Executive Edition 2026
The AI Control Plane Handbook
Architecture, orchestration, and zero-trust security for the agentic enterprise
About this handbook
This handbook sets out what an enterprise AI control plane is, why it has become the next infrastructure advantage, how to architect one, the vendor-neutral target reference architecture to steer by, one way to realize it with MuleSoft Agent Fabric and Informatica, and how to sequence the work over the next 12 to 24 months so that every milestone brings a strategic win.
Sections 11 to 13 apply the architecture to three reference use cases: a professional services firm, a capital markets firm and a natural gas pipeline operator.
It is written for CTOs, CIOs and the architects who work with them. Each chapter combines the architectural framework with the research behind it, and every source is listed in the endnotes.
In brief
-
The new infrastructure advantage
Advantage will come from the governed system that decides which model runs each task and under which constraints, rather than from any single model.
-
Three capabilities
Architecture exposes what the enterprise can do as secure, governed services. Orchestration coordinates models, agents, workloads and the decisions they make, and trust, security and resilience controls verify every actor and output continuously.
-
Paradigm before platform
Embedded, dedicated and hybrid/federated patterns each carry trade-offs, and most large multi-vendor enterprises end up with central policy and distributed execution.
-
MuleSoft and Agent Fabric
MuleSoft’s API-led connectivity exposes enterprise applications as governed APIs and MCP tools, and MuleSoft Agent Fabric discovers, governs, orchestrates and meters agents from any vendor.
-
Value in every phase
Over the next 12 to 24 months, each phase puts a slice of a real use case into production. The work covers an inventory of agents, one gateway for all model and tool traffic, autonomy tiers and agent identities, with post-quantum migration started ahead of the 2030–2031 federal deadlines.
What’s inside
Every section and appendix is a page of its own. Start with the executive summary, or go straight to the part you need.
Start here
Companion materials
Three companions build on the handbook. The executive deck and the assessment toolkit share its endnotes and its reference use cases.
-
Executive deck
The briefing for CTO and CIO audiences, slide by slide, with the slide text and the speaker notes.
-
Assessment toolkit
The method, thirteen templates and a worked example for architects and project teams, with a client deck and a client report template.
-
Quantum briefing
The federal post-quantum deadlines, what they mean for the agent control plane, and what boards should ask for.
About John Fisher
I’m John Fisher. At Salesforce, I help enterprises architect the data foundation that AI depends on: MuleSoft for integration and APIs, Informatica for data management, and the MCP and A2A protocols that connect agents to tools, data and each other. Before Salesforce, I spent fourteen years at TIBCO leading solution engineering for enterprise accounts. Across both companies, my work has sat where enterprise architecture, integration and data meet business strategy.
My writing is where I work through what happens when AI agents move past experiments and start acting across real enterprise systems. The questions that interest me most are architectural: how an agent proves who it is, what policy decides what it may reach, how its cost is measured and controlled, what evidence it leaves behind, how agents built on different platforms work together, and what it takes to run all of that at scale.
Some of what I write names Salesforce, MuleSoft or Informatica technology. It’s the technology I know most deeply, so it’s often the quickest way to make an idea concrete. I use it as one example of a broader pattern, not as the only answer, and the target architectures I start from are vendor-neutral by design.
Disclosure. John Fisher works for Salesforce. He wrote this handbook independently, on his own time and not as part of his work at Salesforce, using only public information. The views in it are his own, not those of Salesforce or any other company named, and it is not sponsored or endorsed by Salesforce or any of them.
License
© 2026 John Fisher. This handbook is licensed under the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License (CC BY-NC-ND 4.0). You may copy and share it, in whole or in part, without changes and for non-commercial purposes, if you credit John Fisher and link to the license.
Suggested credit: John Fisher, The AI Control Plane Handbook (2026), CC BY-NC-ND 4.0.
The license covers John Fisher’s own text and figures. It does not cover trademarks or the quotations, data and other material of the sources cited, which belong to their owners. The handbook is general information for architecture and technology planning, not legal, regulatory, compliance, financial or investment advice.