How I Use AI in My Code Workflow (Real Setup of a Freelance Dev)

When I say I use AI to code, the reaction is usually one of two:

1. “So you don’t really know how to code”
2. “Teach me how”

This post is for person #2. But if you’re person #1 — read it too. Maybe you’ll change your mind.

My AI stack

I use 3 tools daily:

• GitHub Copilot — autocomplete directly in VS Code
• Claude (Anthropic) — code review, refactoring, architecture
• ChatGPT (OpenAI) — quick debug, research, code generators

Each one has a specific role. None replaces the other.

GitHub Copilot: the turbocharged autocomplete

Copilot sits in VS Code suggesting code as I type. It’s like a pair programmer who reads my mind.

Where it shines:

• Obvious implementations (utility functions, types, unit tests)
• Pattern autocomplete for patterns I’ve already established in the project
• Boilerplate I’d write anyway

Where it fails:

• Complex business logic (that’s where I need to think)
• Legacy code with unconventional patterns
• Sometimes suggests deprecated or insecure code

My rule: I accept Copilot suggestions as drafts, never as final version. I always read and validate before accepting.

Claude: the reviewer and architect

Claude is my main tool for heavy code work.

My workflow with Claude:

1. Code review: I paste an entire PR and ask “review this code, focus on bugs, performance, and security”
2. Refactoring: “This React component has 400 lines. Suggest how to break it into smaller components maintaining the same functionality”
3. Architecture: “I need to implement X feature. What are the architecture options? Pros and cons of each”
4. Documentation: “Generate JSDoc for these functions based on the implementation”

Claude with 200K tokens of context handles entire projects. I can paste 10 files and ask for a cross-analysis — and it understands.

ChatGPT: the Swiss army knife

ChatGPT comes in when I need quick answers or specific tools.

My workflow with ChatGPT:

1. Debug: I paste the stack trace and ask “what’s causing this?” — solves it in 80% of cases in 5 minutes
2. Regex: “Write a regex that does X” — never wasted time on regex101 again
3. Research: “What’s the difference between X and Y in React 19?” — with built-in search, cites official docs
4. Generator: “Generate a Zod schema for this JSON” — saves 20 minutes of typing

What changed in my productivity

Before AI:

• Boilerplate: 30 min/day writing repetitive code
• Debug: 45 min/day on simple errors
• Docs: 20 min/day documenting
• Research: 60 min/day reading Stack Overflow

After AI:

• Boilerplate: 5 min (Copilot does it, I review)
• Debug: 10 min (ChatGPT finds it, I verify)
• Docs: 5 min (Claude generates, I adjust)
• Research: 15 min (ChatGPT summarizes, I confirm)

Savings: ~2h/day. In a month, that’s 40+ hours. It’s like having an extra week of work.

What AI does NOT do for me

• Think through architecture: AI suggests options, I decide
• Understand business context: AI doesn’t know why the client wants X feature
• Guarantee quality: AI suggests, I test, I validate
• Accountability: If code breaks in production, it’s my fault — not AI’s

Tips for those who want to start

1. Install GitHub Copilot (has a free trial) — it’s the most natural way to start
2. Create a Claude account (free) — paste code and ask for review
3. Stop copying from Stack Overflow — ask ChatGPT and request an explanation
4. Always read the suggested code — AI makes mistakes. You need to catch the error
5. Invest in good prompts — 2 minutes on a clear prompt saves 20 minutes of rework

Conclusion

AI doesn’t replace knowing how to code. It multiplies those who already know.

If you understand what you’re doing, AI makes you 2-3x more productive. If you don’t understand, AI can fool you with code that looks right but isn’t.

The key: use AI to execute, but keep thinking with your own head.