Thoughts on AI coding

The Situation

As you have already heard many times before, “AI is gonna wipe software engineers”, or “software engineering is gone”, let us put some pressure on our brains and contemplate about this. The software engineering job market is going tthrough one of the worst moments. Companies are doing massive layoffs. Hence, what comes to mind is, “claude can write better code than a junior engineer”.

What I Think of AI Coding:

AI can write code, even good code. But there are some issues with it, to which I don’t think a clear answer is there, but we will come to know as time unfolds. Don’t get me wrong, I use AI to code almost daily, but I use it systematically when I’m doing something serious, because I don’t want to mess it up later. Here are some concerns with AI coding in the long run:

The AI Slop Code Loop:

Today’s AI models were trained on thousands on repositories comprising of code written in different programming languages and technologies. AI coding has come to a point where it’s now mainstream. But does everyone verify the code AI writes and pushes. This same code might be used to train future coding models, which means, these models will possibly be trained on AI slop code, and everyone knows, if the model is trained on slop, it emits slop. If this loop continues overtime, it will be the destruction of software, and the real engineers of that time, who can clean the code, will be in high-demand.

The Senior Judgement in Software Engineering:

With AI now able to execute the syntax writing, people who have developed the judgement to identify good code and architecture can execute at the speed of lightning. But, these engineers developed this judgement after years of time spent fixing bugs in production. The question is, can a junior skip all that hard work, and suddenly become a software engineering guru, powered by claude code, cursor, or codex?

AI Code and It’s reliability in Complex Systems:

Imagine, you’re sitting in the NASA headquarters, trying to enhance a software that has to do with real-time communication, must be very reliable, but you prompt claude code, “fix this, or I’ll go to jail, please”. Doesn’t sound a great situation to be in.

To avoid such a situation, you must know your craft. That requires you to go deep in your domain, understand the code, the lower mechanics of how such softwares work, and use AI systematically, with a great eye for bad code, and that insight is only developed when you’ve spent time debugging and building actual stuff.

Conclusion:

So, those were some of my thoughts on AI coding. I know it’s a hard time to be in, but for all the software engineers out there, or those who are becoming one, you must know how to code even when AI can do it for you. It is worth it. Not saying that you shouldn’t use AI, but use it to implement your own solution, not AI’s. That’s how you keep that spark alive, and you become sharper overtime.

Note: The expressed ideas are my thoughts, anyone can disagree with them without hesitation.