March 27, 2026
Salesforce Reveals Its AI Game Plan with the Launch of AI Foundry
Salesforce AI Research has introduced AI Foundry, a new initiative that brings together researchers, strategic customers, and academic partners to develop and validate AI capabilities at a pace the company says would be impossible through a conventional product cycle.
The launch also gives the market a window into Salesforce’s broader AI strategy. The three stated areas of investment for AI Foundry reveal where the company believes the future of enterprise AI lies. It has also signalled a fundamental shift in its focus on AI development away from competing at the model level and towards building systems in which multiple AI components work together at scale.
Beyond the Model
For more than a decade, Salesforce says it has witnessed progress in artificial intelligence measured predominantly by model performance. The best solutions were the largest, fastest, and most capable models. Salesforce AI Research contributed to this too, having developed predictive models for customer behaviour and generative tools for developer productivity. The company now argues that frontier models are approaching commodity status, and that the more pressing problems for enterprise customers exist not within individual models but across the systems those models power.
Silvio Savarese, Chief Scientist at Salesforce, connects this industry transformation to its AI Foundry launch: “The problems that matter most for businesses don’t live at the model level anymore. 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.”
Key AI Investment Areas
Simulation environments: The first area Salesforce is backing is simulation environments. Enterprise AI agents require exposure to realistic, high-pressure business scenarios before they reach production, and Salesforce has already built ‘eVerse’ to do this. eVerse is a simulation environment that puts agents through thousands of edge cases, handoffs, and complex judgement calls. It has already been used to stress-test Agentforce Voice across large volumes of simulated conversations, and to support a billing agent pilot at UCSF Health’s contact centre.
Ambient intelligence: The second is ambient intelligence; context-aware, proactive AI that surfaces relevant information at precisely the right moment without overwhelming the people it serves. The goal, as Salesforce frames it, is AI that operates continuously in the background of enterprise workflows without requiring constant instruction or generating information overload.
Agent-to-agent ecosystems: The final area is agent-to-agent ecosystems. This is the infrastructure required for AI agents to operate across organisational boundaries securely and autonomously. Salesforce is building what it describes as an enterprise multi-agent semantic layer, incorporating standardised protocols, guardrails, decision logging, and coordinated escalation. The company’s legal team and its Office of Ethical Use of Technology are also involved in defining the legal frameworks that autonomous agent negotiation will require.
The Way Ahead
Enterprise AI operates under requirements that consumer applications never face. Complexity, trust, and operational reliability at scale are demands that benchmark scores do not capture, and that off-the-shelf model performance alone cannot meet. Salesforce argues that AI Foundry is designed to address precisely this gap, not by chasing benchmark improvements, but by building the systems, protocols, and validation infrastructure that make AI agents viable for business at scale.
The announcement arrives as Salesforce continues to build out its broader agentic ecosystem, despite investor concerns about AI. The company has recently deployed AI agent teams into its sales product, brought voice, CRM, and contact centre capabilities together under Agentforce Contact Centre, and acquired a string of AI companies to strengthen its Agentforce offering. AI Foundry positions the research layer as the engine underneath all of it, with simulation, ambient intelligence, and inter-agent collaboration as its strategic priorities.
