# Salesforce Just Helped Microsoft. You're Picking The Winner Of A Trillion-Dollar Fight.

## Executive summary

The enterprise AI landscape is entering a 'trillion-dollar fight' over who controls the agentic control plane—the framework that mediates and coordinates AI agents across a company's existing software stack. Instead of a single winner, the trend is toward 'frenemy competition' (e.g., Salesforce helping Microsoft's agents) that emphasizes interoperability and choice. For build engineers, the key takeaway is that architectural focus must shift from merely connecting data sources to establishing a robust, vendor-agnostic control plane that manages agent workflows, corporate rules, and user permissions.

## Key takeaways

- The High Ground is the Control Plane: The competition is no longer about which AI model is best, but which vendor owns the 'agent relationship'—the control plane that manages and coordinates agents across the entire business process. This control plane is the most valuable asset, as it dictates how agents operate within the corporate environment.
- Interoperability Drives Value: Major rivals (Microsoft, Salesforce, Oracle) are opening doors for each other to ensure customers have choice and flexibility. This 'frenemy competition' is beneficial for customers, allowing them to build systems that integrate multiple vendors and avoid vendor lock-in.
- Be Intentional with Interoperability: Customers must be intentional about requiring interoperability when purchasing or building agentic systems. The ability to swap out agents or components without rebuilding the entire workflow is critical for long-term architectural resilience.

## Technical details

- Agentic Control Plane: This is the software layer responsible for overseeing, coordinating, and bounding the work of AI agents. It determines when and where agent work can happen, what the rules are, and how the results are reviewed, much like how Windows shaped the workplace experience in the 1990s. It is the mechanism that makes agentic experiences enterprise-scalable.
- Vendor Ecosystem Integration: The trend involves vendors building workspaces that bring their proprietary applications and data into external AI platforms. Examples include Oracle developing a workspace to integrate its apps into ChatGPT, and Salesforce's agent context engine (in preview) aiming to supply shared business context to agents across multiple platforms.
- Architectural Shift: The focus is shifting from siloed B2B applications (e.g., Outlook, Salesforce CRM, Oracle Finance) to integrated workflows. The challenge is supplying the agent layer with the necessary context, instructions, and approval mechanisms, even when the data and people involved are spread across multiple products.

## Practical implications

- When evaluating enterprise AI solutions, prioritize interoperability and the ability to connect multiple, disparate systems over deep integration with a single vendor's stack.
- Architectural planning must account for the 'agentic control plane' as a core, non-negotiable component, ensuring it can manage permissions, rules, and workflow boundaries.
- Advocate for solutions that allow for choice and multiple agents to participate, rather than accepting a single-vendor, 'winner-takes-all' architecture.
- Focus on how the system preserves and utilizes accumulated 'useful knowledge' (instructions, corrections, process improvements) rather than just accessing raw data.

## Topics

AI Agents, Enterprise Architecture, Control Plane, B2B Software, Interoperability, Cloud Computing, Nate's Newsletter, Nate's Library MCP

Source: https://www.youtube.com/watch?v=KDGxQtfjBj0