When AI devours software

In 2011, Marc Andreessen prophesied in the Wall Street Journal that "software is eating the world." He was right. In a decade, every industry, from finance to agriculture, has been transformed by lines of code. But at the dawn of this new year, that quote already seems like ancient history. A new force has entered the fray. As the leading figures of Silicon Valley now point out, the new reality is clear: "AI is eating software." This development marks a turning point for the digital economy. We are moving from an era where humans used software as a tool to an era where AI generates, replaces, and drives software.

Insight: written by Delphine Seitiée

The central challenge of the interface: towards a new universal OS

The first area of this transformation is at the interface level. The goal of digital giants is no longer just to offer a conversational assistant, but to become the operating system of our digital lives, making traditional applications virtually invisible.

• The dream of the Super App (OpenAI): with the rollout of generative search engines (such as SearchGPT) and developments around proprietary browsers, the ambition is clear: to create a unified experience. Why navigate between ten apps when a single conversation window can orchestrate all these tasks?
• The integration strategy (Google): Google's approach is equally significant. By injecting AI into the heart of office suites and mobile operating systems, the long-standing player is not asking users to change tools. Thanks to its ability to analyze massive amounts of data, AI already processes our emails and documents to automate office work, profoundly changing how we use traditional software.
• The emergence of "Assistant Browsers" (Perplexity): New players such as Perplexity and its Comet browser offer a different vision. These tools position themselves as "Personal Assistants" that browse for you. Instead of displaying a list of links, they synthesize information and perform complex tasks. The interface fades into the background in favor of a direct response.

The end of technical barriers

It is the very structure of value creation that is changing. Jensen Huang, CEO of Nvidia, summed up this change at the World Government Summit: "The programming language of the future is human language."

In concrete terms, access to technology is becoming simpler. Until now, creating a digital tool required interpreters (developers) capable of translating an idea into complex computer code. Today, AI reduces this intermediary. All you have to do is describe what you want in French or English to generate an application or script. The value no longer lies in technical mastery of syntax, but in the ability to define the right problem and imagine the solution.

A reminder from science: intelligence is not (yet) the world

While the industry is getting excited about software transformation, the founding father of the discipline is calling for a more nuanced approach. Yann LeCun, Turing Award winner and former director of AI research at Meta, regularly reminds us that current language models are not a panacea.

His argument is scientific: current models master syntax, but do not understand physical reality. They lack "common sense." LeCun advocates for a change in architecture toward "World Models" and goal-oriented AI.

In concrete terms, tomorrow's AI will not only have to produce probable text, but also be able to plan, reason, and understand real-world constraints before acting. This distinction suggests that software will not disappear, but will become more robust. We will still need engineers to validate that what AI "imagines" is physically possible.

🇨🇭 Alp ICT's perspective: the Swiss map of "physical"

For the economic fabric of Western Switzerland, this paradigm shift is both a threat to the traditional SaaS model and a major opportunity.

Yann LeCun points out that current AI lacks physical "common sense." However, the Swiss economy is firmly rooted in reality: precision robotics, watchmaking, medtech, energy infrastructure. Unlike a chatbot that can hallucinate a poem with no consequences, AI that controls a machine tool or drug dosage in a hospital has no room for error.

The opportunity for our SMEs could therefore take a different form: rather than trying to compete with international giants in the field of text generation for the general public, they could become champions of "last mile" AI.

The developer of tomorrow in French-speaking Switzerland may no longer be someone who writes standardized code, but rather an architect who knows how to combine the generative power of models with Swiss engineering rigor. The challenge is to create reliable, certified, and secure "World Models." AI may be devouring "talkative" software, but it needs experts in the field more than ever to confront reality.