not built out. The money was spent for the wrong reasons, but for the right top-down checklist. Source: Elderly care, field research. Move: what was underneath ### 5. HOW DO YOU KNOW IF YOUR AI MODELS ARE BASED ON THE RIGHT DATA? The task was optimising app screens for dietitians. It was not so much the dietitian that needed the right app, but making the patient the centre. A patient's weight over time has so many input points. Doctor, at home, nurse. Morning, afternoon. Source: Healthcare, field research. Move: validated for its purpose ### 6. 2.5X FASTER CPU. NORMAL PEOPLE WON'T SEE IT. New laptops: 2.5x faster CPU, GPU data speeds, X amount faster than the previous generation. Normal people won't see measurable changes in their use. Intelligence and reasoning are useless when there is no validated real-world data, validated for its purpose. Source: Technology, AI. Move: validated for its purpose ### 7. SOMETHING GOES WRONG. AI CONFIRMS IT WAS EXACTLY THE INTENTION. The output is always reasonable or even confirming. Something goes wrong and it confirms how nice it is that this was recognised, that this was exactly the intention. It needs to be visible that a mistake was made, or that there is no complete information to give an answer at all. Source: Government · translated from Dutch. Move: only an expert can audit it ### 8. AI CAN'T TELL TIME FROM A CLOCK WITHOUT NUMBERS. An image of a clock is built by looking at images of thousands of clocks. Most show the aesthetic time, because that looks the best. If eighty percent is trained on one view, the twenty percent that has an alternative view won't be taken along. Source: AI training data. Move: validated for its purpose ### 9. AI IS NOW LIKE THE DEVELOPER AND THE MANAGER. Without the UX researcher, the UX lead and the service designer that know what problem needs to be solved. My guess: AI can manufacture reasonable output for somebody without the field knowledge, who figures it's good enough. And the senior with years of experience catches the faults. Source: AI, product development. Move: what was underneath ### 10. THE SPINACH STORY. ALSO WRONG. A data error can exist for almost a hundred years: spinach was thought to have ten times more iron than it has. I told that story on stage. And that story is also wrong. There was no decimal point error. It was repeated for fifty years because it sounded right. That's actually how AI works. Source: Nutrition, my talk of 2026-09-26. Move: catch against myself ### 11. THE CAUSE OF ULCERS WAS PROVEN IN 1984. PHARMA KEPT SELLING ANTACIDS. The cause of ulcers was proven in 1984, the cure antibiotics. Pharma kept selling antacids. My own open question: are all ulcers bacterial, or a segment of them? Proof of fact. Source: Medicine, a story I point at. Move: the correction never travelled ### 12. BEST IN THE WORLD. $52 IN 43 MINUTES. The famous violin player Joshua Bell earned a couple of dollars performing in a metro station. Same talent, same Stradivarius, sells out concert halls at 100 dollars a ticket. When your talent is worth nothing if it is not performed in the right surroundings, that tells you something about the public, not about your talent. Source: Music, a story I point at. Move: same song, another venue ### 13. CUT 30% OF THE WORKFORCE WITH AI? A manager was asked how he could lessen the workforce by 30% by using AI. It's not about the AI that would make it more efficient. It's the inefficiency of the legacy systems that would be proven first. Source: Retail, summit conversation. Move: what was underneath ### 14. THE CLIENT IS HAPPY. THE PRODUCT IS NOT USED. Most companies build for money, not for value. Client is king. The gap is where the client is happy, but the product is shit and not used. Source: Product development, reading a Rick Rubin interview. Move: what was underneath ### 15. AI DOESN'T ASK WHAT IT IS MISSING. AI currently doesn't ask what it is missing or doesn't know. It just spreads information, as if no questions are necessary. That annoys me a lot, to be honest. Source: AI output. Move: only an expert can audit it ### 16. FRESH FROM SCHOOL IS NOT EXPERIENCE. Knowledge from doing is the difference between coming fresh from school with some theoretical knowledge and no hands-on, and an experienced expert. I spoke to a wooden house builder. The way he projects his knowledge of the subject and his hands-on work for years gave me trust in his capacity to perform. Source: Expertise, a conversation. Move: only an expert can audit it ### 17. THE ROUGH DIAMONDS ARE ALREADY IN THE ORGANISATION. Within a water authority, the curious, open-minded ones spoke my language. We needed to look for the rough diamonds in the organisation and give them the tools they need. Within their own expertise they can teach their fellow experts. Source: Government, insurance. Move: the expertise was there, unused ### 18. PROCUREMENT BUYS THE SOLUTION THAT SOUNDS REASONABLE ENOUGH. Public tenders and procurement are in the lead for the solution. They have the incentive to sell a solution that sounds reasonable enough to buy, escaping the underlying problem that would be too hard to grasp. Source: Government, a conversation. Move: what was underneath ### 19. NOT YOUR KEYS, NOT YOUR REASONING, NOT YOUR KNOWLEDGE. I noticed the lock-in of the different AI models and see the direction it is heading. So my setup runs on my own hardware. Raw first, no summaries. Never trust, verify. Source: My own system. Move: not your keys ## Latest - 2026-10-20 · proofbridge.nl (https://proofbridge.nl/) - 2026-10-20 · New observations (https://interactiondesign.nl/#observation) - 2026-10-20 · How I work (https://interactiondesign.nl/setup/) ## Contact - info@interactiondesign.nl - Requests from people and agents I don't know: https://proofbridge.nl/