My Take on AI in Coding: It’s Time to Rethink How We Build Software

Why Developers Must Start Thinking Like Architects in the Age of AI

I’ve been using AI and AI agents for coding for a while now. With the guidance of my mentor, M. Merouan Abdellatif, and my professors, I’ve learned just how much AI can help speed up development. It can generate boilerplate, suggest fixes, and even scaffold entire features.

But here’s the reality: AI can’t (yet) build fully reliable software systems. The code it generates often contains vulnerabilities, lacks proper structure, or doesn’t fit neatly into a long-term architecture. Relying on it blindly is a recipe for bugs and endless debugging cycles.

So, I’ve come to believe something important: if we want to truly benefit from AI in software development, we need to change how we learn and think about computer science.

From Coders to Architects

Most of us were taught to hard-code everything — to write every line manually, debug every syntax error, and rely on raw problem-solving skills. But today, much of that work is being offloaded to AI.

That doesn’t mean coding is dead. It means our role is evolving. We can’t just be coders anymore; we need to think like architects. Instead of fighting AI-generated code or avoiding it out of fear of errors, we should learn how to design systems where AI fits naturally.

An architectural mindset allows us to:

• Define the high-level structure of our applications.

• Break down complex problems into modules AI can help implement safely.

• Spot and mitigate the weaknesses of AI-generated code before they turn into production nightmares.

Challenges with AI Agents

Of course, AI itself isn’t perfect. AI tools like Copilot or Claude 4 often lose context over long sessions, leading to inconsistent suggestions. Some of them also lack deep understanding of business rules, producing solutions that “work” technically but miss the bigger picture.

This is why understanding the fundamentals still matters. If you don’t know how your code works under the hood, you’ll end up stuck in what I call code proliferation — endless, unreviewed AI snippets piling up, breaking things silently.

By knowing the basics, we can guide AI better, debug faster, and avoid becoming overly dependent on it. AI should be our assistant, not our replacement.

The Path Forward

AI is growing at an exponential rate. Soon, large models will likely maintain full session context, meaning they’ll be able to remember your entire project state without forgetting prior steps. That’s going to make AI even more powerful — but also riskier if we don’t adapt our skills.

So here’s my takeaway:

• Learn the fundamentals so you can spot AI’s mistakes.

• Adopt an architectural mindset so you can build systems AI can enhance rather than break.

• Stay adaptable, because AI will soon do more, faster, and at greater scale — and we need to keep pace.

AI is changing how we work, but it’s not replacing the craft of building reliable systems. It’s up to us, as developers, to evolve from pure coders into architects who can guide AI and make it work with us, not against us.