How AI Tools Changed My Development Workflow
My experience using Claude Code, Cursor, and GitHub Copilot in production. What works, what to watch out for.
Introduction
Since around 2025, AI development tools have reached a practical level. I use them daily now, so here are my thoughts on how they work in practice.
Tools I Use
Claude Code
A terminal-based AI agent. My most-used tool.
# Just describe what you want
claude "Add a honeypot field to ContactSection for spam protection"
What's great:
- Autonomously reads/writes files and runs commands
- Understands context thoroughly before implementing
- Code review and refactoring suggestions are accurate
This portfolio site was actually built with Claude Code. From component design to ESLint fixes, I delegated a significant portion of the work.
Cursor
A VSCode-based AI editor. Convenient for using AI inline while coding.
- Tab completion: Context-aware code suggestions
- Cmd+K: Refactoring and transforming selected code
- Chat: Ask questions about code right in the editor
GitHub Copilot
Can't imagine working without it. Especially fast for boilerplate generation.
How I Use Them in Practice
| Tool | When I Use It |
|---|---|
| Claude Code | New features, design decisions, multi-file changes |
| Cursor | Day-to-day coding, partial refactoring |
| Copilot | Autocomplete, test generation, standard patterns |
My general approach: "AI for thinking, AI for assisting."
Things I'm Careful About
1. You Must Be Able to Read Code
Never merge AI-generated code you don't understand. This is non-negotiable.
AI can write code that looks correct but is subtly wrong. Types might check out, but the business logic could be off.
2. Prompt Quality = Output Quality
❌ "Build a login feature"
✅ "Add login using Firebase Auth with Google sign-in.
Manage auth state with React Context. Redirect to /login when unauthenticated."
The more specific your requirements, the better the output.
3. Design Decisions Are Yours
AI presents options, but you choose. Only humans can factor in product context, team skills, and future extensibility.
Following AI suggestions blindly tends to lead to over-engineering. Always ask yourself: "Do I actually need this pattern?"
Skills Engineers Need in the AI Era
Three things I believe matter most:
- Problem Definition — Clearly defining what needs to be solved. Can you decompose the problem correctly before handing it to AI?
- Code Reading — Judging the quality of AI-generated code. You can't review what you can't read
- Architecture Thinking — Seeing the big picture. AI excels at local optimization, but global optimization is a human job
Wrapping Up
AI tools have definitely boosted my productivity, especially for personal projects. What used to take "a weekend" now takes "a few hours."
But AI is just a tool. Deciding what to build and why is still on you. As long as that's clear, AI becomes the ultimate partner.