Dos personas analizando datos en una computadora y un monitor.

Copilot and more: how AI is changing the way we write code

  • 9 July 2025
  • 4 minutos
  • Blog

Programming has changed radically in the last decade, not only because of the evolution of languages or frameworks, but also because of the emergence of new forms of collaboration and automation, from the popularisation of agile methodologies or working with containers.

However, few have been more disruptive than the application of artificial intelligence capabilities to code editors. Of these, nothing has had more impact than the transformation that GitHub has undergone.

A landmark purchase

On 4 June 2018, Microsoft signed one of the most interesting acquisitions in its history. On that day, the American multinational announced the acquisition of GitHub for 7.5 billion dollars.

In this way, Microsoft took over one of the main version control services, which among other things made it easier for developers to follow the complete history of changes in the code, or for teams distributed around the world to work together.

Five years after that purchase, GitHub has become the world's largest collaborative development platform, used by approximately 150 million developers working on one of more than 420 million registered repositories.

OpenAI and GitHub Copilot

But the story doesn't end there. In 2021, with the support of OpenAI, GitHub Copilot was launched, a programming assistant based on Codex (an artificial intelligence model trained on billions of lines of open source code, mostly hosted on GitHub).

Its launch was a revolution in software development, introducing for the first time a system capable of suggesting code snippets, completing functions or generating entire structures from natural language comments, which promised to exponentially increase developer productivity.

This was especially useful for repetitive or standard tasks such as loops, mathematical functions or conditional structures in languages such as Python, JavaScript or TypeScript and promised to exponentially increase developer productivity by integrating as an extension in popular editors such as Visual Studio Code .

However, not everyone embraced this "revolution" from the start. Firstly, because, as with all artificial intelligence models, GitHub Copilot could "hallucinate". Indeed, in its early versions, it could generate code that was functional but also error-prone, insecure or redundant. This meant that programmers had to have solid technical knowledge to review and validate suggestions, which could end up impacting those theoretical productivity gains.

On the other hand, many programmers questioned the ethics or even the legality of Copilot being trained with public code from repositories hosted on GitHub, since in its output the model reuses code snippets without citing the original author. In addition, some experts warned that the intensive use of Copilot could discourage deep code learning, especially among junior developers, encouraging shallower and less comprehensive programming.

But since those early versions, GitHub Copilot has improved dramatically. One of the most notable advances has been its ability to understand the full context of a project, not just the file being worked on. This allows it to generate suggestions that are more coherent and adapted to the style of the existing code. It has also improved its accuracy and security, avoiding suggestions that may contain common errors or poor practices.

Starting with Copilot X, developers also have Copilot Chat at their disposal, a conversational environment in which users can interact with the AI to resolve doubts, ask for code examples or refactor functions directly from the editor. At the same time, automatic generation of unit tests, explanation of code fragments and extended support for more languages and environments have been incorporated, although JavaScript and Python are still the ones that offer the best results.

Programming is a matter of... AI

Beyond GitHub Copilot, the last few years have seen the development of a good number of coding assistants based on artificial intelligence, offering varied approaches (from local development privacy to collaboration in cloud environments) and adapting to different user profiles. Here are some of the most popular ones:

Cursor: alternative editor to VS Code with integrated AI for semantic code navigation. It uses GPT-4 and allows code editing with natural language.

Amazon CodeWhisperer: offers real-time code suggestions integrated with AWS and multiple IDEs. Includes security analysis and filters out suggestions that may violate licenses.

Tabnine: works locally or in the cloud, respecting source code privacy. It is compatible with VS Code, JetBrains, and more than 20 programming languages.

Windsurf: formerly known as Codeium, this editor has the advantage of being free for individual use. It integrates features such as full function generation and automatic documentation.

Replit Ghostwriter: ideal for beginner developers and educational environments, it combines AI with a cloud-based collaborative development environment.

As we have seen, these assistants represent a significant advance in productivity, efficiency and learning. However, it is important to emphasise that they are not a substitute for the judgement, creativity and language understanding of the human developer. It is experience and critical thinking that continues to make the difference in creating truly innovative technology solutions.

More information

Degree in Full-Stack Software Development

Related Insights

Una oficina moderna con un escritorio ordenado, un portátil y diversos libros y revistas sobre diseño arquitectónico.
Videojuegos, Branding, Interiores, Creación gráfica, Producto, Ciencia y Tecnología, Moda, Animación & UX

How to choose a Master’s degree: what to look for beyond the programme’s name

28 October 2026

Choosing a master’s degree requires comparing what a programme advertises with the information actually provided in its academic documentation. Comparing degree programmes is not simply a matter of looking at their names or syllabuses, but of checking five key aspects: the focus of the curriculum, the prior knowledge required, the skills and projects covered, the delivery method, and the academic nature of the qualification.

La imagen muestra un fondo abstracto que combina elementos de gráficos de datos y patrones dinámicos en tonos azules y dorados.
Ciencia y Tecnología

From data analysis to a decision: what to check before taking action

20 October 2026

A data-driven decision uses analysed information to compare alternatives or justify a course of action, but does not in itself eliminate uncertainty. Having well-organised records, dashboards or numerical estimates does not guarantee that a course of action will solve the anticipated problem, nor that its costs will be outweighed by the expected benefits. Before turning a result into a recommendation, it is worth asking the following question: what makes a data analysis sufficiently reliable and relevant to justify a decision?

Un grupo de estudiantes sentados en un aula prestando atención.
Videojuegos, Branding, Interiores, Creación gráfica, Producto, Ciencia y Tecnología, Moda, Animación & UX

How to tell if a creative or tech career is right for you: take a look at the work that goes into it

30 September 2026

To find out whether a degree course is right for you, it is not enough simply to ask yourself whether you like the outcomes. It’s worth finding out which tasks recur during your studies and in your work, what difficulties are part of the process, what problems arise in the curriculum, and what projects you’ll need to learn to tackle. Making a better choice means replacing an idealised image of the profession with evidence that’s closer to how the work is actually done.

Un grupo de jóvenes prestando atención en un aula moderna.
Ciencia y Tecnología

What goes on behind the scenes: software, data and artificial intelligence within a digital product

25 September 2026

A simple function within an app may depend on dozens of invisible components. The software receives an action and executes logic; the data reveals what is happening and what happened previously; and, when the problem requires predicting, classifying, recommending or generating a response, an artificial intelligence model may come into play. Understanding a digital product requires distinguishing between these layers and, above all, understanding how they depend on one another.

Un orador presenta frente a un público joven en una biblioteca.
Videojuegos, Branding, Interiores, Creación gráfica, Producto, Ciencia y Tecnología, Moda, Animación & UX

Learning by doing in design and technology: from the initial idea to a solution you can defend

4 September 2026

Learning by doing does not mean replacing theory with projects. In Design and Technology, it means turning an idea into something that can be observed and tested, gathering information about its shortcomings, revising decisions and explaining why changes have been made. Learning takes place within this cycle of action, evidence, feedback and iteration, not solely in the final outcome.

Una multitud de personas se reúne en un espacio amplio y luminoso, interactuando entre sí.
Videojuegos, Branding, Interiores, Creación gráfica, Producto, Ciencia y Tecnología, Moda, Animación & UX

What to study if you’re into design and technology: a guide to finding your field of study

31 August 2026

If you’re interested in both design and technology, you don’t need to choose between creativity and technique, but rather to identify the type of problem you want to tackle. Disciplines such as Fashion, Interior Design, Product Design and UX, Visual Communication, Animation, Video Games, and Data Science and AI apply varying degrees of both tools depending on whether they focus on the body, space, use, meaning, time or systems.