Salesforce previewed something on 10 September that has less to do with Salesforce specifically than the headline suggests. The Trusted Enterprise AI Harness is built around a new AI Control Plane, a layer that discovers and registers agents and AI capabilities, establishes identity and policy, manages lifecycle, evaluates performance, observes behaviour and outcomes, and controls cost, across Salesforce's own agents and third-party AI alike. The detail enterprise architects should actually be paying attention to is that Salesforce built this because, as VentureBeat put it, companies already run an average of three agent platforms, and nobody is governing all of them consistently. That is an architectural problem, not a Salesforce problem, and it exists in your environment whether or not Salesforce is one of your vendors.
Enterprise architecture has spent the past year absorbing the reality that agentic AI does not arrive from a single vendor. Organisations pick up agent capability from their CRM, their productivity suite, their coding tools and increasingly from purpose-built agent platforms, each with its own identity model, its own policy engine, and its own observability tooling. A control plane that sits above all of them, rather than inside any one of them, is the architectural pattern this fragmentation has been pointing toward. Salesforce's announcement is one vendor's specific implementation of that pattern. It is not the only one that will exist, and it will not be the last.
What a Control Plane Actually Does That a Single Platform Cannot
The six capabilities Salesforce named, Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security and Trusted Models, map onto functions that any serious multi-agent environment eventually needs regardless of vendor: a shared understanding of who the customer or business context actually is, control over what an agent is authorised to do, a record of what it actually did, policy enforcement, security controls, and governed access to the underlying models. Built inside a single platform, these functions only govern that platform's own agents. A control plane built to sit above multiple platforms, accessible through MCP, APIs, Skills and Plug-ins and reaching into environments like Claude, Slack and Microsoft Teams, is attempting to govern the fragmented reality most enterprises actually have.
The Important Caveat Most Coverage Is Skipping
Salesforce's own framing makes clear this is substantially a roadmap announcement. New capabilities and a unified experience are planned to begin rolling out in early fiscal 2028, not now. Enterprise architects reading this as an available product to adopt immediately are working from an incomplete picture. The more accurate read is that Salesforce has previewed the direction it intends to build in, which is useful signal for planning, but not yet something to build a current reference architecture around as a delivered capability.
Why This Matters Even If You Never Buy It
The specific product timeline matters less than the pattern it validates. A major enterprise vendor building a control plane explicitly designed to govern agents across other vendors' platforms is a strong signal that the "one platform governs everything" assumption many organisations have been quietly making is not going to hold. Enterprise architects should treat multi-platform agent governance as an architectural requirement to plan for now, independent of whether Salesforce's specific implementation, or a competitor's equivalent, ends up being the one an organisation adopts.
What to Actually Do About It Now
Rather than waiting for a specific vendor's control plane to mature, the more useful immediate step is auditing how many distinct agent platforms are already live inside the organisation, and whether identity, policy, observability and cost control are currently consistent across them or fragmented per platform, which is the more common state. That audit produces the actual requirements a future control plane, whichever vendor builds the one an organisation adopts, will need to satisfy.
What This Means for Your Organisation
What we see across the enterprise architecture engagements we run is that most organisations already have the fragmentation problem Salesforce is describing, several agent platforms live, each governed differently, well before any single vendor's control plane is mature enough to actually solve it. The organisations getting ahead are documenting that fragmentation now as an architectural requirement, rather than waiting for a specific vendor's 2028 roadmap to arrive and solve a problem that is already live inside their environment today.
Key Takeaways
- Salesforce previewed a Trusted Enterprise AI Harness on 10 September, built around a new AI Control Plane that registers agents, sets identity and policy, manages lifecycle and controls cost across both Salesforce and third-party AI.
- The underlying pattern, a control plane governing agents across multiple platforms, responds to the reality that companies already run an average of three agent platforms with inconsistent governance across them.
- Salesforce's own timeline places full rollout at early fiscal 2028, meaning this is currently a roadmap signal rather than an available product, an important distinction enterprise architects should not skip.
- The practical step now is auditing how many agent platforms are already live and how consistently identity, policy, observability and cost control are applied across them, independent of which vendor's control plane an organisation eventually adopts.
How Trusenta Can Help
Enterprise Architecture audits how many distinct agent platforms are already live inside an organisation and documents where identity, policy and observability are fragmented across them.
Enterprise Architecture in the AI Era builds the reference architecture requirements a future control plane will need to satisfy, independent of which vendor's implementation an organisation eventually adopts.
Fractional Enterprise Architect gives organisations without a dedicated architecture function the ongoing judgement to evaluate control plane offerings as they mature toward general availability.
Conclusion
Salesforce's AI Control Plane is not yet something to adopt. It is a signal about where every serious enterprise agent environment is heading, toward governance that sits above individual platforms rather than inside any single one of them. Enterprise architects who start documenting their own multi-platform fragmentation now will be ready to evaluate whichever vendor's control plane, Salesforce's or otherwise, actually arrives first in a mature, generally available form.
