tl;dr AI can ship fast. Like… really fast. But if you don’t layer in structure, intent, and human refinement, that speed can come at a steep cost.
Move fast—but pay your future self by building in check-ins, code patterns, and feedback loops from the start.
1. “Vibe Coding” Is Real—and Risky
Right now, we’re seeing people with no traditional coding background vibe their way into working apps.
That’s a good thing—for speed, creativity, and access.
But…
AI is often great at making something work, not making it right.
If you rely purely on a single AI tool to generate your code (especially without proper structure), you’ll likely pay the price later: - In feature collisions - In unscalable patterns - In “where-the-heck-is-this” code chaos
2. Speed ≠ Sloppiness
Here’s the pattern I use when building fast features for BoxBoard while still keeping my codebase clean:
✅ Small updates only
Even though AI wants to do a lot at once, I try to chunk my changes. One idea per commit, even if it’s minor.
✅ Refactor checkpoints
After a set of commits, I take a beat and do human-driven polish: - Break long views into partials or components - Normalize CSS - Re-structure messy logic - Apply naming consistency
✅ Design first, refine later
For views and layout:
- I let the AI scaffold the initial pass
- Then I refine layout and CSS patterns manually
Especially if it’s custom. Tailwind-based React projects? AI does great. Custom-markup Rails views? I take the lead.
3. Use AI Tools as a Team, Not a Single Brain
If you’re just throwing code into one AI, you’ll miss the power of roles.
Here’s how I think about it:
- Builder AI – ChatGPT or Cursor helps write initial logic
- Reviewer AI – Claude or even another ChatGPT instance checks for structure, naming, or optimization
- Me (Human) – Guides structure, product vision, UX thinking, and subtle polish
This approach prevents bad code from compounding. One AI builds. One reviews. I lead.
4. Add Rules When You Find the Gaps
AI learns from you—if you train it right.
Whenever I find myself fighting an AI-generated pattern (especially with CSS or file placement), I create new rules:
❌ “Don’t create random files in
app/assets.”
✅ “Useapp/frontend/componentsfor reusable JS and CSS.”
✅ “All new views should use theBoxboardLayoutand follow X pattern.”
I keep a living prompt primer and reference it in each session.
I’ve even created diagrams of my folder structure and included that as visual input to help the AI better understand the app.
5. AI Is Speed. You Are Structure.
The best way to think about it is this:
| AI Strengths | Your Role |
|---|---|
| Rapid scaffolding | Set the vision |
| Syntax + boilerplate | Shape the structure |
| Suggestions | Define the rules |
| Getting unstuck | Know when to pause and rethink |
6. Final Thought
Yes—AI can help you ship a working MVP in a day.
But someone is going to pay the tech debt bill eventually.
That someone might be you… in two weeks.
So set the structure early. Build fast. Refactor often.
And treat your AI tools like junior engineers—great at execution, but always in need of guidance and a good product lead.
Got your own rules or ways to avoid AI-induced chaos?
Let me know what’s working (or not) in your workflow—DM me or connect over LinkedIn or X.