The Trials and Tribulations of an AI-Assisted Developer
This article chronicles the humorous, and at times frustrating, experience of a developer attempting to leverage AI tools like Cursor to build a simple To-Do list application. It's a rollercoaster of hope, despair, and the realization that AI might not always be the silver bullet it's made out to be.
Diving In: High Hopes and Initial Setbacks
Initially, the developer seeks to offload some of the work to AI. The hope is that by providing clear instructions and guidelines, the AI can effectively assist in the development process. However, the initial results are far from perfect, leading to frustration.
- The developer explicitly instructs the AI to focus solely on the login feature.
- Past experiences and approaches seem to be discarded quickly, with the plea "try a completely novel approach".
The Allure of AI: A Brief Moment of Optimism
After some initial skepticism, a glimmer of hope emerges. The developer considers using Cursor but balks at the price, stating "$20 for a month, that is insane. i could hire a developer for half an hour". The developer, however, decides to pursue the AI route after all.
The To-Do List App: A Case Study in AI-Driven Development
The task is simple: build a To-Do list app. The developer provides specific instructions and stylistic preferences, initially using Node.js before reverting to other more preferred frameworks.
- "Good, really you're gonna use Node.js for this."
- "No Buffer overruns? No memory leaks? Maybe AI does have a future."
- "What did I learn C for?"
Navigating AI Quirks and Unwanted "Features"
The developer soon discovers that working with AI requires constant oversight and correction. The AI often introduces unexpected elements and deviates from the specified requirements, leading to a series of "no, no" responses.
- The AI unexpectedly rewrites code to Lisp.
- The UI section is assigned as is, while the rest of the code is assigned to be written with HolyC.
- Blockchain integration emerges out of nowhere: "what? did i say blockchain?"
Style Guides and Hallucinations: The Pitfalls of AI Collaboration
Maintaining consistency and adhering to coding standards becomes a major challenge. The AI seems to have a mind of its own, often ignoring the provided style guide and even "hallucinating" code elements.
- "Do not change the color scheme."
- "I swear i hear clear:both one more time."
- The developer gave the AI 23 files of .cursorrules
Regression and the Call of Familiar Tools
Frustrated with the AI's unpredictable behavior and tendency to introduce errors, the developer eventually returns to familiar tools like VIM, seeing it as a "happy place."
- "That's it, i'm going back to VIM"
- "$ open vim"
- "Ah my happy place"
Back to Basics: Debugging with AI and Exploring Optimization
Even while relying on VIM, the developer continues to seek assistance from the AI, now focusing on debugging and optimization, which reveals some shortcomings of the AI itself.
- "The way you would implement the backend... would be with reversed lists."
- "Users will probably skip the section."
- "I was hoping that you would know."
The Final Straw: A Complete Abandonment
Despite repeated attempts to guide the AI, the developer reaches a breaking point. The AI's persistent errors and disregard for instructions ultimately lead to the project's abandonment, emphasizing the limitations of current AI technology in complex development tasks.
- "Why did you delete my repo? It was good code!"
- "NO! Not Vercel! No, not npm install! Nooooo..."
- The developer acknowledges that everything he knows is now invalid.