tldr
It is to no one’s suprise, but as a developer I am using AI more than ever to complete my day to day work. And how am I doing it? Agents.
My Dumb Version of AI history
So right now I am 24, when I was in college ChatGPT just dropped and that was my introduction to generative Ai and LLM’s, probably like most people. At this point (~2022), to complete my school work I mostly would google things when I needed to research an answer to something, or go to chegg if I really needed an answer. Once ChatGPT was introduced, it was pretty insane. I could ask it questions and it would be comparable to finding a stack overflow post that was almost exactly like the problem I was facing. And for completing non-major classes like some multicultural requirement class, I would start dropping the quiz questions in ChatGPT and it was suprisingly good most of the time. So my college workflow was to have open a tab for google and a ChatGPT tab. I would interact with the ChatGPT tab when either the answer to my query was pretty simple so I could trust the answer without further investiation, or the exact opposite situation when I couldn’t find much use of the google query results and I was desperate for some sort of answer.
Eventually I began my full time job as a SWE in early 2023. AI usage in the corporate setting I was in, was not a thing. People would joke around like, “Haha yea this thing will be doing all my work in the future”. But these type of comments were mostly tongue in cheek and not taken seriously. I am not sure most people believed it when they said it at that point. Things changed quick, probably by my 4th month we had a tool where we could prompt Anthropic models, but we weren’t allowed to use the code it generated as it posed legal and security concerns. By my 4th-ish month I am using Ai to at least iterate and think on coding problems. The Ai responses were equatable to a much better google search that could generate usable code, but I’d have to do the heavy lifting of blending it in how my code base was.
Months went by. The Anthropic models got better, so prompting them in a chat interface allowed me to become more productive but I was still handwriting most of my code. But there were more moments where I could copy and paste the code from the chat interface into my ide and have it running with minimal changes. This was a huge productivity boost then.
I think almost a year passed by and I had a buddy who told me that Cursor is wickedly good. I knew of Cursor, my opinion then was that it was the “Github copilot clone” thing that was probably worse and it was in vscode still. Ignorance is bliss. My buddy said no, Cursor is now a fork of VsCode and it has something called chat mode where you tell it do stuff and then it changes code in your editor. Honestly, I was very skeptical and judgemental. My mindset was there was no way something could write all this code, it had to be poor quality and crap that I would not want in my codebase. And the people who used these tools, they must not be technical and can’t write code. But still my friend was persistent, he said just install it, its really simple if you Vscode set up and you can just try it. So I did. I started prompting and I coped with using it since I could use different models with one tool. Eventually I moved from Neovim, and Cursor was my go to Ide. I was sold. I set up vim motions and all my key bindings in Cursor and I was off running. Quickly I ran out credits on the free plan by like the second or third month of using Cursor and I officially subscribed to Cursor in March 2025. I was slowly giving Cursor more and more control to write code. At first it went from prompting in the chat mode to understand things about potential features, to eventually with the release of agent mode to having it complete full on features. By this point, I was at work using a chat interface to help me handwrite code and at home I was using an “AI editor” to write code for me. It did feel like my tools were better for my personal work. This quickly changed as more tools were introduced and available to broader corporate/enterprise audiences like Cline, Q in IntelliJ, and Q Cli.
So by the second quarter or so of 2025, AI was probably writing 30-40% of my code I produced in my editor or through a chat interface that is comparable to ChatGPT at work. And at home, AI was writing 80% of the code I pumped out in my limited time.
In the last few months, August through end of September, I’ve almost exclusively been using terminal agents. At home I am mostly using Cursor Cli and at work Q Cli. I’ve gone through some fads like prompt engineering, vibe coding, spec driven development, and now context engineering. I can somewhat confidently estimate that 80% of the code I produce at work has been produced by Ai and the other 20% is thoughtfully handwritten so Ai can copy it, extend it, and take inspiration from it in other areas of the codebase. But even this 20%, I am iterating through different ideas by chatting with a LLM and brainstorming different approaches. Or I am quickly prompting for help with syntax or making something more “functional” since I am functional program pilled.
In my next post, I will describe my workflow in detail and somethings I’ve learned along the way. I felt compelled to write my history with Ai since it has changed and improved so fast, I want to be able to look back and see what I was thinking and doing. Plus, I always tell people my anecdote that Ai was a really good google search like last year and now it does 80% of my work…I had to put this pen to paper.