2026-03

How do I work with AI in a smart way as a developer

A quick note on how I use AI as a developer staying in control, writing better code, and avoiding the “vibe coding” trap.

AI is now part of my daily workflow as a developer. But I often see a gap between two approaches:

• people who “vibe code” and let AI drive everything
• and developers who use AI as a tool, not as a replacement

So I asked myself:

Where’s the line between a vibe coder and a developer?

This is a quick note on how I try to use AI in a controlled, efficient, and sustainable way.

---

Create Everything, Break Nothing

One of my main rules is simple:

Create everything, break nothing.

As a developer, I want to be confident that anything generated by AI:

• doesn’t break existing features
• doesn’t introduce unnecessary code
• respects the project’s standards and best practices

That’s where testing becomes critical. I rely a lot on tests to validate AI-generated code, especially when making changes across existing features.

My approach

• I use vitest for testing
• I try to cover critical logic before making changes
• I validate outputs instead of trusting generated code blindly

import { describe, it, expect } from "vitest";

describe("sum", () => {
  it("should add two numbers correctly", () => {
    expect(sum(1, 2)).toBe(3);
  });
});

Of course, AI can also help generate tests (with tools like Claude Code or Cursor), but I still:

• review them
• adjust them
• make sure they actually test something meaningful

This naturally leads into a TDD-like mindset (Test-Driven Development), which I’ve found really effective when working with AI. (I’ll probably write a dedicated article about my approach to TDD.)

---

Step by Step — Setting Rules & Good Practices

Another key difference I see:

• A developer guides the AI
• A vibe coder follows the AI

I always try to bring my experience, my product understanding and my architectural decisions into the process.

AI is great at execution, but it doesn’t always understand long-term vision and doesn’t care about consistency unless you enforce it. So instead of prompting randomly :

• define rules
• structure my project
• guide the outputs step by step

AI becomes an assistant, not a decision-maker. (I’ll go deeper into this in another article: My Approach to Using Cursor and AI Agents in my workflow)

---

The Importance of Context

Something I’ve learned the hard way:

Context matters more than prompting.

A vibe coder will often focus on writing better prompts. Personally, I focus more on:

• providing context
• structuring information
• giving constraints

The more context the AI has: the better the output & the more consistent the results. I try to include things like:

• existing code
• project structure
• conventions
• expected behavior

Also, defining clear rules upfront allows me to:

• avoid repeating myself
• reduce input tokens
• get more predictable outputs

Again, this ties into how I use tools like Cursor and AI agents in my workflow: My Approach to Using Cursor and AI Agents in my workflow

---

Code Review Is Not Optional

This is probably the biggest mindset difference.

For a vibe coder AI is the developer, for me AI is more like a junior developer (even if sometimes it’s better than me).

And like with any junior dev:

• I review the code
• I question decisions
• I validate correctness

Code review is not optional, it’s part of the process. Yes, you can delegate code review to AI agents (again, with tools like Claude Code or Cursor), but I still read the code myself, make sure it aligns with my expectations and every check edge cases.

---

Final Thoughts

AI is incredibly powerful—but only if used correctly. The difference isn’t the tool. It’s how you use it.

What I try to do:

• stay in control
• rely on tests
• provide context instead of just prompts
• review everything

I don’t try to code faster at all costs. I try to code better, with AI as a multiplier—not a crutch.