Building software with an AI team: how I built a working web app with Claude & ChatGPT
Prelude
I’m a professional techie, but no longer a professional developer — my badge and git permissions were handed in over a decade ago. This post reflects my experience using Generative AI tools to build a working web application from the perspective of someone who is extremely rusty on the day-to-day tasks of being a software engineer. However, I am writing this because I found the experience of using off-the-shelf Large Language Models (LLMs) was transformative for me; I’ve written more working applications in the last 6 months than over the last 6 years, and I’ve really enjoyed doing it. This piece chronicles the development of the most complex of those applications, a vanilla Javascript, HTML and CSS client side application for visual concept mapping. I’m going to cover what I learned about the process of working within a team of GenAI agents to build a simple web application.
Why build an application with GenAI anyway?
I’m writing a book. My co-author, Andrew, and I are currently about 80,000 words (or 400 pages) of content into the manuscript. Ahead of one of our weekly calls I decided that I wanted to sketch out the current structure visually. If I had a physical white board, it would be full of Post-it…