# Build the Thing That Builds the Thing

**Published:** August 20, 2026 | **Authors:** Cristian Correa

---

Everyone is shipping fast with AI. Very few are shipping smart. The difference is not speed. It is what you spend your tokens on.

## The Slop Trap

When you have no direction, it is normal to just prompt and ship. You throw tokens at the problem, the model produces something, you move on. This is fine for exploration. It becomes a trap when you stay there.

The trap has a name: AI slop. Not because the output is bad, but because nothing compounds. Every new feature costs the same as the last one. Every project starts from zero. You are renting intelligence by the hour and building no equity with it.

Speed without leverage is just motion.

## Patterns Are the Signal

You do not need a grand plan to escape the trap. You need to notice repetition.

The third time you write the same kind of bash script, that script wants to exist as a tool. The third time you explain the same context to the model, that context wants to be a skill. The third time you wire the same architecture, that architecture wants to be a framework.

Repetition is the market telling you where the abstraction is. Ignore it and you pay the same cost forever. Capture it and the cost drops to near zero.

```mermaid
flowchart LR
    A[Repeated task] --> B[Script]
    B --> C[Skill plus script]
    C --> D[Framework]
    D --> E[New things cost almost nothing]
```

Each

That is the whole game. Build the thing that builds the thing.

## Tokens Are Capital

Most people treat tokens as fuel. Burn them, get output, done. Better to treat them as capital.

Spent on raw output, a token is consumed once. Spent on an abstraction, a token keeps paying. The framework you build today makes every future feature cheaper. The skill you write today makes every future session shorter. This is the difference between income and investment.

Token maxing is not about using more tokens. It is about thinking in the future. Ask of every session: am I buying output, or am I buying leverage?

```mermaid
flowchart TD
    A[Tokens] --> B{Spend on what}
    B --> C[Raw output]
    B --> D[Abstractions]
    C --> E[Consumed once]
    D --> F[Cheaper tokens tomorrow]
    F --> A
```

The

## The Subsidy Window

There is a second reason to think this way, and it is temporal.

Right now, subscription based AI is subsidized. You pay a flat fee and get access to intelligence that costs far more to serve. This will not last forever. Windows like this never do.

The rational move is arbitrage. Convert cheap, subsidized tokens into durable assets: scripts, skills, frameworks, systems. When the window closes and tokens get priced at their real cost, the people who only bought output will feel it immediately. The people who bought leverage will barely notice, because their abstractions already did the expensive thinking.

Mine the subsidy. Do not just consume it.

## Where This Comes From

I am not writing this from theory. I am writing it in the middle of shipping our internal framework at Croma. We call it loom. There is a fun story behind the name, but that is for another day.

![A laptop open on an airplane tray table, code on the screen, the seat back in front labelled A320-200](/changelog/build-the-thing-that-builds-the-thing.jpg)

*For what it is worth, this one was written and shipped from a plane seat. Time maxing is the same idea, applied to hours.*

Croma connects AI to [real world sources](/co/fuentes): government registries, courts, procurement systems, company data. The naive way to grow is to hand build every connector. Ship one, celebrate, start the next from zero. We did that for a while. It works, and it does not scale.

Loom is our attempt to build the thing that builds the thing. Instead of building connectors, we are building the machinery that produces connectors. If it works, going from hundreds of sources to thousands, and eventually to millions around the world, stops being a heroic effort and becomes a pipeline.

I say if it works because we are still proving it. But even half proven, the lesson holds: the connector is not the asset. The system that produces connectors is.

## Ship Fast, Ship Smart

None of this argues against speed. Ship fast. But if it is something you want to maintain over time, ship smart too. Sustainable is not the opposite of fast. Sustainable is what lets you stay fast.

The compounding builder and the slop builder look identical in week one. By month six, one of them is shipping in hours what the other still ships in days, and the gap only grows.

So the discipline is simple. Explore freely when you have no clue. But the moment you see a pattern, stop and capture it. Turn the task into a script, the script into a skill, the skill into a framework. Spend tokens like an investor, not a tourist.

Build the thing that builds the thing. Everything else is rent.

---

**More updates:** [View all changelog entries](/en/changelog/sitemap.md) | [Croma](https://usecroma.com)