Why AI still makes coding mistakes

A few years ago, the idea of artificial intelligence (AI) creating fully functional code sounded like science fiction. Today, it’s happening at scale. Tools like GitHub Copilot, Amazon CodeWhisperer, and our very own Hostinger Horizons are transforming how websites, web apps, and online software come to life.

A few sentences describing your idea are all it takes to get a functioning prototype. Chat with AI a bit more, and your project is ready to accept members and collect payments, becoming your new side hustle or even a business.

However, those who have tried these tools know they are not flawless, and it’s smart to keep an eye on them. Why does AI make coding mistakes, how often do they happen, and how to deal with them? Let’s explore.

Blame the LLMs

At the heart of each AI coding tool is a large language model (LLM). It has read billions of lines of code, learned the patterns, and uses that knowledge to guess what should come next.

Notice I said "guess" – because that’s exactly what it does.

These models don’t truly “understand” code. Instead, they estimate likely code completions, make predictions based on past examples, and sometimes just make things up – this is called hallucination. Plus, they don’t always see the full context of a project.

These issues lead to common coding errors like missing or incorrect code blocks, wrong logic, or non-existent functions or libraries.

In a recent study¹, different AI models made different errors on the same task. This isn't a glitch – it’s how these systems work.

The good news: AI is getting better

There’s reason to be optimistic: LLMs are improving rapidly. According to researchers at Vanderbilt University, newer models like GPT-4 (2023) now generate correct code on the first try nearly 86% of the time. That’s a huge improvement compared to older models that often failed entirely.

What’s driving this progress? Better training data, automated testing during code generation, and smarter filters to catch obvious errors, to name a few.

Still, this is a journey. AI won’t always write perfect, production-ready code in a week or a month. Mistakes will happen, but that doesn't mean there is no point in using AI tools. On the contrary, they are more useful than ever – and once mastered, the benefits are obvious.

Working with – not against – AI coding tools

So, what’s the smart way to use vibe coding tools? Treat them like junior developers: incredibly fast, occasionally brilliant, but needing oversight.

Test the result. If it makes a mistake, a good first step is to just ask it to fix the problem. With that philosophy in mind, we’ve built automatic, free common error fixing into Hostinger Horizons. When an error is detected, Horizons attempts to fix it – resolving about 80% of issues automatically, without any user action.

Here are a few more practical tips I recommend in case errors still happen:

• Ask for a fix. Just copy the error or describe what’s wrong and ask AI to fix the issue.
• Try a new approach. Revert to an earlier, error-free version of the code and update the prompt.
• Be specific. Write clear, detailed instructions. Break tasks into smaller steps, give examples, and add restrictions when prompting.
• Test often. After each change, make sure everything works as intended. A quick test will catch many bugs and logic errors before they sneak into production.
• Ask for a rewrite. If AI can’t fix the error, ask it to explain or simplify the code. Rewriting code is great for cleanup and may reduce bugs.
• Review the code. If the AI-generated code is visible and understandable, take a moment to look it over, especially when it involves data, logic, or anything related to security.

AI is your co-pilot, not the pilot (yet)

The future of software development is undeniably AI-powered, and AI is already changing how we work, think, and build. But even the most advanced AI is still a tool – one that needs thoughtful use, not blind trust. Human oversight is still needed.

Use AI to boost your productivity and creativity. With the right approach and tools, you can create anything you want – faster, smarter, and more securely. Even if your co-pilot sometimes gets a bit ahead of itself.

‍

About Dainius Kavoliūnas:

Dainius Kavoliūnas is the Head of Hostinger Horizons. Dainius has nearly 10 years of experience leading global tech brands. His expertise lies in launching and scaling new tech products. He specializes in rapid go-to-market strategies, product growth, and business scaling.

About Hostinger:

Hostinger provides products and services that help clients achieve their online goals using world-class tools for ecommerce, blogs, portfolio websites, and more. Trusted by over 3.5 million clients in more than 150 countries, Hostinger is one of the TOP 3 web hosting brands worldwide and one of Europe's fastest-growing companies. Hostinger’s team consists of around 900 curious and high-spirited professionals.

‍