Salesforce is making a major bet on the next phase of artificial intelligence: as powerful AI models become easier for companies to access, the real competitive advantage may shift from who has the smartest model to who has the most valuable business context — and the strongest control over what AI is allowed to do.
The cloud software giant has unveiled its Trusted Enterprise AI Harness, an architecture designed to connect artificial intelligence agents with a company’s customer data, business processes, security policies and operational systems while giving enterprises greater control over how those agents behave.
The announcement signals an important evolution in the enterprise AI race.
Instead of trying to convince businesses that one AI model will dominate every task, Salesforce is positioning itself as the layer that can sit around multiple models and agents — supplying them with proprietary business context and governing how they operate.
Salesforce thinks the AI model itself could become less important
The logic behind Salesforce’s strategy is straightforward.
Artificial intelligence models are improving rapidly, while companies increasingly have access to models from multiple providers. What competitors cannot easily replicate, however, is an enterprise’s internal information: customer histories, contracts, transactions, workflows, permissions, institutional knowledge and business rules.
Salesforce President and Chief Platform and Engineering Officer Rohan Kumar argues that this proprietary context could become one of the most durable competitive assets in the AI era.
In other words, two companies may eventually have access to similarly capable AI models — but the company whose AI has better access to accurate customer information, clearer policies and better-integrated workflows could produce dramatically better business results.
That distinction becomes even more important as AI evolves from simply answering questions to actually taking actions on behalf of employees and customers.
The problem Salesforce is trying to solve
Consider a deceptively simple question: Can this customer’s order be fulfilled today?
The answer could depend on information spread across multiple systems.
A CRM may contain the customer relationship. An ERP platform may hold inventory data. Contracts may define delivery commitments. Analytics systems may contain revenue or account-health definitions. Previous customer interactions may reveal important context, while internal policies determine what an employee — or an AI agent — is authorized to promise.
Connecting an AI model to all those systems is only part of the problem.
The harder challenge is ensuring that the AI understands what the information means, chooses the right action and stays within corporate security and governance rules when it acts.
That is the gap Salesforce says its Enterprise AI Harness is designed to address.
Six layers sit behind Salesforce’s AI strategy
The architecture brings together six broad capabilities: Trusted Context, Trusted Agency, Trusted Action, Trusted Governance, Trusted Security and Trusted Models.
Trusted Context supplies AI with customer data, metadata, knowledge, semantics and organizational memory.
Trusted Agency covers the reasoning, planning and orchestration needed for AI agents to complete tasks.
Trusted Action connects those agents with applications, APIs, workflows and other business tools.
Governance and security layers are intended to enforce data quality, permissions, privacy rules, policies and guardrails.
Trusted Models, meanwhile, is designed to let companies use different AI models depending on factors such as accuracy, performance, cost and specific business requirements.
The architecture draws on technologies already spread across Salesforce’s portfolio, including Data 360, Informatica, MuleSoft, Tableau, Agentforce, Salesforce Guardian and the Salesforce Platform.
The significance is that Salesforce is not presenting the Harness as a single new AI model. It is attempting to turn its existing data, integration and application ecosystem into an operating layer around AI.
Then comes the AI Control Plane
Perhaps the more strategically important piece is Salesforce’s planned AI Control Plane.
As companies deploy more autonomous agents, executives will increasingly need answers to basic but difficult questions: Which AI agents are operating inside the company? What data can they access? Who authorized them? What actions can they perform? How well are they performing? And how much are they costing?
Salesforce says the Control Plane will allow organizations to discover and register AI agents, assign identity and policies, manage their lifecycle, evaluate performance, observe behavior and control spending.
Crucially, Salesforce says those controls are intended to extend beyond its own AI ecosystem to third-party agents and models as well.
That could become increasingly important because large businesses are unlikely to rely on only one AI provider.
A July 2026 VentureBeat Intelligence survey involving 107 respondents from organizations with at least 100 employees found that 85% were already operating two or more agent-orchestration platforms, with an average of 3.1 platforms per enterprise. Some 53% expected their primary agent-control architecture to be hybrid by the end of 2026.
VentureBeat cautioned that the self-selected sample was weighted toward large technology organizations, so the results should be treated as directional rather than as a broad market-share measurement. Still, they illustrate the challenge Salesforce is targeting: businesses may soon have dozens or even hundreds of agents operating across several platforms.
Salesforce also wants to avoid locking companies into one AI ecosystem
Another notable element of the strategy is Salesforce’s emphasis on openness.
The company says the Enterprise AI Harness will expose capabilities through technologies including Model Context Protocol, APIs, Skills and plug-ins, allowing Salesforce data and business functions to appear inside other AI experiences.
Salesforce specifically cited environments including Claude, Slack, Microsoft Teams and Agentforce.
That matters because enterprises increasingly want the freedom to switch models or combine AI providers rather than rebuild their technology stack every time a stronger or cheaper model appears.
The approach effectively turns Salesforce’s massive installed base of customer data, workflows and business applications into infrastructure that outside AI models may potentially use — while Salesforce retains a role in governing the context and actions surrounding them.
But there is one major catch: much of the unified system is still coming
Businesses should not mistake the announcement for a fully finished product available today.
Many technologies underpinning Salesforce’s Enterprise AI Harness are already available across the company’s portfolio, but the company says new capabilities and the unified experience are expected to begin rolling out in early fiscal 2028.
Salesforce’s fiscal 2028 begins in early 2027.
Detailed pricing, packaging, availability and upgrade paths have also not yet been disclosed.
That leaves Salesforce with an execution challenge.
The enterprise AI market is developing extraordinarily quickly, and competitors are also racing to become the control layer through which companies manage fleets of AI agents.
SiliconANGLE noted that the Harness is intended to make agents more reliable and secure while the Control Plane focuses on areas such as monitoring, cybersecurity policies and inference costs.
Why this matters for businesses — including those in the Philippines
For enterprises in the Philippines and across Southeast Asia, Salesforce’s strategy highlights a broader shift in the AI conversation.
The first phase of generative AI focused heavily on which model could answer questions most intelligently.
The next phase is increasingly about whether AI can safely do actual work.
Banks, retailers, telecommunications companies, BPO operators and other large organizations frequently have customer and operational information spread across CRM platforms, databases, analytics systems, legacy applications and internal documents.
An AI agent becomes significantly more useful when it can understand that information together — but also significantly more dangerous if it can act without the proper controls.
That makes identity, permissions, governance, auditability and context just as important as raw AI intelligence.
The bigger battle may be outside the model
Salesforce’s latest move reflects a potentially profound change in the AI market.
Foundation models will continue competing on reasoning, speed and cost. But as their capabilities converge, enterprise buyers may increasingly care about everything surrounding the model: the data it understands, the applications it can access, the rules governing its behavior and whether management can see exactly what it is doing.
That is where Salesforce believes its decades of customer data, workflow and enterprise-software infrastructure can become an AI advantage.
The company is effectively betting that the winner of enterprise AI may not be whoever builds the most powerful brain.
It may be whoever controls the context that tells that brain what the business actually means — and what it is allowed to do.
Leave a Reply