As the AI agent workforce grows, Salesforce offers more control and more predictability in AI operations.
Salesforce is significantly enhancing its approach to managing the burgeoning landscape of AI agents, aiming to provide enterprises with greater control and predictability over their AI operations. The company’s latest suite of updates to its Agent Fabric, built on the MuleSoft AnyPoint Platform, introduces robust governance tools and deterministic controls designed to tame the potential chaos of "agent sprawl" and ensure efficient, secure, and cost-effective AI deployments. This move addresses a critical need for businesses as they integrate multiple AI agents from various vendors, a practice that can lead to redundant workflows, data silos, and operational inefficiencies, ultimately hindering safe and responsible AI scaling.
The challenge of agent sprawl is not new for Salesforce. Last year, the company first attempted to address it with the introduction of Agent Fabric. This initial suite of capabilities was designed as a central hub for enterprises to register, monitor, interconnect, and govern their AI agents. The evolution of Agent Fabric has been a strategic response to the rapidly evolving demands of enterprise AI. Launched in September 2025, Agent Fabric’s foundational goal was to unify disparate AI agents into a manageable ecosystem. This was followed by significant enhancements in January, which included the addition of a deterministic scripting tool and agent scanning capabilities, allowing for the automated discovery and onboarding of new agents into the registry. These advancements were crucial steps in bringing order to the increasingly complex AI environments enterprises were building.
However, as businesses moved beyond initial experimentation and into scaled production, the need for more sophisticated control mechanisms became apparent. Recognizing this, Salesforce is now rolling out a new wave of features designed to provide deeper levels of operational oversight and predictability. These additions are poised to give enterprises more granular command over how their AI agents function, ensuring that AI deployments align with business objectives and regulatory requirements.
Expanding Deterministic Controls for Enhanced Predictability
A cornerstone of Salesforce’s latest offering is the expansion of deterministic controls through Agent Script for Agent Broker. Agent Broker, an intelligent routing service within Agent Fabric, is engineered to connect AI agents across different domains, dynamically matching user tasks with the most suitable agent. The introduction of Agent Script for Agent Broker allows developers to codify workflows within multi-agent systems. This capability is critical for ensuring consistent and reliable outputs, moving away from the inherent unpredictability that can arise when relying solely on probabilistic AI models for decision-making.
Traditionally, probabilistic agents make decisions based on the likelihood of outcomes, which can introduce an element of randomness. Agent Script for Agent Broker provides a mechanism to steer these decision-making processes according to predetermined rules. This approach not only enhances predictability but also offers potential cost savings. By leveraging predefined logic, enterprises can reduce the reliance on computationally intensive large language models (LLMs) for every decision point, thereby optimizing resource utilization.
Robert Kramer, managing partner at KramerERP, views these deterministic controls as a vital development for production AI environments. "Pure autonomous agents don’t necessarily work in production as enterprises need to ensure predictable outcomes," Kramer stated. "The deterministic controls should facilitate a secure handoff of control and rules while still allowing the model to engage in reasoning when it’s appropriate. It’s a balance between control and flexibility, which is the norm for most real deployments." His perspective highlights the practical need for a hybrid approach that combines the analytical power of AI with the assurance of controlled processes.
Rebecca Wettemann, principal analyst at Valoir, echoes this sentiment, emphasizing the strategic advantage of offering both deterministic and probabilistic options within Agent Fabric. She suggests that this flexibility empowers developers and agent builders to pursue the most cost-effective path to achieving accurate and predictable results from their agentic systems. This dual approach caters to a wide range of use cases, from highly sensitive operations requiring strict adherence to protocols to more exploratory tasks where emergent behavior might be beneficial.
Despite the significant potential, enterprises will need to wait a bit longer to implement these deterministic orchestration features in their production environments. Currently in beta testing, the feature is slated for general availability in June 2026, indicating a thorough validation process before wide release.
Centralized LLM Governance Tackles Escalating Costs
Beyond orchestration, Salesforce is bolstering its AI governance capabilities with a new LLM Governance feature within AI Gateway. AI Gateway serves as the central control layer for Agent Fabric, offering consolidated visibility into crucial operational metrics such as token usage, costs, and data flows for third-party AI models. The LLM Governance capability, now generally available, is designed to help enterprises maintain their AI operations within budget, a growing concern for many organizations.
The imperative for centralized AI governance has intensified as CIOs grapple with the proliferation of disparate AI systems and the escalating costs associated with their implementation and operation. Without a unified approach, different departments within an organization might independently select AI models, negotiate their own API contracts, and manage token budgets locally. This fragmented approach can lead to significant cost overruns, inconsistent security postures, and a lack of enterprise-wide policy enforcement.
Scott Bickley, advisory fellow at Info-Tech Research Group, warned about the pitfalls of decentralized AI management. "This results in sprawling costs, inconsistent security postures, and no enterprise-wide policy enforcement," Bickley explained. "By positioning AI Gateway as the choke point through which all LLM traffic flows, enterprises gain visibility into AI usage patterns, the models in use, purpose of the usage, and cost data." This centralized vantage point is crucial for strategic decision-making, allowing IT leaders to identify areas of overspending, optimize model selection, and ensure compliance with internal policies and external regulations. The ability to monitor and control token usage is particularly important, as LLM consumption can quickly become a significant operational expense.
MCP Additions Simplify Integration with Legacy Systems
Salesforce is also enhancing the integration capabilities of its AI agents with the introduction of new Model Control Protocol (MCP) features. Among these are MCP Bridge, designed to facilitate easier access to legacy APIs, and Informatica-hosted MCPs, which aim to streamline how AI agents interact with enterprise data and existing APIs. These advancements are expected to save developers considerable time and simplify the complex task of building cross-environment, multi-agent systems.
For organizations burdened by extensive legacy systems, MCP Bridge offers a compelling solution. Bickley highlighted its significance for enterprises with thousands of legacy APIs built on technologies like REST, SOAP, and GraphQL, predating the advent of MCP. "Agents speaking MCP cannot call those APIs natively, so they require wrappers around the API endpoint, which would be a massive engineering lift. MCP Bridge allows these APIs to be exposed as MCP-compatible tools without modifying the underlying code," he stated. This capability significantly reduces the technical debt associated with integrating older systems into modern AI workflows, making it more feasible for businesses to leverage their existing infrastructure alongside new AI capabilities.
Rebecca Wettemann pointed out the further benefits of Informatica-hosted MCPs, suggesting they will reduce development overhead by embedding data quality and governance capabilities directly into agent workflows. This is particularly critical for enterprises operating in regulated industries or those with heightened risk management concerns, where data integrity and compliance are paramount.
However, Bickley also cautioned that while MCP Bridge offers a valuable shortcut, enterprises should conduct thorough testing to understand how it handles edge cases and the nuances of legacy API behavior. Similarly, he noted that even with integrated data quality and governance from Informatica, these processes are not instantaneous. Data validation, deduplication, and cross-system matching can introduce latency, measured in milliseconds or even seconds, which needs to be factored into application design and performance expectations.
A Strategic Pivot for MuleSoft
These ongoing enhancements to Agent Fabric signal a broader strategic pivot for MuleSoft, which Salesforce acquired in 2018 for $5.7 billion. Industry analysts suggest that Salesforce is repositioning MuleSoft from a traditional API integration platform to a foundational infrastructure layer for enterprise AI agents. By integrating robust orchestration, governance, and connectivity capabilities into Agent Fabric, Salesforce appears to be aiming to establish MuleSoft as the definitive system of record for AI agent discovery, routing, and governance across the enterprise. This elevates MuleSoft’s role beyond mere API management to that of core AI infrastructure.
This strategic reorientation may not be universally welcomed by all CIOs. The increased dependency on Agent Fabric for managing AI agents could lead to higher switching costs. As organizations register more agents and define more orchestration rules and governance policies within the Salesforce ecosystem, migrating to an alternative solution becomes increasingly complex and resource-intensive.
As with any critical infrastructure dependency, CIOs are advised to critically evaluate the long-term implications. Key questions to consider include the exit path from the platform, the portability of Agent Fabric components, the lock-in potential, the pricing model, and the depth of integration with non-Salesforce agents and data sources. A thorough understanding of these factors will be essential for making informed decisions about adopting and scaling AI operations within the Salesforce ecosystem.
The evolution of Salesforce’s Agent Fabric underscores a critical industry trend: the growing need for robust management and governance frameworks to support the widespread adoption of AI agents. By providing tools that balance control with flexibility, and by addressing the complexities of integration and cost management, Salesforce is positioning itself to be a key player in enabling enterprises to harness the full potential of AI safely and effectively. The continued development in this space will likely see further innovations aimed at simplifying AI operations, enhancing security, and maximizing the return on investment for AI initiatives.
