KM4BI

Salesforce Is Helping Microsoft: Who Will Control Enterprise AI Agents?

The enterprise AI contest is shifting toward control planes, business context, and interoperability.

Published

Recent announcements point to a new kind of competition. Salesforce is making its systems easier for Microsoft Copilot agents to use, while Oracle and OpenAI are working on a workspace that brings Oracle applications into ChatGPT. These partnerships do not erase rivalry; they show how vendors can help one another serve customers while competing for a more strategic position.

That position is the agent control plane: the layer that supplies information and tools, applies company rules, governs permissions, and routes work for review. When an agent’s task spans CRM, email, spreadsheets, and finance, value comes not only from each application but also from coordinating the work and preserving the context that makes it meaningful.

Each vendor starts from a different strength. Microsoft has workplace documents and conversations; Salesforce understands customers and sales processes; AI labs benefit from users already bringing them tasks. None of those advantages alone settles the contest. Owning data does not guarantee that the owner will also become the place where work is orchestrated.

Accumulated context matters too: instructions, corrections, team practices, and connections between tools. A system that preserves and reuses that knowledge can become genuinely useful—and costly to replace. But that knowledge must be maintained; raw data access will not carry it forward automatically.

When evaluating platforms, companies should test interoperability, agent replaceability, fit with existing systems, and permission controls. Building an internal control plane may be viable, but only with a plan to staff and maintain it. Keeping options open is the best defense against lock-in.

**Watch next:** the Salesforce–Microsoft and Oracle–OpenAI partnerships, shared-context features, and whether customers can run multiple agents without rebuilding their workflows.

Source