My AI-Driven Coding Workflow: Fast, Dirty, Smart
For years, I’ve followed a simple, effective development rhythm:
- Refactor the existing code to make space for what’s coming.
- Spam code the new part—quick and dirty—just get it working.
- Refactor again, clean it up, make it readable and reusable.
- Commit.
It’s always worked. It still does.
But now there’s AI.
The last few years have made AI coding assistants usable, stable, and practical. So I’ve adapted my process. Here's how it looks today:
Step 1: Prep — Clean Code, Clean Repo
Start with a clean working tree. No changes, no mess.
This has always been good practice, but it’s critical now.
Why? Because AI can (and will) generate garbage.
If it does, you need to scrap everything fast. You can’t "undo" AI-generated logic. You just reset.
Step 2: Refactor — Set the Stage
Whether by hand or with AI help, prepare the codebase.
Don’t bolt new code onto a shaky structure.
Make room. Rename things. Break things apart.
This step clears the path and gives the AI (or your future self) better context to work with.
Step 3: Code — Vibe Hard, or AI Fast
Now comes the messy middle: vibe code or AI code until it works.
Requirements first. Style later.
If you're fast, you'll get something functional in minutes.
Use AI here as a collaborator, not a crutch. Let it write snippets, functions, scaffolding—whatever helps you move fast.
Keep changes small. Bridge small knowledge gaps, not large ones.
The more context the AI has, the less it hallucinates.
Step 4: Refactor Again — Own It
This step cannot be skipped.
Go over everything.
Touch every file.
Understand what was written.
Make it readable—not for your future self next year, but for you this afternoon.
This is where you take ownership of the code.
AI can write, but only you can understand.
Bonus Step: Study the Tech
The time you save using AI? Spend it learning what's under the hood.
If you’re building with embeddings or RAG or LLMs—understand them.
Know what they’re doing, how they work, where they fail.
It makes you a better developer.
It also makes you a better AI prompt engineer.
TL;DR
- Start with a clean repo
- Refactor first
- Spam code (or use AI)
- Refactor again—hard
- Learn the tech behind the tools
This workflow is fast, safe, and scalable.
It keeps you in control.
It’s AI-powered, but developer-owned.
Top comments (0)