When did you start taking an interest in AI in development, and how has it changed the way you work?
I started paying more attention to AI when ChatGPT appeared in 2022. I was curious about what it could do and how it could be practically used in software development.
Compared to my earlier work, the change was quite significant. Before AI, I wrote most of my code manually, at most with the help of basic structure suggestions provided by the development environment itself. Today, AI helps me with more routine parts of coding, which allows me to focus more on the overall architecture and the product itself. I consider this shift to be one of the biggest changes in my day-to-day work.
Which AI tools do you currently use in development?
My main development environment is Zed, a modern code editor designed specifically for developers. Even in its basic form, it offers integrated AI features that help with writing and editing code. In addition, I use OpenCode, a tool based on AI agents – automated helpers that can work with code based on given instructions. This combination suits me best at the moment and I use it long-term.
As for language models themselves, I work with LLMs – large language models that power most of today’s AI tools. I use models such as Claude Sonnet 4.5 or Gemini 3 Pro and I approach them cautiously. I formulate prompts – the textual instructions I give to AI – so that they describe a clearly defined task. As a developer, I remain responsible for decisions about further development of the solution. With this approach, I usually get outputs that can be continued with without major adjustments.
Where does AI help you the most today?
Most often, AI helps me speed up working with documentation and writing code. In both cases, it’s mainly about saving time. Instead of lengthy searching or repeatedly writing similar pieces of code, I can move on to the next task much faster.
An important part for me is also working with the prompt itself – the instruction that describes what the AI should do. I mainly use this when working with individual functions with clearly defined inputs and outputs.
Formulating a prompt forces me to precisely define what the final solution should do and under what conditions it should work. While writing it, I clarify what problem I’m actually solving, what result I expect, and where its boundaries are. It’s similar to explaining a problem to a colleague. Thanks to this, in many cases I have clarity even before the AI generates its response.
I primarily use AI as a tool for code generation. Thinking through the design of a specific solution remains an essential part of my work, and one I’m responsible for.
Where do you see AI’s limits today?
One of the main limitations is working with information about the broader context of the problem being solved. Even identifying that context can be difficult, and without sufficient overview of the project, finding a solution becomes very challenging. AI often works only with what the developer explicitly provides and cannot fill in the necessary context on its own.
Another limitation lies in the language models themselves. Even if it may seem otherwise, AI does not think in the same way humans do and cannot search for connections between topics that appear unrelated at first glance. That’s exactly why I don’t give it a free hand.
I usually have a fairly clear idea of what the AI should produce. Once it generates something, I always review the code and try to understand exactly what it does. I consider this a necessary step. Code I don’t understand cannot be passed on or used in a project.
What advice would you give to a developer who is just starting with AI?
If I were to advise a developer who is just starting out, I would recommend using AI mainly as a learning and orientation aid. Typically for finding information about a language, libraries, or individual functions one is just getting familiar with. It doesn’t make sense to me to let AI write an entire piece of code and use it immediately without understanding it, because one quickly loses track of what is actually happening in the code.
As experience grows, however, the approach to AI gradually changes. More experienced developers eventually find their own way of integrating AI into their work in a way that suits them in the long term.
