I’ve been curious for a while about the new wave of AI-powered coding tools. There’s been so much buzz about them: promises of building apps in minutes, generating clean code from natural language, and turning ideas into products faster than ever. Plus the overarching question ”is vibe coding removing the need of software developers all together?”

‍

So, I decided to take a proper look, and rather than starting with a hypothetical project, I thought I’d put these tools to the test on something real: a reporting application we’ve been talking about building for over two years. This internal app would centralise data from our different systems project management, communication and billing, all in one place so we could get vital business information instantly. We’d even started a build once, but never finished. At the time, we estimated it would take about 6 weeks to complete. If AI coding tools could meaningfully speed that up, I wanted to see it first-hand.

‍

Starting in Google AI Studio

I began in Google AI Studio (powered by Gemini). Within about 30 minutes, I had a clean front end with all the features I wanted. It was exactly the kind of “AI magic” I’d hoped to see fast, accurate, and well-structured. And I’d done this by simply prompting the AI just like I was using ChatGPT or Copilot. 

But when I tried to connect the backend, I hit my first friction point. The workflow required exporting the repo to VS Code or Cursor to proceed, which broke the seamless, in-browser, native language experience I was aiming for.

‍

That’s when I decided to pivot to another AI tool.

‍

Switching to Replit

I moved the project into Replit, which is designed for coding directly in the browser. It manages everything from Frontend, Backend, Databases and Deployment. 

Rather than manually rebuilding everything from scratch, I asked Google AI Studio to write me a detailed prompt that would replicate my frontend inside Replit. 

It worked perfectly. In less than 60 seconds, my UI was rebuilt inside Replit, ready for me to continue development. It did bring across a few un-needed Gemini code packages which I kindly asked to be removed. 

For those unfamiliar, Replit’s layout is clean and simple:

- Left: Code editor

- Middle: Chat panel for natural language prompts

- Right: Live web preview

Replit is primarily for web applications (though I’m still exploring Bolt for mobile apps), and its “AI pair programmer” vibe is exactly what I was looking for.

‍

Building the product

When I started adding functionality, I noticed Replit had set the project up as a mono-repo meaning the frontend and backend lived in the same repository. I hadn’t asked for it specifically, but it’s worth noting for anyone using the tool, as it may shape how you structure your code.

From there, I integrated two external APIs from our existing software stack. This turned out to be the most time-consuming part, mainly due to configuring the correct access scopes.

In just two hours, I had:

- A deployable internal product

- Secure authentication

- Live data from two external APIs

- A clean, functional interface

We’re rolling it out to the team on Monday, something that would have taken at least a month in a traditional workflow. A pretty incredible result and an awesome example of how powerful AI code tools can become. 

‍

Lessons learned on AI coding

Plan before prompting:

Similar to maximising your ChatGPT, Gemini or CoPilot experience, the clearer you are about your features and workflows before you start, the better your AI output will be.

‍

Go one step at a time: 

Rather than trying to build large MASTER prompts, In Replit, I built features one prompt at a time rather than sending giant requests. It made it much easier to track changes and test functionality.

‍

AI isn’t replacing developers:

AI coding tools can generate a huge amount of boilerplate and even working features in minutes, but I found in my experience that there is still a need for a developer:

- Complex business logic — AI can write code, but translating deep business knowledge into robust, scalable systems still requires human understanding and experience. The most time I invested in this project was figuring out how to word my prompt correctly to maximise the likelihood of the functionality being built correctly. Prompt engineering will continue to be one of the most important aspects of all AI usage.

- System architecture — Designing a reliable, secure, and maintainable system isn’t just about writing code; it’s about making strategic decisions that AI can’t fully own.

- UI/UX Design: When using both AI Studio & Replit, I found the ability to customise the design and interface a little finicky. I went into the project with no idea of what the interface should look like and found making adjustments on the fly wasn’t super efficient. I really see the power in having a UI/UX designer creating an initial concept for AI to analyse then base its build off. Without this, things do feel a little templated or ‘stock’.

- Deployments: Replit offers a Deployment as a service solution meaning you’re deploying from within the Replit environment. This means you have limitations on your own server environment e.g. Server Region, Additional Security standards etc. If you want to get outside of the Replit ecosystem a developer would likely need to set up your bespoke cloud infrastructure. 

For me, the takeaway is that AI dramatically accelerates the build phase, but it’s developers who ensure the result is reliable, secure, and fit for purpose. Right now, I see Replit as the perfect tool for building either internal apps or MVPs where the stakes aren’t too high. 

‍

Final Thoughts: 

Before I deployed my Replit App, I pushed the code to GitHub and had one of our senior engineers review it. They were pleasantly surprised the code met standard security expectations, including:

- No hard-coded credentials

- Proper authentication and role-based access

- Input validation to prevent injection attacks

- HTTPS-only API calls

- No sensitive error messages exposed

- Dependencies without known vulnerabilities

So clearly the code standards are there. Now it’s just seeing how these tools continue to balance complex technical needs such as custom cloud infrastructure and perhaps even complex relational database creation. 

My goal was to explore AI-powered coding tools and see how far they’ve come. Replit, combined with Google AI Studio for the initial setup, delivered a complete internal tool in two hours that would otherwise have been a months-long project.

While there are still limitations (especially in more complex workflows), the speed, functionality, and flexibility I experienced make it clear: AI-assisted coding isn’t just hype, it's a genuine productivity booster. When we’re looking to build our next internal business tool, we’ll be sure to use an AI code gen tool as the first port of call. 

‍