After many years of excuses, in January 2025 I decided it was finally time to learn how to code. As I did research to find the best online resources for learning, I was confronted with many articles claiming “Coding is dead,” or “ChatGPT is here to take your job.” These warnings delayed my decision to start by another couple weeks, but eventually I decided I had nothing better to do.
I was instantly hooked. The ability to manipulate machines that have become so ubiquitous in my life was empowering. My engineering brain was enamored by the structure and logic. It was an exploration of a world that I previously knew so little about. After several months of learning the basics, I decided it was time to test out the latest tools and see what all the hype was about.
I fondly remember my first agentic experience. I provided a prompt to Cursor, and then a magical entity took control of my laptop and began writing code. I opened my browser to see if it worked, and everything appeared to be running smoothly. Once I got the hang of it, I continued to iterate at a rapid pace.
After an exhilarating three days of work, these were the final stats:
- 5,000 new lines of code
- 2 git commits
- 0 unit tests
If you’re familiar with programming, you will realize the agent was not exactly following software engineering best practices. When I finally sat down to analyze the outputs, I realized most of the codebase was what the experts refer to as slop: 1,500 line files, duplicated functions, and many unused variables and imports. The agent created a lot of cool new features within those three days, but I had to spend the next three cleaning up its mess.
This is not to say that AI tools are useless. They just require some wisdom to be used effectively.
In the 1990s, an inexperienced programmer would have had to spend hours parsing through a book on HTML just to get “Hello, World!” to appear inside a URL. Now, a 12-year old with a knack for technology can produce a full stack application in an afternoon. But what that 12-year old doesn’t have is decades of trial and error solving complex problems in the real world.
The advantage of combining domain expertise with artificial intelligence is not limited to software engineering. A graphic artist with great taste can rapidly produce images in Midjourney that still maintain their distinct style. All of the minute details are abstracted away. Now, one's mental energy can be focused on the bigger picture.
I created the prototype for this website in ten minutes without writing a single line of code. I have subsequently spent dozens of hours meticulously crafting the contents to match the aesthetic I initially envisioned. Claude Code allowed me to rapidly iterate between the various options, but the final result was achieved through hours of careful consideration.
I think it's important to set boundaries on the cognitive load I’m willing to hand off. I try to limit my use of AI during the creative writing process. I will allow it to advise me on the structure of my arguments, or to check my grammar, but first drafts are always written without assistance. If I allow the model to dictate my style, then my writing will begin to sound like millions of other people using the same tools.
In an age of unlimited content creation, distinct style will stand out from the rest.
If the work you do can be accomplished by chaining together words from the contents of the internet, you might be in trouble. But if you’re working on something that can only be accomplished by someone with your unique lived experience, the future is full of opportunity.