Salesforce AI Foundry: System Reliability Beats Model Power

For a long time, the central narrative in global tech was the model wars, where companies competed to build the biggest AI models. 

Now, business leaders are shifting their focus. 

The most important breakthroughs are no longer about a single AI model’s power, but about whether the entire system works consistently and reliably in real-world operations.

This shift from focusing on an individual model’s performance to its integration across a whole system is the driving force behind Salesforce AI Research’s latest project, AI Foundry. 

For global companies trying to manage the move from experimental lab demonstrations to dependable business tools, AI Foundry aims to connect the foundational research with practical business use.

Looking beyond the model 

The tech industry has seen the transition from predictive analytics to generative productivity tools and, most recently, to autonomous agents. 

But as LLMs reach a state of maturity, a bottleneck has appeared. 

The gap between what an AI model can do in a vacuum and how it performs within a complex corporate workflow has become the primary hurdle for enterprise-wide scaling.

“The problems that matter most for businesses don’t live at the model level anymore,” says Silvio Savarese, Chief AI Scientist at Salesforce. 

“They live at the system level, where components work together to deliver accuracy, consistency and reliability at scale. 

“AI Foundry is the engine we’ve built to make that a reality.”

The architecture of AI Foundry

AI Foundry is structured around three pillars designed to address the unique rigours of the enterprise environment:

1. Simulation environments: Autonomous agents cannot be deployed into live business environments without validation. Through eVerse, AI Foundry provides a high-fidelity simulation environment that stress-tests agents against thousands of edge cases and complex handoffs. 
2. Ambient intelligence: The next generation of enterprise AI aims to be context-aware and proactive without becoming intrusive. AI Foundry is researching ways to embed intelligence directly into workflows, focusing on human-AI interaction patterns that offer just-in-time insights while avoiding information overload.
3. Agent-to-agent ecosystems: Perhaps the most complex frontier involves agents communicating across organisational boundaries. Salesforce is developing standardised protocols, such as agent cards, to create a multi-agent semantic layer. This includes establishing the necessary legal and ethical frameworks for autonomous negotiation, ensuring that as agents act on behalf of companies, they do so within strict guardrails.

Accelerating the research-to-product pipeline

Traditionally, the journey from academic research to a functional software feature is a multi-year process. 

AI Foundry seeks to collapse this timeline by fostering a “rapid iteration” ecosystem that brings together internal researchers, academic partners and customers.

“Many of the old rulebooks simply don’t apply anymore,” says Itai Asseo, Head of Incubation and Brand Strategy for AI Research at Salesforce.

“AI Foundry connects foundational research to real business problems by collaborating closely with our strategic customers in rapid iteration cycles.”

Through AI Foundry, Salesforce is delivering a clear message that competitive advantage in the AI era will not come from having the largest model, but from having the most reliable system.