Why AI won't replace developers—but it has changed the industry

My journey back into development—thanks to AI and a few mimosas

I originally learned the core fundamentals of development back in 2002 in college. However, after graduating, I drifted away from coding and didn't touch it again until about two years ago. One day, after I had a couple of mimosas, I had an idea for an app. Having recently played around with ChatGPT, I felt bold enough to dive in. 
I only had experience with HTML and CSS.  So when I asked the AI which language I should use, it recommended Python.  There was no looking back from there. After a week of working with AI, I had a functioning app running locally on my machine. 

But that’s when both the AI and I hit a wall: deploying it to a production environment. Neither of us could figure it out, and it took the help of a few developer friends to finally get the app live…at which point it quickly broke. It couldn’t scale.  It didn’t have an access limitation.  It didn’t even have a database and instead used the browser cache to store everything.  It was only an app in the simplest of terms.

Since then, I’ve grown in my abilities to use AI to create web applications.  I’ve built several other apps using Flask, Django, Laravel, React, Next.js, Hugo and even Go—all without deeply learning any of those languages/frameworks. Over time, I graduated from copy-pasting in and out of ChatGPT to using more advanced tools like Copilot and Cursor and even experimenting with local AI models on my computer.

But here’s the biggest lesson I’ve learned from this experience: AI is still just that—artificial intelligence. The only reason I’ve been able to create the apps I have is because of the help of my extremely experienced developer friends and the problem-solving foundation I built back in college. My understanding of core programming concepts like functions, variables, loops, data structures, control flow, algorithms, operations, and networking allows me to give the AI precise instructions. When I don’t guide it carefully, the AI can go completely off the rails, producing either something overly complicated or something far too simple.

I still have to hold the AI’s hand every step of the way.

AI is not coming for every developers' job—just the bad ones

AI development tools are evolving at a rapid pace. Today, platforms that integrate AI agents promise the ability to build fully functional apps without writing a single line of code. And from experience, I can say—yes, it's possible. But only with a lot of human oversight.

This is the paradox of AI in software development: while AI makes development faster and more efficient, it still relies heavily on human intelligence to guide the process. Developers aren’t going away anytime soon—but the role of developers is changing, and in some cases, those changes will mean the end of the road for certain types of developers.

Let’s dive into why this is happening.
 

Building with AI still requires real intelligence

No-code app-building with AI is often marketed as a hands-free experience. The reality is far from it. Even with AI agents designed to automate tasks, I still have to build my apps piece by piece, steering the AI every step of the way. For each challenge I encounter, I can’t rely on the AI to "magically" solve the problem. Instead, I have to figure out the solution and then articulate my instructions clearly.

This resonates with my own experience. When I first deployed that Python app, the AI wasn’t capable of managing the production launch without outside help. While it excels at generating code and automating tasks, AI lacks creativity and problem-solving skills. 

It can’t understand context the way humans do. AI works best when given clear directions—but coming up with those directions still requires a developer with deep domain knowledge.

This reveals one of AI's key limitations: it's only as good as the person guiding it.

Here’s an example for the skeptics at the back of the room

A recent experience I had with AI development involved creating my first command-line interface (CLI) app in Go—another language I had never used before. The goal of the app was to help migrate a Hugo site from one theme to another, where the syntax and shortcodes were completely different.

One particular conversion involved taking guide buttons, which were handled by a shortcode at the bottom of pages, and moving that information to two fields, prev: and next:, in the front matter.

If you listen to some AI influencers, you might think you could solve this problem with a simple prompt: "Create a conversion that changes the guide button shortcodes to prev: and next: fields in the front matter." Easy peasy lemon squeezy, right?

That’s exactly what I tried, because sometimes the simplest prompt is the best. You don’t always need to spend hours on complex prompt engineering to get results. And when things break, I’ve learned a lot through the process of "breaking things fast." But in this case, it didn’t work out that easily.

I had to guide the AI step by step through building out the code. The process looked something like this:

1. Create a function to extract the next page route from the guide button shortcode and store it.
2. Create another function to add the next: field to the front matter with that route.
3. Develop a function to extract the base route for the directory.
4. Write a function to determine the weight of all pages in the directory, allowing the app to figure out which page comes before the current one. (The previous guide button had relied on the browser's back button, which was not a good solution.)
5. Add a function to insert the prev: field with the determined route.
6. Finally, create a function to remove the old guide button shortcode from the page content.

Without clear, iterative guidance, the AI couldn’t produce a fully functional solution. It would consistently miss pieces to the puzzle or produce code that didn’t work. It required me to break down the task, define the logic, and ensure each function did what it was supposed to.

No-code app-building with AI is often marketed as a hands-free experience. The reality is far from it. Even with AI agents designed to automate tasks, I still have to build my apps piece by piece, steering the AI every step of the way. For each challenge I encounter, I can’t rely on the AI to "magically" solve the problem. Most of the time, I have to figure out the solution and then articulate my instructions clearly.

What AI will actually replace: junior and bad developers

If you’re worried about AI "stealing" developer jobs, you’re not wrong. But here’s the truth: AI won’t replace all developers—just inexperienced or underperforming ones.

Junior developers, especially those with no industry experience, are at risk. Companies may prefer to use AI as a cheaper alternative for tasks that would normally be handled by an entry-level developer. Similarly, bad developers—those who struggle with problem-solving and critical thinking—won’t be able to keep up in a world where AI handles routine coding tasks. 

Problem-solving, not coding ability, is becoming the most valuable skill in software development.

My own success with AI-driven development is proof of this. Because of my foundational knowledge in development, I can troubleshoot, guide the AI, and avoid pitfalls. Without those problem-solving skills, I would have been stuck at step one.

Lastly, developers who refuse to embrace AI may face challenges too. Like any other major technology shift, those who adapt to AI tools early will have a competitive edge. Developers who ignore these tools risk falling behind their peers who use AI to accelerate their productivity.

Developers see past the AI hype

There's a reason so many developers are skeptical of AI. We've all seen the flashy demos of AI generating apps in seconds, but those demos are deceptive. They're designed to show off AI’s potential, not its limitations.

In reality, the fully developed AI-hype-train-generated apps are usually simplistic and disconnected from real-world development. They lack the scalability, security, and maintainability required for production-ready software. AI can assist with certain aspects of development—like generating code or automating tests—but without a developer to direct it, the end result is often a fragile, half-baked solution.

Developers understand this because they know how complex production software really is. Developers aren’t just creating simple todo apps or landing pages which sometimes tends to be the example used for AI development. Most developers work on projects for months or even years building it one function at a time.  Each function and variable is a complex web of multiple variables that involves multiple people working on it from various angles.  

Building an app isn't just about generating code; it's about designing reliable customer experiences, on scalable architecture, handling edge cases, ensuring security, managing integrations, and maintaining performance.  Managing all these variables and their relationships falls outside of the capabilities of current AI models (yes even DeepSeek and OpenAi o1) These management of so many variables is where AI still falls short for the time being.

Advice for business leaders: Don't fall for the hype

If you're a business leader contemplating whether AI can replace your development team, you need to take a step back and avoid getting swept up by the hype. AI can certainly assist with development tasks, but it's not capable of fully managing the end-to-end process of creating robust, production-ready software.

If app development is a core function of your business, you need skilled developers to steer the ship. AI lacks the creative problem-solving and complete contextual understanding that human developers bring to the table. 

On the other hand, if all you need is a basic, static website, AI can probably handle that. But don't expect anything groundbreaking. The result will likely be a cookie-cutter solution—like the cheap, identical homes in a property development. Functional, sure. But generic, uninspired, and easy to spot as a low-effort build.  It will look like every other site on the internet, because that is where the AI learned to make those kinds of websites.

Businesses that depend on high-quality, differentiated applications can’t afford to gamble on AI alone. Instead, AI should be seen as a tool to enhance your developers’ productivity, not as a complete replacement.

AI supplements development—it doesn’t replace it

For developers who embrace AI, the future looks promising. By automating repetitive tasks, AI allows us to focus on high-value work like solving complex problems and improving user experience. It enhances productivity but doesn’t eliminate the need for technical expertise. The best developers will use AI as a tool, not a crutch (like I do).

On the other hand, developers who lack problem-solving skills—or refuse to work with AI—may find themselves phased out. This isn’t a bad thing for the industry. AI will help elevate software development by removing inefficiencies and allowing talented developers to focus on innovation.

Conclusion: Adapt or be replaced

The next time you see a demo of AI creating an app, take it with a grain of salt. AI isn’t going to take over the entire development process anytime soon. Real intelligence is still required to direct artificial intelligence.

AI is changing the way we develop software, but it isn’t the end of the developer as we know it. The developers who succeed in this new era will be those who understand how to use AI effectively, focusing on problem-solving and strategic decision-making. Those who fail to adapt—or who never had strong development skills to begin with—may find themselves replaced by the very technology they fear.

As developers, our role is evolving. But with the right mindset and skills, we can harness AI to build better, faster, and more scalable applications than ever before.

Useful links

• 5 reasons why I'm building our RAG application on Upsun
• AI deployment strategies: balancing efficiency and environmental impact