Fortarezza (Salesforce): ‘Writing code accounts for just 10% of software development’

They call it the “SaaS Apocalypse”. It is the increasingly widespread belief in Silicon Valley that generative artificial intelligence and so-called vibe coding could wipe out traditional enterprise software. All you need to do is talk to a model, explain what you need, and the application takes shape almost by itself. The financial markets are the first to believe it. Since the start of 2026, the shares of the software giants have lost around double-digit figures. Every announcement from the AI giants seems to feed into the same narrative: traditional software is destined to become obsolete, replaced by AI agents capable of creating applications on the fly thanks to so-called vibe coding.

Yet, whilst Wall Street swings between enthusiasm and fear, the major platforms are trying to reinvent themselves. Salesforce, one of the icons of enterprise software over the last twenty years, is attempting to transform AI from a threat into the new infrastructure of the agent economy. We discussed this with Vanessa Fortarezza, CEO of Salesforce in Italia, a leading figure in global enterprise software, who is now attempting to turn what might seem like an existential threat into a new strategy of embracing AI agents.

The notion of the “SaaS Apocalypse” emerged following an incident in which $285 billion in market capitalisation evaporated in just 48 hours. However, the data tells a different story. A Morgan Stanley survey shows that since 2024, the number of companies continuing to rely on traditional software leaders has doubled. The point is that writing code represents only a small part of the problem: around 10% of the software. The remaining 90% consists of security, governance, compliance and business processes. The value of platforms such as Salesforce lies not only in the lines of code, but in the business logic accumulated over decades of development and in the trust built with customers. Consider GDPR certifications or the ACN QC2 certification for the public sector. A public LLM, on its own, is unable to guarantee this level of reliability and security.

We have decided to fully embrace external LLMs by integrating the MCP protocol across all layers of our architecture. This is a significant step because it enables models such as Claude, ChatGPT or Gemini to access Salesforce data in real time and in a conversational manner, whilst remaining within our security infrastructure. The idea is that these models will become a new interface for accessing business systems, without compromising control and governance.

In Italia, there is a great deal of interest and hundreds of pilot projects have been launched, some of which are already in production. Today, we are seeing a genuine ‘agent gap’. There are companies that remain very cautious and are sticking to the testing phase, and others, such as Unicredit, which are embarking on large-scale agent-based transformations. In regulated markets, such as banking, a gradual approach prevails: the starting point is assistive artificial intelligence aimed at employees, to boost internal productivity. Exposing agents directly to end customers requires very high levels of reliability. Companies want these systems to reach standards close to those of humans before taking on reputational or regulatory risks.