I Might Stop Programming, Even Though I Love It
I have been programming intensely since I was around 14 and have absolutely loved every moment of it. However, I’ve also been an avid follower of the development of large language models (LLMs) and have actively used them since GPT-2.
Over the years, my approach to programming has evolved significantly. Initially, I programmed everything manually, as everyone did before the advent of GPTs. Later, I began using GPT for code completion and smaller functions. Nowadays, I seldom write any code myself, whether professionally or privately, even though programming is both my full-time job and my passion.
I believe I am now entering a fourth phase. I am relatively active on my GitHub, creating smaller or larger projects from time to time. Many of these projects are relatively simple in principle. While coding them is not difficult, it might take a day or two to get everything right. This has led me to question why I should write code at all when I could simply write a “Project Specification Prompt” instead, like the following example:
### Project Goal
Create a webpage with a centered grid of input squares for distributing values between friendships and intimately satisfying experiences.### Specifications
- Total value ≤ 25.
- Non-negative integers only.
- Adjust values directly or with +/- buttons.
- Invalid inputs show a red box and border.
- Higher values = greener; lower values = redder; real-time updates.
…
At present, the results are mixed: some aspects succeed while others fail. However, each new GPT model has gradually improved the outcomes. I see no reason why this approach shouldn’t work perfectly in the coming years.
Does this mean that I don’t enjoy programming? Not at all. I still enjoy it a great deal and continue to engage in it frequently. However, I recognize that my time on this Earth is limited. When I reflect on the value my projects might bring to myself and potentially to others, I realize that the underlying code is not as important. It currently enables me to complete projects, but it is not inherently difficult, it just takes time. The real challenge for me lies in generating meaningful project ideas.
The underlying architecture, frameworks, libraries, or even the programming language chosen are not as significant as long as they are selected wisely. I think that LLMs will be able to skillfully and professionally create apps and projects from descriptions increasingly well, and for increasingly complex projects.
With this in mind, it’s likely that I will transition towards a paradigm where I function more like a project manager, writing detailed specifications that future LLMs can execute.