Most businesses running AI in 2026 are not running one AI system. They are running five, ten, sometimes twenty. There is a Salesforce Agentforce agent for sales. A Microsoft Copilot agent handling support tickets. An OpenAI-powered workflow sitting inside a third-party SaaS tool. An Amazon Bedrock agent someone in IT built six months ago and half the team forgot about.

Salesforce is betting this is the biggest unsolved problem in enterprise AI right now: not building agents, but governing them. And its answer is Agent Fabric, which reaches full general availability this month.

What Agent Fabric Actually Does

Agent Fabric is Salesforce’s multi-vendor AI agent control plane. It gives IT and operations leaders a single place to discover, connect, govern, and monitor AI agents regardless of which vendor built them, which cloud they run on, or which team deployed them.

The platform connects to Amazon Bedrock, Microsoft Azure AI Foundry, OpenAI, Google Gemini, GoDaddy, and more through standard APIs and MCP (Model Context Protocol) bridges. An agent running on Bedrock shows up in the same dashboard as one built in Agentforce. You can see what both are doing, set rules for both, and kill either one if something goes wrong.

Four major capabilities landed at the April 2026 TDX developer conference and are now going into GA this month:

Agent Scanner with automated discovery. Instead of relying on someone manually telling the system what agents exist, Agent Fabric now scans connected platforms and MCP servers to find agents automatically. It builds a live registry of everything running across your AI stack. For enterprises that have accumulated agents across acquisitions, departmental projects, and vendor tools, this alone solves a real problem.

Visual workflow authoring canvas. Agent Broker, Salesforce’s orchestration layer, gets a drag-and-drop canvas for mapping how agents hand work between each other, where humans need to approve steps, and how information flows across systems. No code required for basic orchestration.

Guided determinism. This is the governance story. Rules-based guardrails let you define exactly when an agent can act autonomously and when it needs a human checkpoint. High-stakes actions, such as initiating a refund above a certain threshold or modifying a contract, can require mobile-based human authorization before execution proceeds. The agent does not just suggest; it waits. This is a meaningful shift from “AI assists a human” to “AI acts within defined limits and escalates when appropriate.”

Centralized LLM governance. The AI Gateway layer manages model routing, cost controls, and compliance logging across every agent and model in the stack. You can set which models are approved for which use cases, cap token spend by department, and maintain audit trails for regulated environments.

The Number Behind It

Salesforce reported that Agentforce passed $1 billion in annual recurring revenue across 18,500 enterprise customers. That number matters not as a pure revenue figure but as a signal that enterprise buyers are actually deploying agentic AI in production, not just piloting it. At scale, the governance problem Agent Fabric addresses gets worse fast.

The Summer 2026 Salesforce release, available June 15, extends these capabilities further into marketing, revenue cycle management, and multi-agent orchestration across customer-facing teams.

What This Means for Business

Most AI strategy conversations at the executive level right now center on what to buy and what to build. Agent Fabric is a bet that a third question is about to become just as important: how do you manage what you’ve already deployed?

A few things worth tracking as you evaluate multi-vendor AI environments:

You cannot govern what you cannot see. The Agent Scanner capability addresses what is quickly becoming an enterprise shadow-AI problem. AI agents deployed by individual departments or through SaaS vendors often operate without central visibility. Discovering them after something goes wrong is the wrong approach.

Determinism and governance are not the same thing. Guided determinism in Agent Fabric is a step toward configurable autonomy: agents can act freely within defined parameters and pause for human approval outside them. This is more practical than blanket human-in-the-loop requirements, and more defensible than full autonomy.

Multi-vendor AI is the normal state. The assumption that one vendor will provide everything a business needs was probably always unrealistic. Building governance infrastructure designed for heterogeneous AI environments is a more durable strategy than waiting for consolidation that may not come.

The governance layer for enterprise AI has been conspicuously absent from most discussions about agentic AI adoption. Salesforce is not the only company trying to fill it, but reaching GA with a multi-vendor control plane that works across the major hyperscalers puts it in a credible position. Whether the independent observability and governance vendors survive that move is a separate question.

What This Means for Enterprise DNA Clients

If your business is running AI across departments, you likely have agents sitting in tools you have bought, platforms your team has built on, and SaaS workflows that plugged in an AI feature without asking. Getting a clear map of what exists and establishing governance before something goes wrong is not a cautious move; it is the operational foundation everything else depends on.

At Enterprise DNA, our Omni Advisory team helps business leaders audit their existing AI stack, identify governance gaps, and build a coherent operational strategy that does not assume a single vendor. If you are starting to feel the weight of an AI stack that grew faster than your ability to manage it, that is a useful conversation to have now rather than later.

Talk to an Omni Advisor about building an AI governance foundation that scales.