This is how Mario Rodríguez, CPO of GitHub, sees the future of programming

Almost five years ago we asked ourselves Why program when a machine could do it for you?. It was July 2021 and GitHub Copilot was launched, the first major AI assistant that also boasted of being powered by GPT-3. That was quite a turning point for the world of developers, and since then we have experienced the explosion of a segment that has been the first to test the honeys of generative artificial intelligence.

Among those who were at the forefront of that development is Mario Rodriguezan engineer born in Cuba but who emigrated to the US when he was 14 years old. After studying at the University of Miami, Rodríguez began working at Microsoft, and has developed his entire professional career there. In 2018, following the acquisition of GitHub by Microsoftjoined the management team as vice president of product.

Since August 2024, he has been its Chief Product Officer, and therefore he is the one who decides where GitHub goes as a platform. It is an enormous responsibility considering that we are dealing with the collaborative platform that has become the social network for programmers on its own merits.

A few days ago we had the opportunity to sit down to talk with him precisely to talk (“in Spanish, I prefer it, that’s how I practice it”) about the present and especially the future of GitHub, now totally involved in the generative AI revolution.

The competition tightens

Github Copilot was an absolute pioneer in normalizing that code generation support between 2021 and 2023, but the absolute dominance that seemed to have with the appearance of Cursor and, later, in mid-2025, with the release of Claude Code by Anthropic.

Both AI agents have not stopped growing since then, and the popularity is moving apparently to these new platforms although GitHub Copilot still has an exceptional market share in this segment. If we talk about Claude Code, things are even more striking, because his success is such that even Microsoft engineers themselves they have been using it instead of using the company’s own alternative.

The situation was so unique that Microsoft has ended canceling your Claude Code licenses to force their engineers to use Github Copilot, although there is a strong financial argument here: heavy use of Claude Code was becoming too expensive. Microsoft executives recently stated in The Informationwere very concerned about the erosion of their leadership.

Rodríguez is clear that now there is more competition, but clarifies that “we knew that was going to happen“. Not only that, because he added that “competition is good. “It’s exciting for me to wake up every day and see what we have to do to continue leading.”

But GitHub, as he explained, is much more than GitHub Copilot, “it is a platform in itself.” That doesn’t mean they don’t continue to push that part, and in fact in May GitHub announced the launch of the preliminary version of GitHub Copilot App, which, as Rodríguez explains, solves a gap because Cursor or Claude Code (among others) offered “the Integrated Desktop Environment (IDE), which is what we didn’t have.

Beyond the model: why GitHub’s strategy is not to compete in pure AI

At the moment the situation is what it is: OpenAI has its AI agent for programming, called Codexbut it also develops one of the best frontier models in the world, GPT-5.5. Google, the same: it has Antigravity as an IDE, but it also has models like the recent one Gemini 3.5 Flash.

Anthropic is not short, of course: it has Claude Code as an AI agent, but it also has its Claude Opus 4.7 model as a very clear reference in the field of programming and agentic software engineering. Even Cursor, which initially only had its AI agent to program, has ended up launching a surprisingly good model in programming tasks, Composer 2.5.

GitHub has the tool, but not own model.

For Rodríguez this is not a problem at all, because he sees GitHub as something that goes beyond the modelas a native platform for collaboration in development tasks. “For me the code repository is like a garden that is alive and there are always AI agents collaborating with the human in that repository. So, when you change one thing, people say, ‘Oh, you changed it, this has to change.'”

In fact, although GitHub Copilot appeared with OpenAI models as the main protagonists, today it is a multi-model platform that works with cloud models but also with local models. Actually Microsoft does have own models like MAI“but our strategy is not the model. Where we believe the value is in the systems themselves, not in the models.”

In fact, he pointed out, in the model segment things change too quickly. “Tomorrow the best will be OpenAI, the next day Anthropic, then it may be an Open Source model… what’s the difference? Every day it changes, and differentiating at that layer is very complicated, so where we are going to differentiate ourselves is in the platform itself, in our AI agent platform.”

For him, GitHub’s role is differentiating because it is not an IDE or a model, but a platform. One that not only provides tools to share code and work with it, but also focuses on what he calls “macrodelegation and microsteering“(“macrodelegation and microdirection”).

Macrodelegation is high-level autonomy, which makes the developer focus not on looking at each line of code, but on the results. Microsteering is the constant control to correct course, having a human being in the loop (human-in-the-loop) so that errors can be avoided and micro adjustments made.

These are the options that GitHub proposes for the future, and they also focus it on two crucial tasks: “For all this to work, GitHub has to offer a verification and validation platform. This way you don’t have to look at every line of code, but rather you check that the application is doing what you wanted.” The developer is no longer both a developer and an orchestrator.

Engineers will now focus much more on evaluating the application in controlled environments, checking whether the behavior is as expected, auditing performance tests or monitoring what happens in production environments.

Tips for future developers

Given this radical change in software development, the inevitable question arises for the new generations. Does it make sense to go to university today to learn how to program?

Rodríguez firmly defends that exceptional engineers are those who They master the fundamentals of a technology at a very deep level. “If I am an electrical engineer and I am designing a circuit,” he explained, “I have to learn physics and I have to know chemistry too. And the more I know, the better I can make that circuit.”

Thus, a developer needs to understand the lower layers of the system he orchestrates. Knowing in-depth network architecture or the ins and outs of memory management in languages ​​like Rust gives the professional an advantage when guiding AI agents, he stated.

His academic advice for young people who want to program focuses on understanding the real and physical world, knowledge of critical infrastructures and a multidisciplinary profile: the future belongs to those with transversal knowledge, he assures. citing philosophy from legendary investor Charlie Munger.

The massive proliferation of software projects that clone each other suggests that currently ideas may be the only bottleneck. This manager does not agree with that premise: ideas have always been abundant, he explains, and the true value continues to reside in human judgment.

Here AI mitigates the time cost of experimenting, but is not able to predict whether an idea and its execution will be a success or not. Rodríguez made an analogy with Porsche: the brand stands out for its exquisite criteria in engines, but that does not prevent massive brands like Ford from flooding the market with functional proposals.

For him, software development is becoming more malleable and customizable. A developer can build a Markdown editor exclusively designed for their everyday needs, and that is a success in itself even if they don’t aspire to become a millionaire with that project.

As was the case with traditional app stores before the arrival of AI, just the first handful of tools With an extraordinary design and utility they will be a success. The rest of what is usually seen on social networks has more components of technical exhibition than long-term usefulness. “For me what has changed is how quickly I can test an idea,” he concluded.

GitHub wants to empower one billion developers

The ability of generative AI agents to program has made more and more companies delegate programming tasks to these systemsthus replacing the work of human developers. We therefore wanted to address this delicate issue of an industry that seems to have a clear impact on the labor market. Or maybe not.

GitHub maintains a firm position here regarding the nomenclature of its product: they call it Copilot for a reason: with that name they make it clear that the human being must remain at the center of the technological stage. While some predict a devastating outlook for engineering jobs, on GitHub they point to a demographic explosion in this sector.

In fact, its objective is to ensure that GitHub becomes the way for there to end up being billion developers around the planet:

In my opinion, what you are going to see is a renaissance of developers in the world. I mean, you are going to be a developer now, I am going to be one (…). If before you only worked on products, now you will be able to be a developer, you can be a programmer too.

That is already happening, of course, and GitHub wants to be the platform where it happens because one thing is certain: practically Anyone can create apps without knowing how to program. Just three years ago that was simply unthinkable.

In Xataka | AI came into our lives under a “freemium” model: GitHub and Claude are clear that the future is paying for it