Who will control AI agents in the enterprise: the fight over the "control plane"

By Nate (Nate's Substack) · November 1, 2026.

The email is the opening of an executive briefing by Nate on the competition among major software vendors to become the “AI control plane” in enterprises. The body received is only a preview: the full content is reserved for Executive Circle subscribers.

The author begins with three developments from the same week. On October 27, Salesforce launched software that lets it send work to agents built with Microsoft Copilot Studio. That same day, Oracle announced it is developing a workspace to bring its business applications into ChatGPT. The next day, ServiceNow announced a Control Tower Alliance whose founding members include OpenAI, Anthropic, Amazon Web Services and NVIDIA.

Nate's thesis is that these companies are helping each other reach their customers while at the same time competing to be the vendor those customers depend on to get work done across everything else they own. In his view, alliances are one of the ways to compete for that role.

Nate recalls that for years a company could buy Microsoft for employees' day-to-day work, Salesforce for customer relationships and Oracle for finance. Those companies already competed in some areas, but they could maintain solid relationships with the same customer by serving different parts of the business. Agents, he argues, give them a reason to fight over more of the work that happens between applications.

The author refers to his video on the “trillion-dollar fight to run your business,” which focuses on why Salesforce and Microsoft now compete over responsibilities that used to be more separate. In this briefing he widens the lens: Google, Amazon, the AI labs, ServiceNow, business application companies and data platforms would all have credible reasons to take part, with different advantages, and their alliances would reveal what each still needs.

The piece is aimed at two audiences: those who buy software, to evaluate what they are being offered, and those trying to do their work with it, to explain what their company needs. Nate argues that the decisions made under the “AI control plane” label will largely determine how useful agents will be allowed to become.

The full briefing covers five sections. First, what a control plane actually controls: six responsibilities to separate, from receiving the request to reviewing performance, to see which decisions one product can make over others. Second, where each vendor is starting from: what Microsoft, Salesforce, Google, Amazon, OpenAI, Anthropic, ServiceNow and business application companies bring, and what their alliances say about what they still need. Third, how to read an alliance announcement: identifying what passes between the companies and in which direction the connection flows. Fourth, eight questions for a vendor meeting, with tests on authority, handoffs, mid-task failures, portability, cost and availability, applied to a real task. Fifth, a briefing to send to all vendors: a single task description that forces every demo to answer the same questions.

The text received does not include the six responsibilities, the eight questions or the conclusions; that is why this summary does not detail them. As industry context, the battle to control orchestration, permissions and oversight of agents across enterprise applications is one of the most visible lines of competition in enterprise AI.