I Built One Project in a Year and Another in Two Months: What Changed With AI-Assisted Development

Using artificial intelligence to speed up software development has become a hot topic. In this article, I explore how my own experience developing two projects, one taking a year to complete and the other just two months, led me to view AI-assisted coding tools in a new light.

The first project, EngHub, took me a year to finish, with the assistance of GPT-5 throughout the development process. My second project, AlbumMap, a map-based video platform, was completed in just two months using GPT-5.5. The stark contrast in development time between these two projects was striking, and it made me question how much of the difference could be attributed to the newer AI model.

While GPT-5.5 certainly made me feel like I could move from idea to implementation more quickly, I'm hesitant to say it made me six times faster. The two projects were fundamentally different, built at different times with distinct requirements. They weren't controlled experiments, so I can't draw definitive conclusions about the impact of AI on development speed.

What I did learn, however, is that AI coding tools are becoming increasingly adept at bridging the gap between an idea and a working product. There are numerous factors that influence the length of a software project, including scope, developer experience, existing knowledge, available time, and the definition of what constitutes a finished project. Simply looking at the time it takes to complete a project doesn't tell the whole story.

One key insight I gained is that evaluating AI coding assistants based solely on the code they generate can be misleading. While the generated code may appear clean and functional at first glance, it doesn't guarantee that the feature will work as intended. My current project, AlbumMap, is an excellent example of why end-to-end behavior matters more than just the quality of the generated code.

Imagine a scenario where an organizer receives two video contributions. They approve the first one but reject the second. At first glance, everything seems to be working correctly - the approval and rejection buttons change state, and the changes are saved. However, the true test lies in the final video generation. Did the approved contribution make it into the final product? Was the rejected contribution excluded as expected?

This brings me to another important realization: testing the workflow, not just the screen, is crucial when working with AI-assisted development. A more thorough test would verify if the approval status survives a browser refresh, if the selected order remains unchanged after changes, if rejected contributions are still excluded, and if the preview and final export match the user's intentions.

This distinction becomes even more critical when dealing with AI-generated code. While AI can produce an implementation rapidly, the speed increase highlights the importance of having a clear way to determine whether the implementation is truly correct. Defining success criteria before asking an AI to build a feature is essential.

For instance, instead of saying "Make the contribution editor work," a better specification would be "Build the contribution review flow so that only approved contributions appear in the preview and final export. Preserve the user's selected order after refreshing the page. If saving fails, show a clear error and do not display the change as successfully saved." By defining measurable conditions, we can test the implementation more effectively and identify any issues more easily.

While AI can indeed help speed up implementation, it does not replace the need for product decisions. The developer still plays a crucial role in determining what the feature should actually do, how it should handle failures, and what should persist in the final product.

In conclusion, my experience building EngHub and AlbumMap has taught me that AI coding assistants are becoming incredibly powerful tools for developers. They can significantly reduce the time spent on implementation, allowing developers to focus on higher-level design decisions. However, it's essential to remember that AI is not a silver bullet and should be used judiciously, with clear success criteria and a focus on end-to-end functionality.

As AI continues to advance, it's exciting to think about how it will reshape the software development landscape, but we must approach it with a critical eye and a strong foundation in sound product design.