The line between having an idea and actually building it is blurring rapidly. Modern AI chatbots – including Claude, Gemini, and ChatGPT – now allow users to create functional applications simply by describing them in detail. This process, often called “vibe coding,” is surprisingly accessible, requiring no specialized coding knowledge. The quality of the final product depends entirely on the prompts provided, making it easy for anyone to experiment… though perfection isn’t guaranteed.
This isn’t theoretical. Recently, one individual tested this capability by tasking all three AI models with building the same e-reader application, complete with advanced features like real-time text highlighting and dynamic audio-visual effects. The results were telling: the AI itself matters less than the precision of the instructions.
The Challenge: Building the “Tome Reader”
The project was born from frustration with existing e-readers, specifically Amazon’s Kindle, which lacks simultaneous read-aloud and highlighting functionality. The goal was to create a web application – dubbed the “Tome Reader” – that could read uploaded text (PDF, EPUB, or pasted content) aloud while highlighting the corresponding words in real-time. The app would also generate ambient background music based on content categories (horror, sci-fi, etc.) and trigger sound/visual effects when certain keywords were spoken. All within a single HTML file for ease of use.
The Process: Iterative Prompt Refinement
The experiment wasn’t about picking a winner from the start. Instead, the developer iteratively refined a single prompt by having each AI build the project, then generate an updated prompt based on the results.
First, Gemini created a functional prototype. Then, that project was used to generate a refined prompt, which was fed into Claude. Claude further improved the app, but also introduced unexpected logic – limiting trigger effects to once per sentence to avoid “spamming” the user. Finally, ChatGPT received the latest prompt and built the app, though it struggled with adding a dedicated volume slider.
The Results: All Models Capable, But Not Without Quirks
All three chatbots ultimately succeeded in creating a working version of the Tome Reader. However, the process wasn’t seamless. ChatGPT was the slowest. Claude exhibited unpredictable behavior (requiring 11 rebuilds to resolve loading errors in one instance). Gemini was generally the most reliable, but also the least user-friendly in terms of file delivery.
The key takeaway? The prompt itself is paramount. Performance differences between the free and paid versions of these models were negligible. A well-crafted set of instructions can yield impressive results regardless of the AI used.
The Future of App Development?
This experiment demonstrates that AI is rapidly evolving into a viable tool for rapid prototyping and even functional app development. While not a replacement for skilled programmers, these chatbots lower the barrier to entry significantly. The ability to iterate on ideas with minimal coding experience could accelerate innovation and empower individuals to bring their visions to life without relying on traditional development pipelines.
The future of software creation may well be conversational, with AI acting as a collaborative partner in the design and building process.
