--- title: "What Is the Agent Control Plane and Who Will Own It?" url: "https://goremy.ai/blog/articles/salesforce-microsoft-oracle-openai-agent-control-plane" author: Luis Chavez-Mattos editor: Luis Chavez-Mattos editor_title: Director, Product published: "2026-11-02T18:20:07.778Z" updated: "2026-11-02T18:20:07.778Z" topic: AI Industry & Strategy description: Salesforce, Microsoft, Oracle and OpenAI are opening products to each other. Here is what the agent control plane is and why enterprise vendors want it. --- # What Is the Agent Control Plane and Who Will Own It? > Salesforce, Microsoft, Oracle and OpenAI are opening products to each other. Here is what the agent control plane is and why enterprise vendors want it. Edited by Luis Chavez-Mattos, Director, Product ยท Published 2026-11-02 ## What is the agent control plane? The agent control plane is the software that oversees AI agents inside a company, and sometimes coordinates their work. It decides which information an agent can use, which tools it can reach, which company rules apply, who approves actions, and where results go. Salesforce, Microsoft, Oracle and OpenAI are all competing for that position, even while they cooperate on specific integrations. ## TL;DR - Salesforce made Microsoft's agents easier to use, and Oracle announced a workspace to bring its apps into ChatGPT. **Rivals are opening doors to rivals** because each wants to be the layer customers work through. - The prize is not owning every agent. It is owning **the setup that mediates agents** across a company's systems, permissions and data. - Salesforce's bridge works with Copilot Studio agents, so customers **do not have to rebuild agents** to bring them into Salesforce-coordinated work. - Oracle's ChatGPT workspace is still in development. Its pitch is a finance employee investigating costs and month-end close, **subject to company permissions and approvals**. - Better models do not end the contest, because enterprises care about **context, rules and accountability** as much as raw model quality. - No one has the equivalent of Windows for agent control yet. The customer relationship is being negotiated **product by product**. ## Why are Salesforce and Microsoft helping each other? Salesforce made Microsoft's agents easier to use through a bridge that works with Copilot Studio agents. A customer who built an agent in Copilot Studio does not need to rebuild it to take part in Salesforce workflows. That looks like Salesforce propping up a rival. It makes more sense as a bid for responsibility. If a Microsoft-built agent does part of a job and Salesforce coordinates the overall process, Salesforce stays in the loop. Microsoft's agent gets used more. Both companies show customers they offer choice. The old arrangement helps explain why this is new. Microsoft sold Outlook, Excel and Teams. Salesforce sold the system where sales teams manage customers. Oracle often sold finance applications. They overlapped at the edges (Microsoft has its own business applications, and Salesforce owns Slack), but a customer could buy from all three and each held a different part of the relationship. Agents change that because real work does not respect product boundaries. ## Why does work across tools force vendors into competition? Closing the month crosses spreadsheets, email and accounting systems. Handling a sale does not stop at the customer record. People have always bridged those gaps by reading the email, checking the numbers and deciding what to do next. Much of what made software useful happened in the person moving between applications. When an agent takes over that movement, someone has to supply the layer that answers a few questions: 1. Which information can the agent use? 2. Which instructions are current? 3. Who can approve actions? 4. Where does the result go? Each vendor answers from the position it already holds. | Player | Its claim to the agent layer | |---|---| | Microsoft | Already has your documents, your Teams conversations and your employees' daily work | | Salesforce | Understands your customers and the sales processes that bring in revenue | | Oracle | Runs finance, procurement and planning applications (the Oracle ChatGPT workspace targets these) | | AI labs such as OpenAI | People already ask their products to do the work, so connect the apps they need | These are different angles on the same seat. ## What is Oracle building with OpenAI? Oracle is developing a workspace that brings its business applications and data into ChatGPT. It is still in development, but the direction is clear. Customers keep their Oracle applications while doing more of the work through OpenAI's product. The announced scenario is a finance employee using ChatGPT to investigate costs, understand purchasing decisions and identify what is left before month-end close. The workspace would draw on finance, procurement and planning applications, subject to company permissions and approvals. Those restrictions are the kind of thing a control plane exists to enforce. Both sides gain. Oracle customers get more use from applications they already pay for. OpenAI gets a chance to be where the employee works through the problem. If that goes well, the next request might start in ChatGPT too, and another data source or department can be added. A narrow finance use case can grow into a larger role. ## Why don't better models settle this? An individual developer can pick the best model and move on. An enterprise can't, because a more capable agent still has to operate inside company rules, with the right data and the right approvals. The Salesforce and Microsoft case shows how this plays out. A stronger Microsoft agent can make a Salesforce-coordinated process more useful, and Salesforce benefits without supplying the model. Microsoft still has every reason to sell the customer more of the surrounding system. Both can gain from the arrangement and still fight hard over the next purchase. This also means the control plane can be agnostic about which agent does the work. A company might run Claude, ChatGPT or Qwen agents. What it needs is a common layer to manage them as a corporate asset rather than as a pile of personal assistants. ## How does context fit into the control plane? Access to data is not the same as understanding it. A spreadsheet can hold the correct revenue number while an agent uses the wrong definition of revenue. A customer record can be accurate and still miss a promise someone made in an email yesterday. Consider a customer who was promised delivery yesterday and is now talking to a different employee, or a different agent. Does the memory of that promise survive? Does the new handler know what was agreed, and can they act on it? Software that makes this easy gives customers a reason to keep paying for it. That is the thinking behind products like Salesforce's agent context engine, currently in preview. It aims to supply shared business context to agents on many supported platforms, including agents built elsewhere. Microsoft has related products that gather business data, definitions and rules in one place, though they are not the same thing and are hard to compare directly. All of these efforts try to make existing business information more legible to agents. The business still has to supply the judgment about what the information means for a given decision. ## Is there a clear winner yet? No. The comparison offered is Windows in the 1990s, which shaped our experience in the workplace, just as the control plane is expected to. There is no equivalent for agents yet. Salesforce talked up its control plane approach at Dreamforce, Microsoft has been moving toward one, and Oracle is now starting to talk about it. OpenAI has been pursuing enterprise connections in other ways too, including a product called Frontier launched in February. The money behind the shift is large. The nine biggest cloud companies are on track to spend $957 billion on capacity this year and more than $1.5 trillion next year, according to the figures cited in the source analysis. Whoever mediates agent work gets to monetize that infrastructure through the customer relationship. ## What should teams buying agents watch for? You are effectively choosing a mediator each time you adopt an agent. A few questions help: - Can agents built on another vendor's platform participate without a rebuild? - Where do permissions and approvals live, and can you inspect them? - Does shared context carry across employees, agents and tools, including commitments made in email or conversation? - Who defines business terms like revenue, and can agents see those definitions? - How hard would it be to move to a different mediator later? Vendor openness today does not guarantee openness later. The cooperation is real, but each company is also positioning to be the one customers rely on when the next purchase comes up. ## Frequently Asked Questions ### Why would Salesforce help Microsoft's agents work better? Because the bridge lets Salesforce coordinate work even when Microsoft's Copilot Studio agents do part of it. Customers keep the agents they built, and Salesforce earns a bigger role in how the work is managed. It is a way to gain responsibility without supplying the model. ### Is the Oracle and OpenAI workspace available now? According to the announcement, the workspace is still in development. The described goal is to bring Oracle's finance, procurement and planning applications and data into ChatGPT, subject to company permissions and approvals. ### Does the control plane replace systems of record? No. The strategy described is to meet customers where they are, with the systems of record they already have, and add control over agent experiences on top. The control plane mediates access and rules rather than replacing the underlying applications. ### Does the control plane lock me into one model provider? Not necessarily. The valuable position is described as agnostic about which agent does the work. The lock-in risk sits in the mediating layer itself, meaning where your context, permissions and approval flows end up living.