[Pixel Post] How to Actually Get AI to Build Features Without Losing Your Mind

You can’t just say, “Claude, build X,” and expect magic. That’s like throwing your Figma files at a junior dev and walking away.

If you want AI agents to help build features that don’t suck, you’ve got to coach them like a real teammate. Here’s the workflow I’ve been using (and loving) with Claude in Cursor — it’s structured, repeatable, and surprisingly effective.

Step 1: Set the Ground Rules with .cursorrules

Think of this file as your AI onboarding doc. It spells out:

• Coordinate systems (e.g., “Y increases downward”)

• Key files and roles (“App.tsx is the root – don’t touch Main.jsx”)

• Don’ts (“Don’t create global state unless it’s 1997”)

This alone saves hours of AI-wrangling.

Step 2: Write a Design Doc (Yes, Really)

Before code, write what you actually want.

• What should the feature do?

• How should it look/work?

• What files are involved?

• Any weird edge cases or business logic?

You’re basically writing a ticket for your AI coworker. Be clear and specific.

Step 3: Let the AI Ask Questions

Prompt: “Here’s the design doc. What’s unclear?”

Good AIs will ask things like:

• “What if the user doesn’t have an email?”

• “Should the token expire?”

If Claude shrugs and says “All good!”, press it. Assume it's bluffing.

Step 4: Update the Spec With Clarifications

Take its questions seriously. Update the spec. You can even ask Claude to rewrite it with clearer structure — it’s often better at organizing your ideas than you are.

Step 5: Ask for an Implementation Plan

Before touching a single line of code, ask Claude:

“Break down what files will change, and what you’ll do in each one.”

This gives you a preview of its plan — and a chance to course-correct before you’re knee-deep in spaghetti.

Step 6: Let It Code (But Watch Closely)

Claude now has:

• The rules

• A clear spec

• An implementation plan

Let it rip, step-by-step. Review its changes like you would for a junior dev. Fix things early, stay in the loop.

---

Final Thoughts

Treat AI like a junior teammate, not a genie. Structure and prep matter. Do it right, and you’ll ship cleaner features with fewer headaches — and maybe even enjoy the process.

Try it. Your future self (and your git history) will thank you.