10 Harsh Truths from 6 Months of Building with AI

Today

1. Git is Life Support

At some point, something will break. Cursor might hallucinate your app.js into the void. You might forget what you changed last night.

Version control is your lifeline. Use branches. Use local changelogs per folder. It doesn’t just save code, it gives your AI context breadcrumbs to follow.

2. Clean Your Codebase Like It's Your Kitchen

Every week or so, stop and tidy up. Delete those leftover temp files. Rename folders that got messy. AI works better in a clean workspace, and so do you. It’s like rebooting your brain.

3. Don’t Ask the AI to Build the Whole App

It’s a tool, not a junior developer. It’s great for UI stubs, small logic blocks, and those annoying refactors. But trying to get it to build an entire system in one go? That's like asking a microwave to bake a wedding cake.

4. Write the Deployment Manual Now

Seriously. Before you forget what needs to happen to ship this thing, write it down. Which branch do you deploy from? What env vars? What server config?

It’ll save you from those 2am “Why isn’t this working?” spirals.

5. Scope Your Features Before Coding

Cursor is not a place to figure out what you want. It’s where you execute. When you're planning a feature, map it out in GPT or Claude first. Get ideas, pick one, then go implement it. Don’t just wing it inside Cursor.

6. Start Every Project Like a PM, Not a Prompt Monkey

Before you write a line of code, create a simple product doc, what you're building, why it matters, what tools you're using.

Save it in the root of your project (product.md or instructions.md). Refer back to it constantly. It’s your north star when AI loses the plot (which it will).

7. One Issue, One Chat

Don’t dump everything into one endless Cursor chat. Keep it scoped. One issue = one chat thread. It makes it way easier to get good responses, and prevents the AI from getting “confused toddler with scissors” energy.

8. Ask, Don’t Tell (Yet)

When debugging, don’t jump to solutions right away. Ask the model to explore first. Get multiple suggestions. Then decide. Only then ask it to implement.

Saves you from recursive bug loops and code spaghetti.

9. Tech Debt Happens at AI Speed

You can ship fast with AI, but you can also end up in a mess twice as fast. That quick MVP can rot into a nightmare if you don’t pause to refactor every now and then.

AI helps you move fast, but you still have to steer.

10. Lead the Machine, Don’t Follow It

Cursor isn’t magic. It’s not replacing you. You’re still the one making the big calls. Use .cursorrules to define project rules. Use git checkpoints. And most of all, bring your own product sense and systems thinking to the table.

Keep it up! Lead the machines. 🤘