Multi-Agent System Architecture Blueprint

# Multi-Agent System Architecture Blueprint

You are a senior AI systems architect. Design a production-ready multi-agent system for [PROCESS OR BUSINESS DOMAIN]: autonomous research pipeline, customer lifecycle management, code review and deployment, or financial audit workflow.

## Agent Inventory

For each agent in the system define:
- Agent name and single responsibility (one job only)
- Inputs it accepts: data types, sources, format
- Outputs it produces: data types, targets, format
- Tools and APIs it has access to
- Model tier required: fast/cheap vs. capable/expensive

## Orchestration Logic

Design the controller layer:
- Which agent triggers which: sequential, parallel, or conditional branching
- How context passes between agents: shared memory, message bus, or structured JSON handoff
- How the orchestrator handles an agent that returns null or errors
- Maximum retry count and fallback behavior

## Human-in-the-Loop Points

Define exactly where a human must review before the system continues:
- What triggers a checkpoint
- Information presented to the human: context, options, recommended action
- Maximum wait time before escalation or timeout

## Observability and Debugging

Specify the logging and monitoring design: what each agent logs, how to trace a single job across all agents, and how to replay a failed run from any step.