Will AI rewrite your business?
What if AI's biggest revolution lies in the way we use it to build our own tools? This is the starting point for the thoughts of Andrej Karpathy, one of the most influential voices in the global AI ecosystem (ex-Tesla, OpenAI).
Insight: written in collaboration with Stéphane Fallet, Founder of Rapid Rise & President of AI Swiss
Andrej Karpathy's 3 ideas for Swiss SMEs
His vision of the future of software seems to us to offer a powerful insight. In a field whose frontiers are receding every month, his ideas are not certainties, but benchmarks for thinking about action today. That's why we wanted to explore three of his key concepts to help SMEs in French-speaking Switzerland think strategically.
The concept: the advent of "Software 3.0
Karpathy's fundamental idea is that we are entering the era of "Software 3.0", when computers are programmed in natural language. After "classic" code (Software 1.0) and neural networks (Software 2.0), humans can now build a tool simply by describing their needs to the machine.
This revolution, embodied by platforms such as ChatGPT and Copilot, makes software creation accessible to all. In Switzerland, the number of uses is multiplying: employees with no technical background are now designing application prototypes, automations or data analysis. The role of the developer is becoming that of an orchestra conductor, steering the AI through dialogue and feedback.
Idea 1: Radical democratization through Vibe Coding
Vibe coding is a direct consequence of this change. It enables any employee - marketing, logistics, administration - to materialize their ideas via AI, without depending on the IT budget or rare technical skills.
The figures confirm this trend. Gartner predicts that by 2026, over 75% of applications will be built on low-code or no-code platforms, compared with just 5% in 2021. This trend is already noticeable in Switzerland, where specialized training and services are emerging. By entrusting creation to those who know the field best, SMEs can innovate faster and at lower cost.
Stéphane Fallet's commentary:
From prototyping to production
"Crucially, Vibe Coding is a great tool for brainstorming, creating proofs of concept (POCs) or presenting an idea to decision-makers. However, moving it into production is another matter."
Moving from a personal mini-assistant to a robust enterprise application raises critical questions:
- Maintenance: Who is responsible for updates and maintenance once the tool has been deployed, especially if its initial creator has no technical skills?
- Governance: How do you guarantee data security and compliance, especially when you're creating a strong dependency on an external platform?
- Integration: An application becomes truly powerful when it connects to existing systems (CRM, ERP...). This requires expertise that is much closer to DevOps than simple dialogue with an AI.
Vibe Coding is therefore an exceptional ideation gas pedal, but industrialization requires a clear definition of its scope and anticipation of its technical supervision.
Idea 2: Augmentation, not substitution - Towards the "Iron Man suit
Andrej Karpathy is categorical: AI will not replace humans on a massive scale, at least not in the short term. His metaphor of the Iron Man suit is rather apt: AI acts as a "co-pilot", boosting employee productivity tenfold, without ever eliminating the key factor in their success: human expertise.
Tesla's example speaks for itself: after a decade of R&D, 100% autonomous driving remains a distant goal. On the other hand, man-machine collaboration is already generating major operational gains. For a Swiss SME, this means automating repetitive tasks, speeding up information retrieval and facilitating decision-making, while capitalizing on the unique know-how of its teams.
Stéphane Fallet's commentary:
The pilot remains the expert
"The Iron Man suit only works because Tony Stark is at the controls," Stéphane reminds us. "The AI co-pilot is a great assistant, but he's not infallible. It can make mistakes, 'hallucinate' information or propose solutions unsuited to the context. The vigilance and critical spirit of the human expert are more than ever necessary to validate, correct and guide the AI's work. The aim is not to blindly follow the machine, but to use it to increase its own intelligence."
Idea 3: Prepare for the future by becoming "readable" by machines
The last axis is the most visionary: the advent of autonomous AI agents. Tomorrow, your customers, partners or suppliers could be AIs. To remain relevant in this ecosystem, your company must become "readable" by machines.
In concrete terms, this means :
- Structuring information: Adopt clear formats such as Markdown (a simple markup language for formatting text readable by all) for documentation.
- Open standardized doors: Set up APIs (programming interfaces) so that external services can interact with your data in a controlled way.
- Adopt new standards: Prepare for files like llms.txt (an AI equivalent of robots.txt) that will tell AIs how to interact with your site.
Stéphane Fallet's commentary:
Start with the foundations
"Before we think about AI agents in 2030, we need to make sure that our own systems understand each other in 2025," Stéphane moderates. "Making your company 'readable' starts in a very pragmatic way: structuring your product data neatly, centralizing your documentation, having a clear service catalog. These actions have an immediate benefit for your online visibility (SEO) and your internal efficiency. It's this foundation of clean, organized data that will truly prepare you for the future, whatever that may be."
Food for thought for Swiss SMEs
Karpathy's vision is not science fiction. For Swiss SMEs, three avenues for reflection emerge:
- Experiment, with humility. Try your hand at "Vibe Coding" as a means of prototyping. But keep in mind that while the results may be impressive, they can't yet be industrialized at the click of a button. Integration with your existing systems via connectors (APIs) is the essential first step towards a robust, future-proof solution.
- Augment, not replace. Adopt co-pilots and AI assistants to amplify the quality and speed of your teams. The greatest productivity gain lies in enhancing the value of your human capital, which remains the driver of the tool.
- Structure for the future. Make your information and services accessible and well-organized. It's a competitive advantage that needs to be built today on a sound foundation, even before we start talking about AI agents.
Agility, quality and innovation are in the DNA of the Swiss economy. Turning this corner means strengthening these pillars. In a field that is changing so fast, these principles, relevant in September 2025, offer a compass to build, step by step, your own "Iron Man suit".
