Analyst(s): Mitch Ashley
Publication Date: October 30, 2026
Salesforce expanded MuleSoft’s Agent Fabric on October 27, 2026, adding production evaluation, automated intervention, and hard AI cost controls for agents running across platforms. The update pushes Agent Fabric toward a cross-vendor control layer for enterprise agent estates, reaching agents, models, and MCP servers beyond Salesforce’s own stack. For CIOs, the real decision is which control plane to standardize on before the layer multiplies.
What Is Covered in This Article:
- The capabilities and rollout milestones Salesforce detailed for Agent Fabric, including Agent Evals, API Evals, the Custom Agent Scanner, and Open Policies for Federated API Gateways.
- The new AI cost controls: a FinOps Dashboard, Model Wallet spending limits, and Cost Optimization Policies.
- The intervention model, including the Agent Kill Switch and its triggered variant.
- The availability timing across beta and general availability dates.
- Futurum’s read on what the expansion means for enterprise buyers and the CIO.
The News: On October 27, 2026, Salesforce expanded Agent Fabric, the MuleSoft offering it describes as an AI control plane for managing and governing agents across platforms. Andrew Comstock, SVP and GM of MuleSoft and Agent Fabric, framed the update as combining evaluation, cost control, and intervention in one layer.
The release adds Agent Evals, which generates tests before deployment and then evaluates agents, large language models (LLMs), and Model Context Protocol (MCP) servers in production against quality thresholds, triggering predefined interventions, including an Agent Kill Switch, when results fall short. API Evals generates functional and performance tests from API specifications to catch breaking changes before production. A Custom Agent Scanner extends discovery to systems behind firewalls and third-party platforms, and Open Policies for Federated API Gateways enforces a single policy across Azure, Apigee, and Kong gateways. On cost, the release includes a FinOps Dashboard, a Model Wallet that sets hard spending limits that stop requests before a budget is exceeded, and Cost Optimization Policies that it says cut token usage by up to 75 percent in test scenarios.
Agent Kill Switch, Open Policies for Federated API Gateways, and the three cost controls are generally available now. Agent Evals and API Evals enter beta in November 2026, with general availability set for January 2027; the Custom Agent Scanner is in beta now, with the same January 2027 general availability target. The Triggered-Agent Kill Switch, which links evaluation findings to automatic intervention, is set for general availability in January 2027. ASICS Digital is helping shape the Agent Evals beta; Capita, GE Appliances, and Newell Brands provided testimonials describing the use of Agent Fabric, though the announcement does not establish the production scope or adoption status of the newly announced capabilities at those organizations. Salesforce says Agent Fabric handles over 1.2 billion requests daily, without specifying customer versus internal volume. Full details are in Salesforce’s announcement.
Salesforce Agent Fabric Expands to AI Control
Analyst Take: Salesforce is making a bid to CIOs to be the enterprise AI control plane, the governed layer above agents, no matter which platform built them. This is a positioning move first and a feature release second. Agent Evals scores non-Salesforce LLMs and MCP servers, the Custom Agent Scanner lets enterprises extend discovery to agents behind firewalls and on third-party platforms, and Open Policies spans Azure, Apigee, and Kong rather than MuleSoft alone.
The broader scope answers a problem enterprise buyers are already grappling with. Running agents on more than one platform is increasingly common, and each platform comes with its own console, policy model, and bill. The result is control-plane debt: integration, governance, and cost obligations that accumulate faster than any single team can service. Agent Fabric is positioned to consolidate that sprawl into a single governed layer.
Figure 1: AI Spend Relative to Budget, 2H 2026
The AI Control Plane Is the Hotly Contested Layer
Agent Fabric is itself a control plane, and that complicates the equation. Microsoft already ships Agent 365 as a cross-platform agent control plane, and ServiceNow runs its AI Control Tower, so the layer is already multiplying. As Google, AWS, and the agent-platform startups press their own versions, the debt re-forms one level up. Rather than govern agents centrally, buyers must decide which control planes to standardize on and whose reach across rival stacks is real, not merely demonstrated at product pitches and proof-of-concept pilots.
MuleSoft makes a strong pitch to CIOs, positioning Salesforce as a credible contender for that cross-vendor seat. Integration is the heritage, and the durable control is in governing the connector layer- what an agent is allowed to reach. Cataloging and policing MCP servers and APIs across federated gateways constitute governance at the connector layer, providing the leverage needed to scale agentic deployment. The risk is that the reach narrows back to Salesforce-centric estates once the beta features reach general availability in January 2027.
Discovery and Cost Controls Target the Governance Gap
The Custom Agent Scanner addresses the enterprise exposure of unregistered agents. Origination has dispersed to everyone building with AI, while accountability for securing and governing what they build converges back on IT. Scanners that enterprises build to reach agents behind firewalls and on third-party platforms are a direct answer to that gap, and the risk it exposes surfaces first in audit and incident response, long before procurement sees it.
The cost controls help manage AI spend as a metered variable cost. Model Wallet stops requests before a budget is breached, and the FinOps Dashboard gives budget owners a view that they have mostly lacked.
Futurum’s 2H 2026 CIO & Technology Buyers Decision Maker Survey found 46.9 percent of enterprises over budget on AI and only 5.6 percent under plan, so the appetite for hard limits is real (Futurum Research, August 2026). Salesforce’s up-to-75-percent token reduction comes from test scenarios, so treat it as a claim to verify against customer numbers rather than a delivered result.
For the CIO, Agent Fabric is a platform and governance decision. It sits with the integrator-and-security side of the role, and it rewards buyers who consolidate toward a few layer-dominant control planes instead of letting every platform install its own. The question it forces is whether a governance and cost-control layer now earns its own named budget line, owned by IT, or remains buried within integration and observability spend, where no one is accountable for the total.
What to Watch:
- Whether the competing control planes, including Microsoft Agent 365 and ServiceNow AI Control Tower, move toward cross-vendor enforcement and interoperability, or leave buyers running overlapping registries, policies, and cost controls.
- Whether the January 2027 releases, Agent Evals, API Evals, and the Custom Agent Scanner, deliver the promised evaluation and discovery coverage across non-Salesforce agents, models, and gateways, or the reach narrows to Salesforce-centric estates in practice.
- Whether enterprises fund a named governance and cost-control budget line owned by the CIO, or leave agent oversight inside existing integration and observability spending.
For the capability details and availability dates, see Salesforce’s full Agent Fabric announcement on their website.
Disclosure: Futurum is a research and advisory firm that engages or has engaged in research, analysis, and advisory services with many technology companies, including those mentioned in this article. The author does not hold any equity positions with any company mentioned in this article.
Analysis and opinions expressed herein are specific to the analyst individually and data and other information that might have been provided for validation, not those of Futurum as a whole.
Other Insights From Futurum:
SUSE OpenShell Agent Security: Enforcement Is Table Stakes, Proof Is Not
With Kong Volcano, Kong Hosts the Agents It Governs
Docker Reruns the Container Playbook, This Time for Cloud Sandbox Kit
Author Information
Mitch Ashley is VP and Practice Lead for the CIO & Technology Buyers and Software Lifecycle Engineering practices at The Futurum Group. A multi-time CIO and CTO with 30+ years leading technical organizations, Mitch built and operated production systems spanning cybersecurity for the U.S. Department of Defense, PKI services for the broadband and 5G industries, SaaS platforms, large-scale telecom and banking systems, and a national broadband network. His work with AI began early, developing expert systems that diagnosed and repaired complex mainframe environments. That operator foundation grounds his analysis in operational consequence, covering the technology buyer's world of software engineering, cybersecurity, DevOps, cloud, and AI.
