Home
Blog
AI Agents & Orchestration
AI Governance & Security
Brad Griffith
Nov 11, 2026
An AI control plane includes six components: a gateway layer, a registry and catalog, identity, policy and guardrails, cost controls, and audit and observability. Together they let an organization govern every model call, every tool an agent touches, every agent action, and every API request from one place. Workato’s AI Control Plane delivers the six through three components: the AI Gateway with four gateways, the AI Registry and Catalog, and AI Security and Governance.
Use the list as an evaluation checklist. A product that covers two or three of the six is a point tool, not a control plane.
What is the gateway layer?
The gateway layer is the enforcement point for AI traffic. It authenticates each caller, applies policy, limits usage, and logs every call. Without it, policy is only advice, because nothing stands in the path to act on it.
Workato’s AI Gateway has four gateways: Model Gateway for LLM calls, MCP Gateway for tools, Agent Gateway for agent collaboration, and API Gateway for managed endpoints. Our guide to what an AI gateway is covers how they work together.
What is the AI registry and catalog?
The registry and catalog is the inventory of what you have built and approved. It lists every model, agent, tool, and MCP server with an owner and controlled access, so teams reuse what exists and IT knows what is running.
Workato’s MCP registry shows how this works in practice. Servers stay private while you build them, an administrator publishes them, and developer access requests route to an admin with a note. The registry is itself an MCP server with three tools (search, get details, list tools), so agents discover servers at run time.
The broader AI Registry is designed to extend this to every approved AI asset, built on Workato or on an external platform. It is designed to give admins control over access, usage insights, and approval for listing, and to give end users a front door where they request access and track their own usage.
What does identity cover?
Identity covers who the agent acts as, and whether the source system verifies it. A control plane needs one identity model across the gateways, not a different login for each product. Otherwise you end up reconciling who did what across several systems after an incident.
Workato uses two layers. Workato Identity, with SSO and groups, controls access to the server. Verified User Access then runs the downstream action with the end user’s own connection, so records show the real user and not a service account, and end users stay separate from builders.
What are policy and guardrails?
Policy and guardrails define what content and data an agent may handle. They cover PII handling, content safety, denied topics, and blocked words, and they work best when they apply the same way wherever an agent runs. A guardrail that only exists in one app is a gap everywhere else.
Workato agents (Genies) offer content-safety screening at low, medium, or high, along with 31 default PII types plus custom regex types. For a match, you choose whether the Genie tokenizes, redacts, or refuses. Tokenizing means the agent never sees the real value.
What do cost controls include?
Cost controls include token metering, limits, and attribution. Finance needs to see what AI costs and who spent it, and engineers need limits that stop runaway spend before it shows up on an invoice.
Workato Model Gateway reports token usage by input, output, cache read, and cache write, grouped by model, provider, and route. Rate and token limit policies attach per team, project, or automation. In Workato’s AI Control Plane you can see spend by team, model, and cost center and set budgets that alert well before they are blown.
What does audit and observability include?
Audit and observability give you a record of every action and a way to trace it. That record should say who invoked the call, what they invoked, under which policy, and what happened. If you cannot answer all four for one call, you do not have audit.
Workato’s MCP tool-call log records authentication method, client, user, duration, request, response, and outcome. Genie conversation traces show each guardrail check, LLM call, and tool call. Agent Evals are designed to add continuous evaluation of deployed agents across correctness, tool selection, policy adherence, safety, and outcome quality. Logs can stream to an external SIEM, and audit logs are retained for a configurable period of up to 90 days, with extended retention through an external SIEM.
What does an AI control plane not include?
An AI control plane does not include the execution of the work. It governs what agents may do, but it does not supply the skills, workflows, and connectors that do it. Without them, agents improvise business processes from raw APIs.
Workato’s Execution Plane supplies that layer: Enterprise Skills, recipes, and connectors. That is why Workato is the Control and Execution Platform for enterprise AI. Our piece on why an AI control plane is not enough makes the longer argument.
How do you evaluate an AI control plane?
Ask for a demonstration of each component, not a feature list. You want to see an agent act under a named user’s permissions, a policy blocked at the gateway, an approved asset found in the registry, a spend report by team, and a log line for one call.
Then ask whether the platform that governs the call also does the work. Workato has 25,000+ customers, 1 trillion+ tasks processed, and a 99.99% platform SLA.
Summary: What an AI Control Plane Includes
An AI control plane includes a gateway, registry, identity, policy, cost controls, and audit. The gateway enforces while the registry inventories, one identity model covers every surface, and policy and guardrails apply across traffic. Cost controls meter and limit tokens, audit records every action, and execution of the work remains a separate plane.
FAQ
What are the components of an AI control plane?
A gateway layer, registry and catalog, identity, policy and guardrails, cost controls, and audit.
Can an AI control plane revoke access quickly?
A control plane should revoke access in one place and not five. Workato describes one identity model and “one switch to shut it all off.” Workato’s proxy also revokes brokered third-party credentials at once, and revoking a user revokes their agent’s access.
Is an AI registry part of a control plane?
Yes. It is the inventory of approved assets.
What is the difference between a control plane and observability?
Observability shows what happened. A control plane also decides what is allowed.
Does an AI control plane replace IAM?
No. It uses the identity provider and adds agent and tool permissions.
About the Author
More on AI Agents & Orchestration, AI Governance & Security
Discover the latest product updates and features
What Are Agentic Workflows?
Nov 11, 2026
How Do You Switch LLM Providers Without Code Changes?
Nov 11, 2026
Agentic Orchestration vs Orchestrating Agents
Nov 11, 2026
Table of Contents
