Who this?!

This is my first post out here, so first, and as a documented promise to you: the contents of this blog will NEVER be generated by AI. The only use of AI in this format will be to rephrase posts to make them more readable...

So hi, I’m Jonathan, a software engineer becoming an entrepreneur in one of the most unclear times to be an entrepreneur.

My software engineering career began 9 years ago when, as a computer science student, I started working as a junior developer at Check Point. Since then, I’ve worked at a mid-size tech company, learning the ways of the engineer, and then moved on to work for a startup in the data engineering tooling space. The last two were the lion’s share of my career so far (3 and 4 years respectively).

I had always believed in the path of professionalism. What I think it meant was:

1. Learning a subject to become knowledgeable in it.

2. Practicing that field to become proficient in it.

3. Building on existing studies and leveraging my knowledge and proficiency into other territories (cross-pollination of proficiency)

My greater purpose was always to eventually become an entrepreneur. Don’t get me wrong, I really love software engineering. I love algorithmic thinking, I love breaking down problems and understanding them deeply, I love designing architecture to accommodate my solutions, and I just love finding a nasty bug and crushing it (intelligently)!

But all of this was a means to a cause. I always saw myself as an innovator. Well, to be honest, I never really acted on innovation proactively (although I did win a hackathon with a team working on an idea I came up with), but part of it was a kind of introversion I always had. It wasn’t always apparent, and I got by day to day, but when it came to bigger decisions, it was there all the way.

That all changed when my daughter was born. Life finally slapped my introverted face upside down, and then I realized the ever-so-prominent and simple truth: YOLO (yep, I just wrote it down...).

The last six months at the company I worked for were very tough for me. I worked with people I loved working with, but I knew, very clearly at that point, that I was wasting my time. Having said that, now that I have a lot of responsibility on my hands, I couldn’t just quit. I had to plan it right. So I worked a bit longer, made my calculations, and figured out when I could quit. Then I did it. I quit my job to become an entrepreneur. That was super exciting and super scary in tandem.

I had my first idea that I wanted to ship. It was a passion-meets-profession kind of idea. You’ve heard of those before... I’m a big music nerd: music theory, listening to different genres of music every day, discovering new bands, singers, producers.. you name it. I learned how to produce music, and although I wasn’t a professional, I figured I had what it takes to create something in that field.

But wait, you must be wondering how this fits with my philosophy from before. Well, AI.

I believed that AI could complete me. Be the music producer I wanted to be. Be the audio engineer I needed in order to create a music production service for beginner producers. Be the entrepreneur consultant, market researcher, marketing consultant, and many other things.

I failed with my first idea.

It was for a couple of reasons:

1. I wasn’t proficient in the field, and, realized at this point, AI cannot replace proficiency in non-deterministic by nature fields. AI isn’t enough if you’re creating a startup in a field you have no clue about. Well, this is true if the field has deep roots and is fairly established (you can definitely succeed in all sorts of surface-level-enough fields like screen lockers and whatnot). The lack of proficiency meant that:

   1. I couldn’t understand what beginning producers were actually expecting.

   2. I couldn’t validate with anyone.

   3. I had no track record to show that I could do what I promised to do.

2. I had 0 connections in that industry, and in this industry, you better know someone who knows someone.

3. I thought I could do it because multiple AI agents kind of let me think that I could.

What did AI agents make me think?

Claude, ChatGPT, and Gemini made me think that my idea was plausible enough to go after. You see, these services are adapting to make you feel good. They adapt to empower you, even when it doesn’t make sense. So when I asked about creating Temprom, a service to deterministically master EDM tracks based on the track, genre, and instrument profiles, they all went bananas over the idea. Then when I raised a few questions, they killed it. Then when I answered their answer, they went bananas once more. And then again. And again. Both directions.

Those intervals kept happening throughout the implementation of the idea. I reached the point of clarity when I started thinking without them. Then I realized: “Well, why wouldn’t the competitors do the same? Are they so big that they can’t move quickly enough to implement such a thing? Why are they all using in-house models? Oh, because you just can’t do it deterministically... the whole IP is in the model’s weights, since it’s absolutely not possible to understand the artist’s purpose and thoughts based on a track’s DSP.”

Yes. And just like that, it was gone. Sadly, but without looking back.

Something quite similar happened to me with the second idea I had: Parrot Studio, a service that creates Spotify and Apple Music short videos based on the track’s DSP, lyrics, or a short description written by the artist. This time I wasn’t deep in the music part but more in the video deduction. This was, again, something I wasn’t very proficient in, and while some record labels I approached and offered free videos to told me they liked it, none actually used the videos, and I believe actions more than words.

The most beneficial response came from a record label owner who told me the honest truth: this is AI slop. I thought that as well. Dropped it.

At that point, I just wanted to get something out. Something useful, easy to create, and something that would teach me something new. Here came my third idea: I’d scratch my own itch and create something I’d benefit from using immediately: Marinador!

Marinador is an iOS app for chicken breast marinades. Yes, exactly. As a gym-goer, I’m thawing a lot of chicken breast every week, but I’m not sure what to do with it each time, and I also don’t like running around buying ingredients for marinades- I want to use what I have. This is what Marinador is all about.

I won’t go too deeply into it, but it was a nice project to learn what it’s like to create and ship an iOS app. It also taught me what I need to do in order to get people to use it, and to be honest, I’m not into social media marketing. And that idea too, was passed by me.

At some point during the implementation of Parrot Studio, I realized that I cannot trust AI services’ opinions. I know I sound like the noobest noob to ever walk on earth, and that this kind of thinking isn’t exactly expected out of a software engineer, especially when I do have a deep understanding of the internal workings of these services, but what can I say, I fell into the trap. It made me forget the fundamentals. It made me forget the simple truth: I’m talking to math functions and reviewing the answer that was weighted more heavily than the others (and at times, not even that).

This was a triggering point for me. From now on, I will only ask models for evidential info. No opinions, no market research (the results were opinionated), no evaluations. Only facts.

After completing my Marinador endeavours, I started thinking more deeply about the whole thing. What could even be considered a moat for a company in days where Claude just eats the world, and the other services try to eat Claude?!

Don’t expect a clear answer. First, because no one knows. Second, because we, human beings, are changing and responding to this massive shift in the world. When changes arrive with such force, and from so many angles, all I could think to do was try things out.

I used Claude Code to create a framework for idea generation (I’ll open source it when I get the chance), in which the AI service only asks you questions about ideas you have, industries you’re thinking about, or some other random thoughts you come up with, in order to help you get to a complete idea. It’s not allowed to express opinions and not allowed to rate your ideas, only ask questions that help you estimate the moat and necessity of your idea. It then runs the idea through a validation pipeline that gathers all sorts of information (updated to the last 6 months), and passes it through Smart Bear’s Viable Business Model to get more hints about it...

All in all, I’m pretty proud of it, and I have one idea that I’m about to test in the wild.

My next ideas will focus on target audiences I know and love, and that I have deep knowledge of: developers and product managers. Well, at this time of AI disruption, it’s better to focus on functions and not roles, since everything is changing and the audience you create software for might not even be human. So I’ll rephrase: my next ideas will be about helping software development and product improvement.

It’s a very interesting time we’re living in, and we’re all trying to figure out where we fit in the puzzle (and where the pieces are, and how many pieces there are, and whether solving a puzzle is even a moat...). I still think it’s a great time to go and ride the entrepreneur road, especially as a software engineer who actually knows how to ship products and work at scale (you’d be amazed to see the results of Claude Code, performance-wise, and not in a good way).

So this is basically what I intend this blog to be about: my progress, thoughts, and findings about the new AI era, things I learn, things I succeed at, things I fail at, etc.

I hope I’ll have good news to share with you real soon.

Jonathan.