How AI Removed the Execution Barrier for My Hobby Projects

andre

July 28, 2026

I will be the first to admit that AI has its flaws. It hallucinates, overengineers solutions, and loves duplicating code. But despite these quirks, the latest models have genuinely increased my overall motivation to build things.

My brain is always generating ideas for new tools or projects. The problem has never been a lack of imagination. It is the groundwork. Even the simplest website requires boilerplate, configuration, and repetitive tasks. This work is unavoidable and incredibly tedious. It is exactly the kind of friction that kills a hobby project before it starts. AI shines here. If you give it an exact specification, preferably with tests to measure the results, it will churn out decent code for the mundane stuff.

People throw around wild productivity metrics. Some claim a tenfold increase, others say double. My experience is more nuanced. For certain tasks, AI is actually a distraction. But for other tasks, I simply would never have undertaken them without it. The real win is not just doing old things faster. It is doing things I previously avoided.

Take a specific hobby project I have used almost daily for years. I originally built the interface using a niche technology. The community is supportive but tiny. Library updates happen maybe once a year. If I wanted new functionality, I had to wait or write it myself.

Recently, I decided to rewrite the entire interface using a more standard framework. The catch is that I did not know the new framework and had zero interest in learning it. Building interfaces is not my forte.

So, I pointed the AI at my old codebase and told it to replicate the behavior in the new framework. After a short iterative loop, it got the job done. I still had to review the output, catch regressions, and fix inconsistencies. But I reached feature parity with my old app very quickly.

That is when the real fun started. The whole point of the rewrite was to add features I had been putting off. Now, I just tell the AI to build them. It handles the interface framework easily, likely because this is the kind of work it is trained upon. I even pushed it further and generated a complete mobile version of the app. What used to be a web page is now a native mobile application. I have a long list of future plans, and I know the AI will handle the heavy lifting.

This highlights a major shift in how I work. For a long time, the pain of wanting a new feature was not enough to overcome the pain of actually coding it, even when I knew exactly how to do it. Time is always the limiting factor in my rather busy life, so hobby projects are very low on the priority list. I would always think how many months, possibly years it would take me to complete certain projects in the limited free time that I had. That assuming I almost gave up spending time with friends and family, which I’d rather not. Now, the execution barrier is almost gone. As long as I can describe the desired outcome in detail in plain English, I can build it. My imagination is no longer constrained by the physical hours required to sit down and type out boilerplate.

AI is able to give an average bottom-of-the-greasy-pole developer an ability to feel like a high-flying project manager with a whole team at his total disposal. This is probably how these guys feel when they have a task and just divide it among their subordinates and then watch them sweat while doing the actual work. The main differences are twofold. AI is way cheaper, because even a very incompetent junior developer still isn’t very cheap, even if he or she comes from a country with a low average income. And, second, even the best developers are rarely prepared to work 24-7 and without being grumpy or very idiosyncratic. At best, you can expect them to work eight hours a day effectively and efficiently. The truth is, based on many years of my personal observations in the field, that most people work well only for a few hours a day and slack for the rest of their allotted hours. That’s even when there are any good. There are plenty of those who, even after spending a whole day in the office, often produce nothing but bugs that others have to fix.

AI is not perfect, but it’s already on the level of an average middle-level developer and way better than almost any junior, except for very bright and talented things. And it is there 24-7, more or less, as long as you have internet and AI credits. AI costs can quickly escalate, the internet if full of such stories today. However, if you use it smartly, then it’s not that expensive either, and it’s definitely cheaper still than the humans.

We all hear about AI companies bleeding money and, in effect, subsidizing their customers. Basically, the tokens that they sell, even though they’re not cheap, they’re still not fully covering the full outlays those companies are making to build all the infrastructure and to pay all their super expensive brainiacs. Therefore, the situation might change in a few years when they finally run out of everyone else’s money. But for now, the odds are still in our favor, and we definitely should make the best use of them.

Another area in which AI has make my life so much easier is debugging. What AI does really well is solving all these mysterious bugs and failures. Sometimes your code just doesn’t work, and you know you can find out the root cause, but because the original problem is not very obvious, you also have the gut feeling that it will take quite a while before you get to the bottom of it, and this causes you to feel rather low, because you know the next few hours will not be very enjoyable. Seeing a huge backtrace occupying a few screenfuls is never the highlight of the day. Now, instead of digging into logs and perusing the code time and time again in search of this little minus instead of plus, or this comma instead of a period, you can task the AI with this problem. Let it read the logs, let it read the code. Surprisingly often, it produces results pretty fast and pretty accurate in the debugging area. As usual, this is never guaranteed, and sometimes it can go off on a wild goose hunt, but more often than not, and especially for those very obvious in the hindsight problems, it’s pretty good.

I have been riding this wave of enthusiasm since the middle of last summer. AI coding assistants were decent back then, but they are genuinely amazing today. It makes you wonder what coding will look like in two or five years. Everyone is speculating about the future right now. I remain optimistic that the tools will just keep getting better. Assuming the machines do not decide to destroy us first, of course.