You're Probably Using AI Wrong for Programming (And Here's the Right Way to Do It)

Google Chrome's head of engineering warns: You're probably using AI to code wrong, and this is the right way to do it.

The key is to give things context. It's that simple.

Something very important is mentioned here: artificial intelligence has become the Swiss Army knife of programmers. But it's not a replacement, it's simply a tool.

The same goes for GitHub Copilot: it's an assistant, the copilot, but you are the pilot. Don't let the tool work on its own; you are the one who uses it.

AI as a tool, not a replacement: The importance of context

In the developer community—and I include myself—the same thing is always repeated: plan before you ask. Giving things context is another way of saying it.

It's not about throwing out a vague idea without context and hoping for the best. It's tempting to ask for something simple and expect perfect code, but that rarely works. The more complex the project, the more likely you are to get an incomplete or confusing result.

Example: If your boss just says, "Make me a task app," without details, you'll do something your way, but it probably won't meet all the project's actual requirements.

The recommendation is to work module by module, little by little, using AI as a support, not as a total executor.

Many people on the internet show how AI creates an application from scratch. This isn't bad, but it's often done only to measure speed, capabilities, or as a technical test, not to achieve a quality final product.

Clear specifications
The article suggests providing a good plan before ordering:

• Purpose and Requirements: What exactly should the code do?
• Inputs and Outputs: Define what data it receives and how it should display it.
• Constraints: Technologies to avoid, APIs to integrate, etc.
• Development Phases: Proceed step by step.
• Style Guides: Indicate how the code should be structured.

This not only helps the AI, but also serves as documentation for the project.

The inverted method

An interesting recommendation is to apply a method similar to TDD (Test-Driven Development): create the tests first, validate them, and then develop the code that passes them.

Although for large projects this may be too abstract, if the tests are performed by AI and you review them, this complexity is reduced.

AI isn't magic. A vague prompt won't give you a perfect result. It's a tool that, used with planning and context, can greatly speed up work.

Post:
https://www.larazon.es/tecnologia-consumo/jefe-ingenieria-chrome-avisa-probablemente-estas-usando-mal-programar-esta-forma-correcta-hacerlo_20250803688b2dd69db33905c4dcc2c0.html