The choice of Agentic AI technology for enterprise leaders is best understood as an ecosystem of tech vendors that align with three functional planes: the need to build agents, to embed them into business workflows, and to manage and govern them at scale. These needs represent functional planes rather than strict layers in a ‘stack’, since vendors often span these or blur the boundaries between them. Even so, enterprises must treat them as distinct problem-spaces in order to design coherent architectures and make sound investment decisions.
The first plane is the ‘build’ plane, which Forrester’s research on AI platforms identifies as a key – though not exclusive – area of focus. This plane addresses the question of how to build, deploy and scale agentic AI systems and applications. In this build plane, AI platforms provide model access, agent frameworks, tool integration, vector stores, evaluation pipelines and the surrounding infrastructure required to create agents or agent-enhanced applications.
The second plane is ‘orchestration’, and it is expressed by what Forrester terms as Adaptive Process Orchestration (or APO). This plane addresses the question of how to orchestrate agentic and non-agentic components inside business processes and workflows. APO platforms let teams model processes, define routing and decision logic, compose integrations and ensure that tasks, data and actions move through a process in a controlled, observable sequence. APO gives enterprises the operational backbone to embed agents across their business processes.
A third plane is now emerging. As enterprises deploy multiple agents from multiple vendors across many domains, a new question becomes unavoidable: how do we apply a consistent envelope of visibility, governance and management across a heterogeneous agent estate. This is the role of the ‘agent control tower’, or ‘agent control plane’. Distinct from a development environment or an orchestration fabric, this plane supervises the entire agent heterogenous landscape in a vendor-agnostic way to keep autonomous behavior aligned with business intent, policy and risk tolerance.
Complex AI systems need out-of-band management
The emergence of the agent control plane follows a core principle in enterprise architecture. As agents proliferate across the ‘build’ plane and the ‘orchestrate’ plane, governance must sit outside both planes in order to provide independent visibility, enforce consistent policies and maintain control when runtime environments behave unpredictably. This creates a distinct need for an out-of-band ‘oversight’ plane that can enforce policy and maintain trust regardless of how or where agents are built and executed.
Enterprises already rely on external control planes in other complex software domains. At Airbnb, an experimentation platform uses independent guardrails that automatically halt experiments whose metrics drift beyond acceptable thresholds. These guardrails operate outside the service being tested. At JPMorgan, a Model Risk Governance function validates and monitors AI and ML models independently from the product teams that build them. In regulated banking environments, frameworks such as Singapore’s FEAT Principles require AI systems to be subject to independent oversight, audit trails and accountable human decision-making. In each case, the governance mechanism sits separate from the runtime environment because that separation is what maintains trust and operational reliability as systems scale. Agents introduce similar and often greater complexity. As interoperability standards solidify, agents from diverse vendor ecosystems and platforms will increasingly collaborate across ecosystem boundaries. As a result, enterprises cannot depend solely on the agent platform itself or on the process orchestrator around it to enforce the right boundaries. Oversight must live outside the agent’s execution loop so that monitoring, policy enforcement and intervention remain available even when an agent or runtime behaves unpredictably.
Introducing Forrester’s ‘agent control plane’ research stream
An agent control plane is an enterprise control plane that inventories, governs, orchestrates and assures heterogeneous AI agents across vendors and domains. It plays a role similar to a control plane in cloud or network architectures, sitting above the underlying systems to provide unified oversight and intervention. Key features include (but are not limited to):
Agent inventory and identity: maintaining a single, accurate catalogue of all agents, to provide a clear view of what exists and what each agent is allowed to do.
Policies and guardrails: centralizing business, risk, and technical policies for agents and applying them consistently at runtime.
Monitoring and insights: offering continuous visibility into agent behavior, performance, and outcomes to help teams understand how agents are behaving across processes and platforms.
Control and coordination: providing operations, risk, and engineering teams with the tools to manage how agents work together across environments and ecosystems, using shared protocols and playbooks.
Risk, compliance, and audit: Recording decisions, interventions, and policy changes in an auditable way and helping translate high-level risk frameworks into concrete controls applied to agent activity.
Agent control planes already exist in an early and uneven form, largely embedded within individual vendor ecosystems and constrained by platform-specific assumptions. Even so, we are beginning to see vendors and industry initiatives move toward broader, vendor-agnostic capabilities that reflect the need for governance across a distributed agent estate. Over the next 12-24 months we believe this will solidify into a clearer market with distinct offerings, evaluation criteria and value propositions.
I am initiating focused research on this emerging space and welcome participation from vendors building any part of this governance plane. If you are developing capabilities that align this space and would like to brief me, please use our site to schedule a briefing. If you have questions about this space and are a Forrester client, please schedule an inquiry or guidance session.
