## Where I started I wrote code in VS Code and switched from consulting stack overflow for answers to asking ChatGPT. I still wanted to learn all the details of the tech stack I was using. (I will read the [quick start on Gemini](https://developers.google.com/gemini-code-assist/docs/overview)). I still used code to help me understand the real world. ## Early 2025 I added Gemini's code assistant to VS Code when they released the free version for individuals. ## 2025-03-24 Let's compare Gemini and the free tier of copilot today. Let's get the project backed up in GitHub Gemini couldn't access .ipynb files and wasn't able to diagnose the problem. Instead I had to tell it that it was operating as a VS Code extension and that I accessed it from the context menu. It had no useful suggestions for me, and only when I tested a .py file did I realize that the issue was actually the file format. Next I tried copilot. This worked better in the Jupyter Notebooks. My first experience was trying to get it to write a simple compound growth calculator for a series of assets. It wanted to iterate through each year rather than use a vectorized approach. I wrote the function signature and asked it to complete it. I realized upon reading the function that I had forgot to provide a starting price. I told it where to find it and asked it to update the function. It broke halfway through and left me with half-finished code. I only had to prompt it again in the same way to get the starting price added in. It was able to get the right answer but I had to specifically tell it to use a vectorized approach for the compound growth calcualtion. My favorite command is just to clean up spacing when I need to add lines and subsequent lines don't follow! ## 2025-05-17 I spent three weeks [[guides/Vibe coding my data science portfolio|Vibe coding my data science portfolio]]. Through the process, I gained more fluency in agentic AI coding tools, downloaded Claude Code, Windsurf, and Cursor, and learned how to better incorporate copilots in my coding workflows. I also began to discern the personalities of each LLM, using Gemini more than ChatGPT but relying on Claude for copywriting.