As frontier AI models become increasingly commoditised, the “arms race” in technology is shifting. While 2025 was the year of the LLM, 2026 is becoming the year of the System. As such, Salesforce AI Research has launched the AI Foundry.
This is a new initiative, specifically designed to bridge the gap between “what a model can do” and “how a system performs in production.” For manufacturing and technology leaders , this marks a critical turning point. Specifically, AI is moving from a standalone tool to a deeply embedded, autonomous layer of the business.
The System-Level Shift: Why Models are No Longer Enough
For more than a decade, AI progress was measured by the size and speed of individual models. However, Silvio Savarese, Chief Scientist at Salesforce, explains, the hardest challenges in enterprise AI cannot be solved by a better chatbot alone. Such challenges include autonomous agents communicating across organisational boundaries.
“The problems that matter most for businesses don’t live at the model level anymore,” he says. “They live at the system level, where components work together to deliver accuracy, consistency, and reliability at scale.”
Notably, this aligns with The Growth Hub’s State of Sales 2026 findings, which highlighted that while 63% of firms have implemented AI, many are struggling with a “reality gap” in complex areas like forecast accuracy and deal velocity. AI Foundry is Salesforce’s direct answer to that gap.
The Three Pillars of the AI Foundry
The initiative is focusing its research on three strategic areas that will define the next phase of revenue growth:
1. Simulation Environments (eVerse)
Before an AI agent interacts with a real customer, it needs to be stress-tested. AI Foundry has developed eVerse, a simulation environment that exposes agents to thousands of edge cases and “judgment calls.” This has already been used to pilot Agentforce Voice and manage complex billing conversations for UCSF Health, ensuring consistency before the first “real” word is spoken.
2. Ambient Intelligence
The goal here is AI that “disappears” into the background. Instead of a user having to prompt a tool, Ambient Intelligence is context-aware and proactive, anticipating a salesperson’s needs or surfacing a just-in-time insight during a negotiation without creating information overload.
3. Agent-to-Agent Ecosystems
Perhaps the most ambitious area of investment is the development of protocols that allow AI agents from different companies to “talk” and negotiate with each other. Salesforce is working with legal and ethical teams to define the frameworks required for autonomous agent negotiation, essentially creating a digital “handshake” across organisational boundaries.
Speed is the New Currency
In short, the traditional product cycle is too slow for the current pace of change. By connecting foundational research directly to strategic customers, AI Foundry aims to move innovation into the Salesforce product roadmap faster than ever before.
Therefore, as our Growth Guru recently noted, the difference between “Quicksand” and “Quicksilver” in 2026 is the ability to enable your team with systems, not just tools.
Salesforce’s move to the “Foundry” model suggests that the future of revenue growth won’t be found in a better prompt, but in a more resilient, autonomous system.
