During the summer I followed my wife and her new job to live in Barcelona. After a few months, I can easily tell you, that we certainly could have ended up in a worse location. I really do love it here!
Before the move, I ended my job in Switzerland, and so far I have not started a new one. When the move had become a certainty, I knew what I wanted to do.
Following up on a belief that I had, that nobody in my occupation ever had the time to do what I felt is right now absolutely necessary: Diving full time into what it really means for a product person to work with AI tools as close and as efficient as possible, without corporate rules made in a previous dark age of knowledge work.
One important rule I set for myself:
Building on and using my own infrastructure.
Nothing like Lovable or any platform that solely wants to make you completely dependent on them would ever be used. Just an AI coding tool that I use to build my own custom tools and my own products. They will still work if the AI stops working, as I still have my own code running on my own servers. It definitely makes it harder, but I was sure that it would be worth it.
Also:
No more posts drafted or even completely written by AI.
Although it takes much more time, LLMs are just not good at writing. And you miss out on all the benefits of what writing itself can do for your own thinking process. So I am back to doing it myself. I will add an attribution disclaimer per post.
My Kind Of University
The image above is the view of the beautiful old university building in Barcelona that I can enjoy regularly, when I sit in one of the cafes on Plaça de la Universitat.
With the speed of new developments happening in the AI space, university courses would need to adapt almost every week. Maybe we will get there one day, but I don't think established institutions are even close to this.
So I chose the traditional learning by doing method. The first step was to find at least one important enough project that would make use of my quite intense love for product building. Besides many personal projects like a finally really good grocery list app or a now irreplaceable notes and dictation app, all extremely customized to my own thinking patterns and habits, I needed a larger project with not only me as the target group.
The Project I Focused On
I found it a while back when talking to a friend about his work and the horrible manual processes involved in doing it. It involves huge PDFs of compliance rules and regulations and cross-matching information in old ERP systems by hand. That was a fascinating insight into a work process I had no idea about. A process that many companies in Switzerland are facing, but that is so niche that most of them are just accepting the cost of doing it manually.
As someone who enjoys the challenge of solving problems, I immediately reacted to the pain involved in doing it by hand. It meant that the provided relief by a working solution should be really sweet. So it would be the ideal use case to base a theoretical two-person company on. One for tech, one for business logic. We would be the perfect combination of complementary skills.
How It Went
Today, the product is already very close to being seriously useful for end users in our targeted niche. But it needs a few more weeks of polishing until we can really stand behind our demo. So it is not public yet. The plan is to soon offer a publicly available version to get feedback.
But even with a great product, commercializing it is a different story. So we are looking for partners right now.
For this blog, the more interesting part will be the learnings I made while continuously adapting my processes and my own toolset: Doing product, software development, infrastructure and security all by myself using only bare-metal linux servers, an AI coding tool and of course some open source software.
The Loop Back
I would like to characterize my last months as a phase of intense study or research. I went to my own kind of university and learned a ton about what today's LLMs and agent harnesses are capable of, how to extract value from them and how to bring back joy into product development.
Outlook
I really feel that this technological change will require more effort to adapt to than previous changes, especially for knowledge workers. So I will keep writing about it. Watch out for the following topics I will try to cover in the near future without a particular order:
My own custom toolset I host on my servers and continuously improve to make the absolute most out of the AI coding tool I use, but without getting too dependent on one provider. I think it will be called "Cheap Enough To Try".
Even though I say I am not that dependent on one provider, of course there is already some real dependency. An interesting topic in itself as it is also about the trust we develop with the heavy use of the same model. Working title: "Trust and Dependency".
The belief that knowledge workers without their own custom agent harness/toolset will soon not be competitive anymore. Agents will mostly just talk to other agents. As a knowledge worker, you will need one tailored to exactly your strengths and weaknesses, one that creates a real competitive edge. Starting to understand this argument is the first step.
My own personal memory system I implemented. Something I previously talked about here.
End user feedback loops and how to nurture them with highly custom built feedback tools that depend on just the creator.
Cutting the two-week development cycle to minutes. The incredible speed of testing and iterating on ideas and the pure joy it brought me.