Just Build It Yourself.

”Ship Before You Hype” Series #0 (Prologue)

Every Week, Same Story

“Let’s build the workflow in n8n for this project.” “Made a mock in Google AI Studio — blazing fast!” “RAG! Let’s build a knowledge base!” “We can orchestrate everything with Dify.”

Every project brings a new tool. Someone walks in, eyes bright, convinced they found the silver bullet.

Every time, I think the same thing.

Just build it from scratch. What are engineers for?

n8n: Stop Connecting Nodes

n8n is a visual workflow tool. Connect nodes with lines. Automate without code. That’s the pitch.

In the time it takes to learn the node UI, you can write a shell script. Shell scripts are readable. When they break, you fix them. When n8n nodes break, you click around a GUI hunting for the cause. The speed difference is obvious.

Visual workflows exist for people who can’t write code. If you can write code, there’s no reason to use them.

Google AI Studio / Dify: Legacy Debt Factories

Build a mock in Google AI Studio. “Prototype done in minutes!” Celebrate.

Then hand it to an engineer.

Too many bugs. Code quality is abysmal. Fixing it eats more hours than building from scratch. Time that should go to new features gets spent cleaning up AI-generated mess. The mockup becomes negative legacy. Would have been better never handing it off.

Dify has the same structure. Wire up processing in an orchestration tool. It looks like it works. Inside, it’s a black box. When edge cases break it, who fixes it? Engineers open the hood and fight the tool’s quirks instead of building the product.

Polish your prompts section by section. Call APIs directly for orchestration. Middle layers create the complexity they claim to solve.

RAG: Before You Add Search

“RAG!” — the word bounces around every meeting like a magic spell.

Retrieval-Augmented Generation. Search external knowledge to improve AI accuracy. The theory is sound. Pull relevant information from massive document collections. Smart.

But first: are those documents actually organized?

Put a high-performance search engine in a messy room, and all it finds is mess. Before deploying RAG, design your file structure. Organize directories. Establish naming conventions. Add metadata.

Write careful prompts against structured files. That alone eliminates most problems people wanted RAG to solve.

Before adding a search mechanism, organize what you’re searching through.

All Middle Layers

n8n, Google AI Studio, Dify, RAG.

All middle layers. Things wedged between you and the code.

Middle layers exist for people who can’t build. People who can’t code connect nodes. People who can’t write prompts configure GUIs. People who can’t design file structures lean on search engines.

Builders have no reason to use middle layers. They get in the way. The garbage middle layers produce gets cleaned up by the people who can actually build. That’s what’s happening in the field right now.

About This Series

Five articles follow, each dissecting a specific tool.

Every one of them, used firsthand. Every one, pushed until it broke. That’s why I can write this.

I don’t hand out blueprints. I show you where the walls are. How to climb them is your job.

Everyone has their own We ship. We own.