Entering AI Development...

Day to day software development has definitely taken a turn in the AI realm. We all know that, it is interesting to see how technical choice is becoming less of a blocker. This article is about observations, it is not offering solutions to anything but something I felt I had to write based on my experience, it does not reflect general industry experience.

Over the years pre proper AI adoption (before 2025), more often than not the technical choice was a huge limiting factor in lots of things we do. When we develop for the web we are constrained by the frameworks we choose to adopt, when we build features we are later held back by the technical debt we introduced, when we build backend we are limited by scaling issues and so on. That’s a lot of things to keep in mind. All of them have solutions and still have however the thinking now is becoming less and less concerned about the concise details and more on how can I use AI to help me do that better.

This is very much interesting to people who spent years specialising in those techniques. AI definitely is not yet a replacement and won’t be for a few years, it still struggles with brownfield projects and bigger projects. You have the concern of token count and how much consumption which really equates to how much it costs. Latest LLMs show us flashy capabilities which is very admirable however it still needs to translate into the real world.

Shifting gears into the topic of building. Us engineers, we always loved (still do) the joy of building and being in the state of flow. Code comes out like a flow of water, never stopping and we gain satisfaction from it. Now it is less the case, now it is an engineer in command mode. A boss demanding to build AI, demanding details, looking after token consumption, heavily focused on expressing the right idea on AI than the details of the code. It is very much shifted although code is still important. They are important, using the correct framework is still as important and having good support of dependencies is still crucial although the concern is not on the pieces as much.

Some may object with that as they would say that’s the next era of development and it needs to be this way. Exactly, it is the next era. It is the future. It is different given the control is being handed more to AI. Code is not being viewed the same and that maybe fine.

Brownfield issues, manual reviews and costs won’t vanish soon, it is becoming more different. We can now have oversight with AI. So the programmer is AI, the reviewer is AI, the QA is AI and so on… That’s not very far and not unrealistic. Us engineers will still be worried about decisions but on a new level of abstraction. The same way Java overtook other programming languages when it evolved back in the day.

It is a bright future ahead with AI. We need to be responsible as well, although that’s a problem that’s difficult to start solving given all the different interests at play. Cybersecurity of AI generated code is a point to think deeply about going forward.