Agent Control Plane – Workato
Speakers: Adam Seligman (CTO) and Zayne Turner (Developer Advocate), Workato
Agent Control Plane Concept
Every agentic system has two layers:
| Layer | Type | Responsibility |
|---|---|---|
| Reasoning Layer | Probabilistic | LLM, agents, decisions |
| Control Plane | Deterministic | Governance, auth, business logic |
"The control plane needs to be correct every single time."
Seven Factors for Agent Control Plane
| Factor | Description |
|---|---|
| 1. Governance Operations | Protocols are thin; handle enterprise concerns yourself |
| 2. Deterministic Mutations | Control plane owns all creates/writes/deletes |
| 3. Intent-Based Communication | Tool calls about intent, not implementation details |
| 4. Founded Access | Least privilege for probabilistic stakeholders |
| 5. Safe Retries | Different retry patterns for stateless callers |
| 6. Appropriate Errors | Error messages designed for reasoning callers |
| 7. Structural Observability | Observability by design, not just logging |
Key Architectural Principle
"Don't put enterprise concerns into the protocol where they don't belong."
Protocols like MCP are thin by design. Enterprises need to wrap them with governance, auth, and business logic.
Integration Platform Perspective
Workato has been an integration platform for 10+ years. Customers are pulling them into the agent governance space.
"Customers are building agents way faster than the governance and controls can keep up."
Resource: White paper at workato.com/sevenfactors