About This Video
A 29-minute video from AI News & Strategy Daily | Nate B Jones, published November 1, 2026. He uses a Salesforce and Microsoft partnership announced on October 27th, and a same-day Oracle and OpenAI announcement, to explain a shift in how big software companies compete. The analysis and predictions are his own opinion.
Summary
Jones argues that AI agents are changing what counts as competition. Companies that used to sit in separate lanes (Microsoft, Salesforce, Oracle, and the AI labs) now overlap because each wants to own the agent relationship, meaning the layer that mediates how people do computing at work. He calls that layer the agentic control plane and says it is the "high ground." Yet these same companies are helping each other reach customers, because customers want choice. His advice is to be deliberate about where your control plane sits and to demand interoperability when you buy.
Key Arguments
- Rivals are opening doors for rivals (00:00). On October 27th Salesforce made Microsoft's agents easier to use, and Oracle said it is building a way to bring its apps into ChatGPT. Jones says nobody does that out of generosity: each is chasing the same seat, the software you depend on to get work done across everything else you own.
- The money behind it (00:00). He says the nine biggest cloud companies are on track to spend $957 billion on capacity this year and more than $1.5 trillion next year. Later (18:00) he cites a Gartner forecast of about $1.46 trillion in worldwide software spending this year, and notes that not all of it is up for replacement or yet agent spending.
- Software used to be complementary (04:03). Microsoft sold Outlook, Excel, and Teams, Salesforce sold the customer system, and Oracle might have sold finance applications. They already overlapped a little, but a customer could buy from all of them.
- People used to be the glue (05:01). Closing the month crosses spreadsheets, email, accounting systems, and the finance department. Much of what made software useful happened in the person moving between applications. When agents do that work, someone must supply the layer that decides what information they can use, which instructions are current, who approves actions, and where results go.
- Each vendor has a different angle on the same position (05:01 to 06:00). Microsoft can say it has your documents, Teams conversations, and employees' working day. Salesforce can say it understands your customers and your sales process. The AI labs can say people already ask their products to do the work.
- The 2026 and 2027 question is managing agents (07:04). Opening data to agents is only the start. The question is how to manage and control agents so they become an enterprise-scale approach to intelligence. He says that is why Salesforce, Microsoft, and Oracle all talk about a control plane.
- Why Salesforce helps Microsoft (08:03 to 09:02). The bridge works with Copilot Studio agents, so a customer does not have to rebuild an agent to use it with Salesforce. Salesforce can coordinate work even when Microsoft supplied part of it. Jones reads this as a way to earn more responsibility from the customer by giving them choice.
- Better models do not end the competition (09:02). A more capable Microsoft agent can make a Salesforce-coordinated process more useful, while Microsoft still has reason to offer more of the surrounding system. The customer relationship is negotiated product by product.
- Definitions and judgment still matter (12:01). A spreadsheet can hold the correct revenue number while an agent uses the wrong meaning of revenue. A customer record can miss a promise made in an email yesterday. Connecting systems gives access, but the business still has to supply what the information means.
- Oracle and OpenAI (13:02 to 14:01). Their proposed workspace would bring finance, procurement, and planning applications into ChatGPT, within company permissions and approvals, to help with tasks like investigating costs before month-end close. Oracle customers get more from apps they already pay for, and OpenAI gets a chance to become where the employee works through the problem, then expand to another department.
- Owning the data is not enough (15:01). Jones says Microsoft's Fabric (as transcribed) can reference external data in place, so a customer does not need a giant migration before another supplier becomes important. That lowers a barrier to competition spreading across enterprise software.
- Accumulated work is the sticky part (16:00). He describes a recruiter (story lightly anonymized) whose agent pulled candidate profiles and drafted shortlists. Over time she taught it how she screens people, what hiring managers want, and who got hired. The raw candidate data was only part of it: her instructions, corrections, skills, and tool setup would not move with the data. A vendor that preserves and spreads that knowledge becomes hard to replace.
- Do not expect winner takes all (19:01 to 22:00). He hopes for "frenemy" competition with partnerships that give customers choice. He thinks the market is big enough for multiple winners, since finance may need controls marketing does not, but says it will be clearer within roughly 12 to 24 months whether a true control plane winner emerges.
What He Tells Viewers To Do
- Be intentional about the control plane (23:01). Put interoperability into purchasing requirements when you build or buy one, so you are not boxed in. He says most people he talks to want strategic optionality and several solutions in play.
- Vet it with questions (24:01). Where does the value in the system come from? How interoperable are the agents? How many options do you have to swap them out? How does it tie into your agent harness, memory and data systems, and the tools employees use every day?
- Advocate even without budget authority (25:00). Think about data access, interoperability with other agents, and workflows, and tell procurement what you need and why. He says the squeaky wheels tend to get the budget.
- Do not default to what you use every day (26:00). Think about how a choice drives the business long term.
- Decide where the control plane sits (26:00). If you plan to build one, he warns it is not a small build and needs talent to create and maintain it.
- Consider your own product (27:02). If you are in software or tech, ask what you put on the table in an agent-mediated world. He says this reaches B2C and physical products too, and promises another video or two on it.
Key Terms
- Agentic control plane
- Software for overseeing agents, sometimes including coordinating their work. It determines when, where, and how agent work happens and what bounds apply (10:05).
- High ground
- Jones's metaphor for owning the relationship that mediates the rest of your computing at work (02:01).
- Frenemy competition
- Companies that partner where it makes sense while still competing, which he hopes continues (20:03).
- Agent context engine
- Salesforce's product, in preview, meant to supply shared business context to agents on many platforms, including agents built elsewhere (12:01).
- Microsoft IQ products
- Related to the context engine but not quite the same, which he says makes them hard to compare (12:01).
- Frontier
- OpenAI's enterprise offering, launched in February, which connects business systems, context, agent work, and permissions in one place (14:01).
Key Takeaways
- Agents turn once-complementary vendors into rivals for the same position.
- The prize is the control plane that manages agents across a company.
- Partnerships between rivals mostly give customers choice, and that is good for buyers.
- The knowledge built up while people work with agents is what makes a vendor hard to leave.
- Demand interoperability and optionality when you buy.
- Speak up with procurement even if you do not hold the budget.
Review Questions
- What does Jones mean by "the high ground," and why is it valuable?
- Why does he say Salesforce helping Microsoft is not simply supporting a rival?
- How does he define an agentic control plane, and what comparison does he draw to the 1990s?
- Why does connecting systems not solve the problem of what data means? Use his revenue example.
- In the recruiter story, what is lost if you only keep access to the raw data?
- Why does he say owning the data does not settle where the work happens?
- Does he expect a single winner? What time frame does he give?
- Which questions does he suggest asking when vetting a control plane?