Software Engineering After AI

About a year ago, I started experimenting with using AI to develop software.

At first, I wasn’t very impressed. I had quite high expectations, partly influenced by the media. But then I began to realize that you have to guide the AI by asking the right questions. Don’t expect the AI to generate readable, maintainable code if you simply say, “Can you implement feature X for me?” It will just do that (sometimes incorrectly) and generate one big blob of code.

Writing code with AI is similar to agile development: you work in small steps. In the diagram below, you can see the workflow I used. I added a brain icon to indicate where human interaction is especially important. You need to be able to ask the right questions to push the AI in the right direction. Without proper guidance, the AI quickly get lost.

Sometimes the AI is simply not able to solve your problem. In those cases, you have to implement it yourself.

I was quite impressed with the results. I built a user interface with React (I normally use Angular) and implemented a fairly complex application in just two weeks. Along the way, I also learned a lot about React. Normally, this would have taken me months. I was able to keep the code manageable by constantly refactoring, and I used AI here as well to speed up the process.

Lesson learned:

• Vibe coding without deep domain knowledge is an illusion. Without solid software engineering fundamentals, you are not guiding the AI — you are wandering with it. The quality of the outcome is directly proportional to the quality of the questions you ask.
• Mastery of a specific language or framework is becoming less critical. I don’t need to be a React expert to build a solid React application. What matters far more is architectural thinking, system design, and a strong understanding of web fundamentals and UX principles etc. These are the skills that allow you to evaluate, refine, and direct the AI toward the right solution.
• The role of the software engineer is evolving. We are becoming conductors of intelligent systems. Instead of writing every line of code, we orchestrate, refine, and shape the output. This shift demands experience, judgment, and the right mindset. The future belongs to engineers who understand systems deeply and can collaborate effectively with AI.

You can see the result here: https://robert-sanders-software-ontwikkeling.github.io/rs-x/

For more information see rs-x opensource project: https://github.com/robert-sanders-software-ontwikkeling/rs-x