Personal Software
While this has been going on since the early days of ChatGPT, when people realized these language models are good at writing code, I think 2025 marks the rise of coding agents.
Coding is one area where these models truly shine due to several factors:
- Code is very well documented, with the entire community putting everything they know—every problem and solution—on publicly available sites like StackOverflow and GitHub.
- Code itself is a language with less ambiguity than natural language; it’s much more well-defined.
- Code is usually testable, meaning whatever the large language models produce can be tested and verified. There are still problems with verbosity in code generation, but that’s not our concern right now.
The focus here is on personal software.
We’ve been living with the idea of personal computers for decades, but software was something only very specialized, well-trained people could create. Now, with coding agents and “vibe coding”, there’s an undeniable opportunity for people to create their own apps.
I’m not going to spend too much time on how realistic or feasible that is, or the large gap between the marketing hype and reality. But I’ve seen _very non-technical _people vibe code useful applications for their hobbies or small work-related use cases, and it makes them genuinely happy because there’s real empowerment there.
I’ve also seen highly technical developers with lots of experience create personal software and be genuinely amazed at how much they accomplish in how little time, and how that opens up the possibility to free us from being locked into platforms and services.
It’s not like the App Store. In the past, if you had a use case, you’d research the appropriate software tool, go through the App Store or buy software, and figure out how it worked with your particular needs.
While vibe coding seems much better at first glance, it introduces the problem of almost infinite choice.
Think of this as analogous to cooking: when you go to a restaurant (the App Store), you choose from the menu, relying on descriptions or images if you’ve never tried the dishes.
If you learn to cook, you can follow recipes but also start changing them, and that’s when it gets really complicated.
A good—or rather, experienced—developer will do magic with these tools because they can provide good and precise instructions, review the output, and draw on tons of professional experience in system design, UX, UI, and more.
The non-technical person, on the other hand, can work with a large language model and build out features, workflows, and solutions that are completely nonsensical if you step back. The entire situation reminds me of The Simpsons episode where Homer gets to choose his dream car and builds this completely ugly and ultimately undesirable customized vehicle.
