You’re not a real developer until you have countless unfinished personal projects and even more half-baked ideas that never make it past the brainstorming stage. Imagine all the million-dollar ideas tucked safely away in Moleskine notebooks around the world…

Or is that just me?

This has been my reality since the days when PHP powered the majority of the web.

Looking at past projects, my attention typically lasts only a month or two, and if building an MVP (Minimum Viable Product) takes longer than that, it’s likely destined to live in a private GitHub repo forever. All that changed once I started using AI-powered development tools.

ChatGPT

I’ve been using ChatGPT since January 2023 to assist me in development. When I got stuck working on personal projects, I’d copy/paste the problematic code or file, describe the wider context, and see if it could unblock me. It was very slow with a lot of back and forth, and even then, I’d still resort to reading documentation regularly.

Cursor

With all the AI buzz going around, I kept hearing about Cursor, so at the beginning of 2025 I tested it out by building a feature-rich bookmarking app in a weekend. Since it’s built from VS Code (which I was already using as my preferred IDE), I was immediately familiar with the UI, and all my extensions work seamlessly.

The second thing I tackled was rewriting a project I had started 3 years ago but never finished. I had originally worked on Telestrator for 2 months off and on, but with Cursor I was able to complete it in less than 5 days. While I had some core logic to reference, I not only surpassed my previous work but also modernized it with current React tools.

After getting the hang of how everything worked, I started tackling various annoyances by building small web apps and scripts to make my life easier, ranging from a scheduling tool for finding times to meet with HOA board members to a podcast feed filtering app (FeedFilter).

With each of these projects, I chose to experiment with new technologies and frameworks to expand my knowledge. For instance, I’d been reading the docs for Tanstack Start, and since FeedFilter is a SPA, I thought it was the perfect project for a crash course on the framework.

I also used AI to validate the concept and write the initial code behind Search 2.0. While the Telestrator project was pretty straightforward, working on this existing codebase felt more like pair programming. I referenced the Upstash Search documentation and outlined how I planned on structuring the data for searching, then worked iteratively with AI to implement it.

Antigravity

Mid-2025, the Cursor team upset a lot of paid developers with their change in pricing. The confusion around the new pricing model and messaging caused an outcry from developers. While I didn’t experience any issues personally, it made me consider evaluating other AI IDE alternatives. With Google’s strong showing with Gemini 3 Pro, I decided to give Antigravity a try. It’s too soon to have any valid opinions yet, but I’ll write about it if I decide to switch from Cursor as my main IDE.

GitHub Copilot

Also at the beginning of 2025, Microsoft made using AI to improve development and operational efficiency mandatory and part of our yearly review cycle goals. So for work, I use GitHub Copilot with Visual Studio 2022 and VS Code. Even though it’s the only agent I can use, one nice thing is I get to choose which AI model I’m working with. Assuming my experience isn’t too skewed due to enterprise restrictions, I’m still left unimpressed. At times, it feels like I’m using an outdated version of Cursor. Regardless, I primarily use Copilot for prototyping, pair programming, validating code I’ve written, improving test logic, and creating test harnesses.

The Evolution of My Development Process

Looking back, as AI models and tools improved and my prompting became more precise, my development process drastically evolved beyond those early ChatGPT days.

I transitioned from using ChatGPT like a Stack Overflow replacement to using tools like v0 as a starting point. I’d give it a general idea of what I wanted and let it choose everything: the tech stack, design framework, and overall architecture. My focus was simply getting it up and running, then importing the project into Cursor to continue development from there. But once I got more comfortable with Cursor, I realized I could streamline things.

Now my process looks like this:

1. Start with an empty directory
2. Include any secrets/tokens/api keys I want to use in a .env file
3. Explain what I want the app to do
4. Outline the frontend design and backend architecture
5. Wireframe the UI using something like Excalidraw to visualize how I want everything organized
6. Use Cursor’s plan mode to verify everything is as I expect it, tweaking things if necessary
7. Use agent mode and let it create everything

I’m constantly evolving my process to include the latest AI breakthroughs while ensuring it remains optimal for achieving my goals and challenging myself in new areas. For example, I’m currently using AI to port a Node.js backend to Go. As I learn Figma, I’ll also be utilizing the MCP (Model Context Protocol) to see if AI can create the Astro template for the next design iteration of this website.

The irony isn’t lost on me that AI, which many feared would replace developers, has instead become the tool that helps me actually finish the projects I start. Those half-baked ideas in my Moleskine notebooks are finally seeing the light of day. The best tools, when used properly, amplify what we’re capable of creating.

The future of AI is not about replacing humans, it’s about augmenting human capabilities.

– Sundar Pichai