Ten years of code and a sleepless editor

The code was ten years old. Not all of it, of course. There were more recent parts, like layers of paint on a wall where no one remembers what color it originally was. But the core, the structure that held everything else together, had been there for a decade. It was written by a developer who back then had fewer gray hairs and more patience for long variable names.

His name was Ramón. Or it could be.

Ramón knew his code like you know a house you've lived in for a long time. He knew where the floor creaked, which door had to be pushed a little harder, in which corner there was a decision made at two in the morning that worked but that he couldn't fully explain if asked in an interview.

Then came a request: adapt the system for two new autonomous regions. Aragón and Castilla y León. Small but delicate adjustments. The kind that seem easy until you pull a thread and the whole fabric moves.

Ramón opened the editor. Looked at the code. And for the first time in ten years, he asked for help.

The machine that reads better than you

The help didn't come from a colleague. It came from an artificial intelligence. Ramón passed it the code, the specifications for each region, and the requirements. The AI read it. All of it. The ten layers of paint, the variable names that no longer meant what they said, the two a.m. decisions.

And it understood it.

It didn't understand it the way a junior understands it when you assign them the project and they take two weeks to ask why there is a function called process_dossier_v3_final_FINAL. It understood it like someone who reads an entire book, retains every page, and can explain the relationship between chapter three and chapter seventeen.

Ramón asked it to prepare the models for Aragón. The AI proposed the changes. They were the correct changes. The ones Ramón would have made, but taking three days to locate all the dependencies. The AI took minutes.

For Castilla y León it did the same. Small adjustments. Delicate ones. The kind where if you get a decimal wrong, a dossier comes out wrong and someone has a real problem. The AI didn't get any decimal wrong.

The quantitative editor

Then came the second part. Ramón had a document. A report with data, figures, percentages. He had written it himself, carefully. But there were many numbers and several sources, and Ramón knew that one error in a data point could ruin the credibility of the whole text.

He passed the report to the AI. He told it: "Check all the numbers. One by one. Cross-reference with other sources. Do your own calculations."

The AI did it. It found two figures that didn't add up. One was an incorrect rounding. The other, a data point that had changed between the original source and the updated version. Ramón would have probably found them too. But it would have cost him a whole afternoon of going back and forth between browser tabs, and in the end, he would have trusted that the other twenty-seven figures were right because he was tired of looking at numbers.

The AI checked them all. Without getting tired. Without deciding that the last ten were probably fine.

He didn't say it with admiration or fear. He said it like someone observing that the washing machine washes better than doing it by hand. It doesn't take away the job of getting dressed. It takes away the scrubbing.

What doesn't change

There is one thing Ramón didn't ask the AI. He didn't ask it to decide what to build. He didn't ask if Aragón needed a new module or if Castilla y León could reuse Murcia's with adjustments. He knew that. Because he had spent ten years talking to the people who use the system, understanding why each field exists, remembering that meeting where someone from the ministry said that data was mandatory and another said it wasn't, and in the end, it was left as optional but everyone fills it out.

That knowledge isn't in the code. It's in Ramón's head. In the conversations of ten years. In the decisions he made with information that was never documented because no one thought it would be necessary.

The AI can read the code better than Ramón. It can adapt it faster. It can check figures without blinking. But it cannot sit in a meeting with the head of an autonomous region and understand that when they say "we want something simple," they actually mean "we have no budget for training and our team is afraid of computers."

The one who knows and the one who executes

Ramón works differently now. He thinks more. He types less. He spends mornings understanding what needs to be done and afternoons reviewing what the AI proposes. Sometimes the AI is wrong. Not in the data, it does that well. It's wrong in the decisions. It proposes a technically correct solution that doesn't fit with how the team works. Or suggests a change that breaks an unwritten convention known only to someone who has been on the project for ten years.

Ramón corrects those things. Without getting angry, without drama. The way you correct a very competent assistant who doesn't know their way around the house yet.

His wife asked him one day if he was afraid the AI would take his job.

"No," he said, putting the plates in the drying rack. "I'm afraid that someone who knows nothing about the business will use it thinking they don't need to know."

What this means if you have a company

If you need custom software, what has changed is not the what but the how. The result is still the same: a tool that solves a specific problem in your business. But the path to get there is faster. What used to take months now takes weeks. What used to be expensive to maintain can now be updated with less effort.

The difference lies in who is steering. If on the other side is someone who understands your business and uses AI as a tool, the result is better and faster. If on the other side there is only an AI with no one who knows what questions to ask, the result is code that works and serves no purpose.

Ramón closed his laptop at seven, which was something that almost never happened before. He put on his jacket. On the screen was a three-hundred-line diff that the AI had generated and he had reviewed in forty minutes. A year ago, it would have taken him two days.

He stepped out into the street. It was cold. He didn't think about code.