How Good Are AIs for helping in my day to day work?

To begin, I would like to discuss ChatGPT. When I initially began writing this post, I provided it with the title of the proposed article, "How Good Are AI for Software Development,"

I expected it to start asking me questions about what I wanted to write but instead it generated a 500-word essay that contained entirely fabricated content. It presented my thoughts on the three AI tools I had been testing: Roo Code, Gemini, and Cursor. However, the generated information was unrelated to my actual experiences with these tools.

Here is what ChatGPT generated for each AI:

1. RooCode: Ambitious but Inconsistent
2. Gemini: Helpful, but shallow in context
3. Cursor: A Hidden Gem for Refactoring and Debugging

As someone who has extensive experience writing CMS systems, I decided to utilize AI to develop a new CMS that leverages my years of expertise in creating websites for companies ranging from simple web pages to enterprise-level websites.

Let's begin with RooCode. I thoroughly enjoyed using this tool and spent approximately five hours building a code base in vanilla JavaScript that would have taken a week to complete. I had a demo-able version of the CMS that could render multiple mock pages. However, when I asked it to convert the code base to TypeScript, it failed to upgrade the project to TypeScript. If I had initially started in TypeScript, the issue would likely have been resolved.

I subsequently switched to Gemini, which successfully modified the project to TypeScript and, within five to six hours, had the code mostly working using TypeScript. However, it continued to cycle through tasks, likely due to certain technical challenges.

During this period, I took a break and conducted some research. While researching, I came across a video about Cursor. I had previously heard positive feedback from other developers and had always intended to test it.

Upon returning to Gemini, I asked it to compose a specification document for the project that I could provide to another AI. I then gave this specification document to Cursor and built a new version of the CMS.

At this juncture, I had constructed the same code employing various AI tools. I highly recommend any of these tools to enhance your work, as they proved to be exceptionally effective. The agents exhibited remarkable capabilities, enabling them to edit files, relocate them, and generally execute the majority of the modifications I required. In my opinion, the productivity boost derived from utilizing these tools is approximately tenfold. Furthermore, adopting a consistent tool would likely enhance my ability to articulate the precise changes I need to make.

Since I spent the most time with Cursor, some the annoying aspects were when I asked it to create a database to store some of the data it created the Primary key as a VARCHAR(36) instead of a UNSIGNED BIGINT. the interesting thing was when I queried the AI about the best way to store a primary key it wanted to then change it to a BINARY and it wrote two functions that would change the GUID it wanted to use into a numerical value, Cursor has a .cursorrc file that allows you to specify some rules that it will use when generating solutions. which I eventually found when it kept making the same mistakes.

When I asked it to generate documentation on the project it wrote a lot of information, most of it wrong it talked about directories that did not exist.

Summary

Our digital age is changing it's time to get on board. A year ago I would have and did describe these tools as drunk students. Now they are Junior Programmers, that you need to keep an eye on. but they will only get better.

Many of these lessons now shape how we build production platforms like PrimeCRM, where AI accelerates delivery while architectural decisions remain tightly governed.

PrimeCRM