Agentic Control Plane: Why Microsoft, Salesforce and OpenAI Want to Run Your AI Agents

Two announcements on the same day look odd at first. Salesforce made Microsoft’s agents easier to use, and Oracle said it is bringing its business apps into ChatGPT. Companies that compete don’t usually make life easier for each other.

This article explains Nate B Jones’s view of the agentic control plane: what it is, why the biggest software companies want to own it, and how you can choose one without boxing yourself in, whether you run a large team or a small business.

What did Salesforce, Microsoft, Oracle and OpenAI actually announce?

Two partnerships, both announced on 27 October 2026, according to the video.

• Salesforce and Microsoft: a MuleSoft bridge, built on an agent-to-agent (A2A) approach, lets Microsoft Copilot Studio agents take part in work that Salesforce coordinates, without the customer rebuilding those agents.
• Oracle and OpenAI: Oracle is developing an Oracle AI Workspace that brings its finance, procurement and planning apps and data into ChatGPT. Nate stresses that it is announced but not yet available.

Nate also points to related moves: Microsoft Agent 365, which Microsoft describes as a control plane for AI agents; Salesforce’s control plane message at Dreamforce; and OpenAI’s Frontier platform from February, which joins business systems, context, agent work and permissions in one place.

What is an agentic control plane?

It is software for overseeing AI agents, and sometimes for coordinating their work. Nate puts it in practical terms. An employee gives an agent a job. The control plane makes sure the agent gets the right information and tools, applies the company’s rules and brings the work back for the employee to review.

To do that, the agent needs answers to four questions:

1. Which information can it use?
2. Which instructions are current?
3. Who approves its actions?
4. Where does the result go?

Nate compares the position to Windows in the 1990s, when almost every office worked through the same system. His point is that there is no “Windows for agents” yet, and every big vendor would like to be it. Crucially, the control plane can stay neutral about which model does the work. You could run Claude, ChatGPT or Qwen agents through the same layer.

Why would Salesforce help a rival like Microsoft?

To earn a bigger role with the customer, not out of generosity. Nate’s reading is that customers want choice. By making Copilot Studio agents work inside Salesforce-coordinated processes, Salesforce stays involved in the work even when Microsoft supplies part of it.

Both companies can gain. A more capable Microsoft agent makes a Salesforce-coordinated process more useful, without Salesforce having to supply the model. Then they compete hard for the next purchase. Nate describes the customer relationship as being “negotiated product by product”.

The same logic applies to Oracle and OpenAI. Oracle customers get more use from apps they already pay for. OpenAI gets the chance to become the place where a finance employee works through month-end close. If that goes well, the next request may start in ChatGPT too, then another data source, then another department.

Why is business context the real prize?

Because connected data is not the same as understood data. Nate gives two simple examples. A spreadsheet can hold the correct revenue figure, but an agent might use the wrong meaning of “revenue”. A customer record can be accurate and still miss a promise made in an email yesterday.

That is why vendors are building context products. Salesforce’s Agent Context Engine, in preview, aims to give shared business context to agents on many platforms, including agents built elsewhere. Microsoft’s IQ products bring together work context, business data and the definitions agents need. Microsoft Fabric can reference supported external data where it already lives, so a business doesn’t need a big migration before a new supplier becomes important.

Nate’s test is a customer promise. A customer was promised a delivery yesterday. Today they speak to a different employee, or a different agent. Does the promise survive the handover? Software that makes that easy earns its place. Our guide on getting your business data ready for AI tools covers the groundwork.

What really moves when you switch AI agent providers?

Much less than you think. Nate shares a lightly anonymised story about a recruiter. Her agent pulled candidate profiles and drafted a shortlist. She reviewed it and passed it to the hiring manager. Over time she taught it how she screens people, what each manager wants and who actually got hired.

When her company considered changing AI provider, her manager hadn’t realised how much work sat in that arrangement. The raw candidate data could move. Her instructions, corrections, skills and the way her tools worked together would not.

Nate also notes that this knowledge needs upkeep. Last month’s correction doesn’t automatically reach this month’s brief. A system that makes that maintenance easy has a lasting job. Our guide on how to keep your AI agent setup when you switch tools shows how to protect that work.

How big is the fight, and is it winner takes all?

Very big, but Nate doesn’t think it has to be winner takes all. He cites TrendForce figures showing the nine leading cloud providers on track to spend about $957 billion on capacity in 2026 and more than $1.5 trillion in 2027. He also cites Gartner’s forecast of about $1.46 trillion in worldwide software spending in 2026, while noting that not all of it is up for grabs.

His view is that different teams have different needs. Finance may need controls that marketing doesn’t. Specialists may want to fit the pieces together themselves. He calls the current pattern “frenemy competition” and hopes it continues, because partnerships give customers choice. He suggests watching the next 12 to 24 months to see whether a single winner emerges.

What does the video not confirm?

The video also doesn’t compare the products in practice. There is no hands-on test of the MuleSoft bridge, Agent 365 or Frontier, so treat this as a strategy explainer rather than a review.

How should you choose an agentic control plane?

Start with your own work, then judge each option on how well it connects. Nate’s checklist, in our words:

1. Ask where the value comes from. Is it the model, the data, the workflow or the tools your people already use?
2. Check interoperability. Can it work with agents and tools from other vendors?
3. Test how easily you can swap parts. Could you replace one agent or model without rebuilding everything?
4. Look at the fit with your setup: your agent harness (the software that runs your agents), your memory and data systems, and the apps your staff live in every day.
5. If you plan to build your own, be honest about resources. Nate warns it is not a small build, and it needs people to maintain it.
6. If you are not the buyer, advocate. Tell your procurement team what you need and why. Nate says the people who ask clearly tend to get the budget.

Our guide on choosing an AI agent platform without getting locked in turns this into questions you can put to any vendor, with a simple scoring table.

Who does this matter to most?

This matters most to you if:

• you influence which AI tools your team or company buys,
• you already rely on Microsoft 365, Salesforce or Oracle and are adding AI agents,
• you build software or services that will need to work with agents.

It matters less if:

• you use one AI chat tool for occasional writing and research,
• you are a sole trader with no plans to connect agents to business systems yet.

Even then, Nate’s third takeaway applies: agents will change how people use anything with a screen. If you sell a product or service online, it is worth asking how it will work when customers arrive through an agent.

Frequently asked questions

The bottom line

The biggest software companies are competing to run the layer your AI agents work through, and that is why rivals are suddenly cooperating. For you, that competition is useful: it means more choice, as long as you ask for it. Choose your agent platform deliberately, make interoperability a must-have and keep your agent setup somewhere you own. Your next step: list the AI tools your team uses today and note which ones could work together if you changed one of them.

Related posts

• Anthropic S-1: What It Reveals About Enterprise AI Adoption
• How to Choose an AI Tool for Your Business: Benchmarks vs Real-World Results
• How to Get Your Business Data Ready for AI Tools

Garry Knight

I'm Garry Knight, the person behind Prodify Digital. I write about email list building, email marketing, SEO, AI search and the tools that connect them. My aim is to make online marketing easier to understand, so creators and small business owners can make informed decisions about building an audience and keeping people engaged. Here you'll find straightforward guides and product reviews that explain what something does, where it fits and which limitations matter. The focus is on clear explanations and useful next steps—not hype, shortcuts or promises of easy earnings.

See Full Bio