Why non-developers fail at building things with AI

The AI coding revolution and its limitations

You may have heard of tools like Replit, which have become really popular and allow non-coders to write and build applications by using AI to write the bulk of the code. I've met people who are building their own systems with these tools - both commercial products and internal systems, tools, and even websites. It's amazing technology, it's incredible, and it can do so much.

What I'm seeing a lot of, though, is exactly what I experience with AI coding tools. I use AI extensively in my day-to-day work - most of my code is written by AI. But I think people are under the slight misconception that AI is doing all the work while I'm just sat back sipping coffee. Unfortunately, that is not true.

What AI can't do: the hard parts of development

The hard bit of coding has always been working out what to build and how to build it. Is the thing of value? Who do we need to include? Which stakeholders are invested in this? How does it need to look? What does the UI need to do? How will this affect existing users? How are we going to market this? How are we going to put it live? How are we going to ensure application security?

All of these things AI can help with, but it can't do independently. It still needs me to decide, architect the solution, design it, and give really specific instructions. AI can do the donkey work - the coding itself.

The house-building analogy

If we use a house-building analogy, AI tools can lay the bricks at lightning speed. For any project where we need a brick, AI can do it - it's done. But building a house isn't just laying bricks. You don't really have a proper house at the end if that's all you do.

You need to:

• Lay foundations
• Get planning permission
• Connect to mains electricity
• Put the roof on
• Ensure it doesn't get damp
• Insulate it properly
• Furnish it properly
• Design it properly

All of these things go into building a house. It's not just laying the bricks. This is exactly what I'm seeing with AI coding tools and software development.

The 80-20-95-99 problem

I think there's a bit of a magic trick happening. There's a slight illusion to it, and people realise this later into their projects with tools like Replit and AI coding.

People find the first 80% of what they're trying to build goes like lightning - they're so impressed. They show their friends, they show their colleagues and peers, and think, "Okay, I can do this."

But then getting from 80% to 95% done takes a heck of a lot longer. They start running into all sorts of issues:

• Security doesn't quite work
• There are gaping logic holes
• Pages break
• Some of the UI doesn't look the same
• Things they're trying to build just aren't working properly
• Other parts of the system start falling down

If they do get to 95%, getting from 95% to 99% takes even longer than the first 95%. The work becomes arduous and needs someone technical - an engineer to look at the code and really directly analyse what's going wrong and what the solution is.

AI as a legion of interns

AI can get it right nine times out of ten, but one out of ten times it needs a technical person - not even to solve it by hand-coding themselves, but to direct the AI properly. AI is like having a legion of interns at your beck and call.

That's great - they're going to be able to do a lot of work. But with work from interns, you don't want to trust everything they do, and they're not always going to get it right. They need supervision, direction, and expertise to guide them effectively.

The disconnect between coding and software engineering

Even with the advancements of AI that have gone crazy in the last year - and I'm sure a year from now, two years from now, three years from now, it's going to get better and better - I still think there's going to be this disconnect between writing code and software engineering. They're not the same thing.

What starts out as someone building their own MVP with a tool like Replit, feeling like they're making great progress, turns into an absolute time sink and time drain. I've spoken to people where everything seemed great, and then they're up until midnight trying to get one thing working, and it's still not working. They're getting stressed, they've already sunk three weeks into it, and quite a lot of money.

It's a slight illusion and false economy.

How I use AI effectively

I use AI extensively, which means I have lower overhead, so I don't outsource as much of my work anymore. I work with a team of expert senior engineers who are specialists in their own areas that I call upon when they're needed. But for the coding donkey work, AI does the heavy lifting.

This is good for clients because their payment and investment for their product mainly goes towards my experience, management, delivery, and architecture. But they get a lower overall cost due to AI being able to take on some of the work. This efficiency is part of how I can deliver MVPs in 8 weeks.

The expertise amplification effect

The key difference is that I understand what to build and how to architect it before I even start coding. When I use AI to write code, I can:

Direct it effectively: I know what patterns to use, what libraries to integrate, and how components should interact.

Spot problems immediately: When AI generates code with issues, I can identify and fix them instantly rather than debugging for hours.

Integrate seamlessly: I understand how each piece fits into the larger system architecture.

Maintain security: I know what to look for and can ensure AI-generated code follows security best practices.

Optimise for maintenance: I can guide AI to write code that's clean, documented, and easy to modify later.

The real value of professional development

Working with someone experienced who uses AI tools effectively gives you the best of both worlds:

• The speed and efficiency of AI coding
• The expertise and oversight of professional software engineering
• Focused feature development that avoids the 80-95-99% trap
• Proper architecture that scales with your business
• Security and maintenance built in from day one

When DIY AI coding makes sense

There are scenarios where building with AI tools yourself might work:

• Simple, internal tools with minimal requirements
• Proof-of-concept projects where quality isn't critical
• Learning exercises where the journey matters more than the destination
• Projects with unlimited time and patience for debugging

But for anything commercial, anything that needs to scale, or anything that represents a significant investment of time and money, professional expertise becomes essential.

The future of AI in development

AI will continue to improve and handle more complex tasks. But the fundamental challenge remains: knowing what to build, how to architect it, and how to ensure it works reliably in production. These are human problems that require human expertise.

The developers thriving in the AI era aren't competing with AI - they're collaborating with it. They use AI to handle routine tasks while focusing on the strategic and architectural challenges that create real value.

If you've got a project in mind - whether or not you've tried building your own thing with AI tools like Replit or similar - let's have a chat. I can show you how professional expertise combined with AI tools delivers better results faster and more reliably than either approach alone.