Salesforce Enterprise AI Harness Puts Proprietary Context at the Centre of Agent Strategy
The Development
Salesforce unveiled the Trusted Enterprise AI Harness at Dreamforce 2026 on September 10, framing it as the foundational architecture for the Agentic Enterprise. The system bundles six capabilities — Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security, and Trusted Models — into a composable architecture drawing on Data 360, Informatica, MuleSoft, Tableau, Agentforce, and Salesforce Guardian. Alongside it, a new AI Control Plane provides a single interface to discover, register, manage, and cost-control agents across Salesforce and third-party systems. The harness is built headlessly, accessible via MCP, APIs, Skills, and Plugins, meaning it can extend into Claude, Slack, Microsoft Teams, and other surfaces. Existing capabilities are available now; new capabilities and the unified experience begin rolling out in early fiscal FY28, which aligns with early 2027.
Our Take
The strategic argument Salesforce is making is precise and worth taking seriously: models commoditise, proprietary context compounds. Every agent interaction that flows through the harness creates new memory and signals that enrich future reasoning — a flywheel that deepens the gap between enterprises that centralise their agent infrastructure early and those that build agent-by-agent. The Claudeforce integration, which brings Anthropic's Claude directly into this architecture, signals that Salesforce is positioning the harness as the context and governance layer for whatever model wins. That is a durable business if it holds. The risk is FY28 availability — early 2027 is far enough out that enterprise buyers have time to evaluate competing architectures from Microsoft and ServiceNow before committing.
What Changed
Enterprises can now operate a single governance and context layer across heterogeneous agent fleets — Salesforce-native and third-party — rather than engineering trust, permissions, and data access separately for each agent deployment. The AI Control Plane adds enterprise-wide observability over agent behaviour, cost, and policy compliance in one interface.
Marketing Impact
Marketing operations and martech teams face the most immediate decision: whether to route campaign agents, personalisation logic, and customer data orchestration through the Salesforce harness or maintain separate agent governance stacks. Teams already deep in Salesforce CRM have a lower switching cost and a clearer upgrade path.
Competitive Implication
Enterprises with significant Salesforce footprints gain a faster path to governed, multi-agent marketing automation than those building on fragmented stacks. Marketing technology vendors selling standalone agent orchestration or customer data platforms outside the Salesforce ecosystem face accelerating pressure to demonstrate comparable governance and context capabilities.
Strategic Outlook
Microsoft's Copilot Studio and ServiceNow's Now Assist are the immediate competitive responses to watch in Q4 2026. Expect Salesforce to use the FY28 rollout window to lock in design-partner commitments from enterprise accounts, making switching costs structural before general availability. Pricing, still unannounced, is the next forcing function.
The Exploit
Action Item
Salesforce enterprise customers running more than three active agent deployments should request harness design-partner access now — before Q4 pricing is announced — to negotiate capability access on current contract terms rather than new FY28 packaging.