Through Techflection, I work as a fractional CTO for startups and scaleups. That is mostly about helping companies make better technology decisions: architecture, hiring, engineering culture, AI adoption, product delivery, and making sure technology actually supports the business. Highlighting Space to Grow, where we are helping CEOs understand how their organization really works, so they can make better decisions about people, roles and growth. I’m building PromptPlanner, which is more personal. It comes from my own need to structure AI work better. A lot of people use AI in a very ad hoc way. I’m interested in turning that into a more deliberate workflow: planning, prompting, iterating, and getting repeatable output instead of random one-off chats. The common thread is leverage. I like building systems that help people or businesses do more with less friction.

I usually start building when I feel the same friction repeatedly and can’t ignore it anymore. That happened very early in my career. For HAR2009, we had to process thousands of visitors with QR-code tickets, and the first idea, using a webcam, was simply too slow. So I replaced it with a proper scanner, wrote the device integration, and built a check-in system that worked under real event pressure. That experience stayed with me: the right solution is not always the fanciest one; it is the one that survives contact with reality. With PromptPlanner, the trigger was different but the pattern was the same. I saw myself and others using AI more and more, but often without structure. Great ideas would disappear. Prompts were improvised. Useful workflows were hard to repeat. That made me think: AI needs a planning layer. Not another chat window, but a way to turn intent into a repeatable process.

Earlier in my career I probably believed too much that if the technology was strong enough, the product and business would naturally work. I cared deeply about architecture, performance, scalability, and technical elegance. I still do. But I’ve changed my mind about what makes something successful. Users don’t adopt architecture. They adopt clarity, timing, trust, and a problem that feels urgent enough. Some of my early projects, like 3D worlds and physics-based games, got attention because they were technically interesting and exciting. But attention is not the same as distribution, and technical possibility is not the same as market pull. Today I try to strengthen my business skill the most. I am learning so much things, and I am putting myself more out in the open.

Another moment was with Phyzle and Diversia. People were not just saying “cool project.” They were imagining uses for it. A teacher wanted to add it to a collection of problem-solving games for students. Others compared it to sandbox and physics games they already loved. That told me something important: when users describe your project in their own world, using their own references, it means the idea has landed.

The thing I’m most stuck on right now is focus and positioning. With AI products, there are so many possible directions. You can build for founders, marketers, developers, consultants, students, teams, enterprises. The technology can do a lot, but that is also the trap. When a product can be many things, the hard part is deciding what it should be first. For PromptPlanner specifically, I’m working through the question: is this primarily a personal productivity tool, a workflow tool for professionals, or a more opinionated system for people who use AI seriously every day? The technical part is solvable. The harder part is sharpening the promise so that the right user immediately thinks: “Yes, this is for me.”

Release early, release often. A good product is not built by thinking harder in isolation. It is built by putting something real in front of people and letting reality correct your assumptions. A high-level of product maturity matters, but only after you are solving a real problem. A beautifully engineered product that nobody urgently needs is still a failed product.