Let’s get straight to the point: AI is taking Software Engineers place, its faster to setup and bugless in the beginning… How about later? How about after some changes and some adjustments in the core of the applications because of business decisions and world changes? What will happen? Get ready and lets analyze it together.
A shift that’s already underway
AI is already changing the way software is built. It’s not a matter of if, but how fast you’re willing to accept, understand, and adapt to it. Many predict that by the end of this decade, AI will be responsible for producing up to 90% of all code. From powerfull code assistants like cursor to vibecoding nocode like Replit…
However, in the next two to three years, that number is unlikely to exceed 30%.
Why the gap? While AI is excellent at producing results quickly, it struggles with the kind of complexity that emerges in large systems. Especially those tied to the messy logic of human behavior and business priorities.
“Systems program building is an entropy-decreasing process, hence inherently metastable. Program maintenance is an entropy-increasing process, and even its most skillful execution only delays the subsidence of the system into unfixable obsolescence.”
― Frederick P. Brooks Jr., The Mythical Man-Month: Essays on Software Engineering
Let’s be clear: AI doesn’t struggle with logic. It struggles with us. With people. Our decisions are influenced by emotion, intuition, and experience, not just data, even the most rational CEO’s and leaders. Think about when your partner says “I’m fine,” but you know they’re not. Or when a CEO kills a profitable project simply because they don’t trust the person behind it. AI can’t make sense of those things. And yet, the software it writes often needs to reflect and support decisions driven by them.
In these situations, the issue isn’t just AI’s limitations, it’s the loop of translation: a human expresses intent in a prompt, an AI writes the code, and then another human needs to adapt and evolve that code within a real system. That back-and-forth creates cracks, especially when the system grows.
The Rush for Speed
Tools like GitHub Copilot and ChatGPT already help developers get more done, faster. Tasks that used to take days now take minutes. One developer using AI can produce what once took a team. This kind of acceleration will continue to attract businesses chasing quick results.
Founders with limited technical knowledge, junior engineers, and even experienced developers are using AI to release features in record time. The benefits, faster launches, reduced costs, and smaller teams , are impossible to ignore.
The Hidden Cost
But speed comes at a price. And that price is complexity, particularly when it’s invisible until too late.
AI-generated code often lacks cohesion when integrated into bigger platforms. It might work in isolation, but struggles to scale or adapt when things change. As systems grow, AI can’t keep up with the ripple effects of design choices, performance bottlenecks, or architecture decisions that require long-term thinking.
Many people using AI tools lack the background to recognize these challenges. They’re deploying code that solves today’s needs, but unintentionally creating massive maintenance costs down the line.
AI doesn’t (yet) have a strong grasp of:
- * How to structure systems for adaptability
- * When to prioritize performance over simplicity
- * How one change can introduce cascading issues across multiple components
- * Design patterns that keep software stable and understandable
Without this knowledge, teams relying too heavily on AI will face rising error rates, longer recovery times, and growing frustration as systems break under pressure.
Hitting the Wall
Yes, we’re heading toward a future where AI writes most of the code. But before that, many companies will face a harsh reality. Over the next few years, teams will hit a wall. Projects rushed with AI will become harder to maintain. Bugs will increase. Flexibility will vanish. The result? A growing demand for experienced engineers who can step in and make sense of the mess.
This is why AI-generated code will likely plateau around 30% in the short term. Not because the technology isn’t powerful, but because building reliable systems takes more than outputting lines of code.. it takes real understanding.
What’s Coming
Eventually, AI tools will improve. They’ll learn to include context, track intent, and help design better systems, not just generate functions. Teams will also become more thoughtful in how they use AI, blending its speed with human insight.
But for now, we need to stay cautious. The coming years will divide those who use AI to strengthen their codebase from those who use it to patch together short-term results.
Software complexity won’t disappear just because AI can write quickly. And writing fast is not the same as writing well.
A Case in Point
Here’s a simple example from a study I’m currently finishing:
A trading startup created a platform for currency exchange. Their five-year plan aimed to capture a large share of the global market. Instead of hiring 20 engineers, they used AI to build the entire platform with just two developers and a small team of product managers and analysts steering the process.
The results were impressive, at first. They reduced bugs by over 40% compared to traditional teams, and their output was 70% higher.
But then the market shifted. A new U.S. president introduced tariffs that made currency trading unprofitable. The company pivoted to focus on energy and oil trading instead.
And this is where the problems began.
Rebuilding the platform for the new strategy wasn’t just difficult, it was nearly impossible. What should have taken weeks would now take months. Why? Because the code was built by AI to solve specific problems with no clear documentation, no architectural foresight, and no long-term planning. The AI-generated scripts weren’t designed to adapt. And the people operating the system didn’t fully understand how it worked under the surface.
Without experienced engineers, the company faced massive delays and rework. They had saved time upfront, but at the cost of flexibility when it mattered most.
Final Thoughts
This is why software engineers will matter more than ever in the years to come. Especially those with mid to senior-level experience. They won’t just write code, they’ll be the ones called in to fix what was built too fast, too recklessly, and with too little thought.
AI might lead the charge, but humans will still be needed to make sense of the chaos that follows.
Top comments (0)