6 Application and integration architecture
This chapter covers
- Locating agents in the application estate and designing the bounded interaction surfaces through which they act
- Allocating responsibility across agents, deterministic services, workflows, policy components, state stores, and human roles
- Using conversational contracts to express intent, context, constraints, authority, outcomes, and evidence obligations
- Mediating agents’ access to enterprise APIs, SaaS platforms, legacy systems, workflows, and data operations
- Coordinating specialist agents through Agent Mesh design, events, explicit structure, durable state, and human re-entry paths
- Recognizing five application and integration antipatterns, including Raw API as Tool and Orchestrator as New Enterprise Service Bus
- Applying the chapter’s patterns through an implementation playbook for a single bounded agentic workload
6.1 Designing agent participation in the application estate
6.1.1 Seven ways agents participate in the application estate
6.1.2 Deciding what the agent owns and what stays elsewhere
6.2 Agentic interaction contracts
6.2.1 Pattern: Conversation-as-Interface
6.2.2 The six elements of a conversational contract
6.2.3 API, event, and conversational contracts
6.3 Governed access to existing systems
6.3.1 Pattern: Governed Tool Wrapper
6.3.2 Wrapping SaaS and legacy systems
6.3.3 Idempotency, retry control, compensation, and auditability
6.4 Event-mediated coordination
6.4.1 Commands, decisions, actions, and evidence signals
6.4.2 Topics, streams, and asynchronous coordination
6.5 Governing specialist-agent collaboration
6.5.1 Pattern: Agent Mesh
6.5.2 Supervisor, brokered, and peer-to-peer models
6.6 Human reentry at integration boundaries
6.6.1 Pattern: Fallback-to-Human
6.6.2 Initiating human re-entry
6.6.3 Handoff payloads and return to flow
6.7 Application and integration antipatterns
6.8 Implementation playbook: designing a bounded agentic flow
6.9 Summary