Satya Nadella explained that these systems offer greater performance in generating new code than in editing or correcting existing code.
Microsoft CEO Satya Nadella revealed during a Meta event that between 20% and 30% of the code stored in the company's internal repositories has been generated by artificial intelligence.
This data was shared in a public conversation with Mark Zuckerberg during LlamaCon and confirms an ongoing transformation in the way software is developed at one of the world's leading technology companies.
Nadella explained that the company constantly analyzes so-called acceptance rates, an indicator that measures how many of the AI-generated code suggestions are ultimately adopted by human developers.
These rates currently range between 30% and 40%, suggesting a significant integration of automated tools into programming workflows.
According to the executive, these systems offer greater performance in generating new code than in editing or correcting existing code.
Nadella also acknowledged that the effectiveness of AI varies depending on the programming language. While languages like Python are highly compatible with current models, others like C++ have greater limitations. Despite this, he emphasized that legacy C++ code remains functional and stable.
Future Projections
Microsoft's chief technology officer, Kevin Scott, has projected that by 2030, up to 95% of code could be generated by artificial intelligence.
This estimate anticipates a paradigm shift that would affect both the organization of work and the training and profile of professionals involved in software development.
At the same meeting, Mark Zuckerberg, CEO of Meta, noted that he does not yet have concrete figures on the use of AI in his company's systems, but anticipated that "next year, probably half of development will be done by AI instead of people."
Job Impact: What
IBM's CEO Says About AI
IBM CEO Arvind Krishna offered a different perspective during the SXSW festival, stating that AI will not eliminate the need for human programmers. For Krishna, these tools can increase developers' productivity, but they will not replace them.
The executive disagreed with estimates such as those of Dario Amodei, CEO of Anthropic, who has suggested that AI could generate up to 90% of code in the near future. Krishna estimated that a more realistic figure would be between 20% and 30%, depending on the type of task and the environment in which the technology is used.
According to the IBM CEO, the role of artificial intelligence should be understood as a complement intended to optimize processes and facilitate product development, without completely replacing human labor.
Technical Limits and Remaining Challenges
Krishna also addressed the current limitations of artificial intelligence models. He noted that these systems learn from existing information, such as text or graphics, but cannot formulate new questions or anticipate unknown phenomena.
In his words, “the current generation of artificial intelligence does not take us to what is called artificial general intelligence,” referring to a hypothetical form of AI with cognitive capabilities comparable to humans.
While he acknowledged the value of research led by companies like OpenAI and DeepSeek, he insisted that smaller, more efficient models could be sufficient to drive useful projects.
This vision contrasts with the trend toward large-scale models, which require high levels of energy and computational resources
The projections and analyses of figures such as Nadella and Krishna agree on one essential point: collaboration between humans and intelligent systems will be key to technological development in the next decade.
No comments:
Post a Comment