Salesforce Launches 'AI Foundry', Pioneering Enterprise General Intelligence Through Agent-to-Agent Ecosystems
Salesforce Launches 'AI Foundry', Pioneering Enterprise General Intelligence Through Agent-to-Agent Ecosystems
Salesforce has unveiled 'AI Foundry,' a strategic initiative marking a decisive shift from standalone models to interconnected Agent-to-Agent (A2A) ecosystems. By prioritizing 'Enterprise General Intelligence,' the tech giant aims to deliver practical, system-level AI architectures that value consistency over speculative AGI hype.
Salesforce has fundamentally realigned its artificial intelligence strategy with the launch of AI Foundry, an ambitious initiative that pivots the enterprise software giant away from isolated AI models and toward highly orchestrated system-level architectures. Announced in March 2026, the initiative cements a strategic transition into the era of Agent-to-Agent (A2A) ecosystems, framing a new paradigm Salesforce is dubbing "Enterprise General Intelligence" (EGI).
Unlike the speculative pursuit of Artificial General Intelligence (AGI)—which aims to match or surpass human cognition across all domains—EGI is purposefully pragmatic. It focuses on the "boring breakthroughs," ensuring that AI systems are highly capable yet strictly bounded by governance, consistency, and enterprise reliability.
The Core of AI Foundry: Moving to System-Level AI
For the past two years, enterprise AI adoption has been plagued by what researchers call jagged intelligence—the frustrating reality where Large Language Models (LLMs) effortlessly pass complex tests but fail to reliably execute routine corporate workflows. AI Foundry is Salesforce’s direct response to this bottleneck, bridging the gap between raw intelligence and reliable business performance.
Instead of merely deploying larger foundational models, AI Foundry focuses on how discrete AI components work together across departments, data silos, and organizational boundaries. Silvio Savarese, EVP and Chief Scientist at Salesforce Research, noted that pushing the envelope of agentic AI requires moving "along the axis of consistency, accuracy, and trust," identifying EGI as the company's "true north star for 2026 and 2027".
Through AI Foundry, Salesforce is doubling down on three critical trends:
- Agent-to-Agent (A2A) Ecosystems: Secure frameworks where specialized agents collaborate and negotiate without human bottlenecks.
- Simulation Environments: Advanced virtual sandboxes, such as Salesforce's eVerse, that train voice and text agents through synthetic data generation and reinforcement learning.
- Ambient Intelligence: Embedded AI that sifts through massive data streams to surface proactive, real-time insights during human-centric workflows, such as sales meetings.
The Agentic Swarm and A2A Architecture
The cornerstone of this new era is the shift from single-prompt, reactive chatbots to the Agentic Swarm. Under the new A2A architecture powered by Salesforce’s Atlas Reasoning Engine, tasks are distributed autonomously among a hierarchy of agents.
In practice, a central Orchestrator Agent interprets complex human intent and breaks it down into sub-tasks. It then delegates these tasks to specialized Helper Agents—such as a billing agent, a shipping agent, or an inventory agent. These agents communicate using standardized protocols like the Agent-to-Agent (A2A) protocol and the Model Context Protocol (MCP).
This orchestration explains how AI Foundry aims to achieve its ambitious goals:
- Decentralized Execution: Narrow specialist agents cost less to run and hallucinate less frequently because their token budgets and instructions are hyper-focused on single domains.
- Secure Handoffs: The A2A semantic layer establishes strict rules of conduct, ensuring that agent negotiations stay within legal, safe boundaries, overseen by an enterprise multi-agent semantic layer.
- Interoperability: By embracing open standards like MCP and A2A, Salesforce ensures that its internal agents can securely exchange data and context with external third-party agents across partner ecosystems.
Enterprise General Intelligence: The Capability-Consistency Matrix
The guiding philosophy behind AI Foundry is Enterprise General Intelligence (EGI). Salesforce evaluates EGI through its Capability-Consistency Matrix.
While raw capability handles complex horizontal reasoning, consistency guarantees that an agent adheres to organizational governance and privacy frameworks. A highly capable but inconsistent agent is an enterprise liability. By routing interactions through systems like the Einstein Trust Layer—which strips Personally Identifiable Information (PII) before it hits an LLM—Salesforce grounds autonomous operations in verifiable safety.
Furthermore, AI Foundry employs rigorous, continuous simulation. Using the eVerse framework, agents are stress-tested against synthetic, highly complex scenarios such as spotty data connections or frustrated customer tones. This allows the AI to fail, learn, and optimize in a closed environment before it is ever deployed into live production, mitigating the performance saturation observed in standard LLMs.
Why This Matters for the Future of Work
Salesforce’s introduction of AI Foundry and EGI is a clear signal that the enterprise AI market is reaching a new level of maturity. The focus has decisively shifted from "what can a model generate?" to "how can an AI system autonomously and reliably execute a multi-step business process?".
By building the infrastructure for autonomous, multi-agent collaboration, Salesforce is not just upgrading its CRM; it is laying the digital plumbing for a new kind of enterprise architecture. Organizations that master A2A deployments will transform their workforces, shifting human employees away from routine task execution and toward managing, mentoring, and auditing high-performing agent swarms.