Practical notes: Inside the Architecture of a Multi-Agent System: A Technical

Use this as an operator-facing rebuild of the ideas in “Inside the Architecture of a Multi-Agent System: A Technical Guide for B2B Crypto Builders”: clear stages, ordered code slots, and recovery notes that survive a handoff. The Overview stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

What Is a Multi-Agent System?

For the What Is a Multi-Agent stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

```text
User Request
     ↓
Orchestrator
     ↓
Research Agent
     ↓
Risk Agent
     ↓
Portfolio Agent
     ↓
Policy Check
     ↓
Execution Agent
     ↓
Blockchain
     ↓
Verification
```

Core Architecture of a Multi-Agent System

For the Core Architecture of a stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

Agent Runtime: Where the Reasoning Happens

For the Agent Runtime Where the stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Prefer structured outputs with schema validation over free-form prose when the next step is code or a tool call. For the Agent Runtime Where the stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

  ```text
Input
 ↓
Build Context
 ↓
Ask Model to Reason
 ↓
Choose Action
 ↓
Call Tool / Agent / Return Result
 ↓
Receive Result
 ↓
Update Context
 ↓
Reason Again
```

2. Orchestrator: The Control Center

When working through the 2 Orchestrator The Control stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

LLM-Driven Orchestration

When working through the LLM-Driven Orchestration stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Cache stable system instructions and tool schemas. Re-sending identical preamble is a common source of burn.

           ```text
        Supervisor Agent
              │
        ┌─────┼─────┐
        ▼     ▼     ▼
   Research  Risk  Analytics
```

Code-Driven Orchestration

When working through the Code-Driven Orchestration stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the Code-Driven Orchestration stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

```text
Research
   ↓
Risk Analysis
   ↓
Policy Check
   ↓
Simulation
   ↓
Approval
   ↓
Execution
```

Why Hybrid Orchestration Works Well

The Why Hybrid Orchestration Works stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

3. Shared State: How Agents Remember the Workflow

The 3 Shared State How stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

               ```text
                Shared State
                    │
         ┌──────────┼──────────┐
         ▼          ▼          ▼
    Research      Risk     Execution
     Agent        Agent     Agent
        └──────────┼──────────┘
                   ▼
             Updated State
```

4. Agent Communication: Handoffs vs Tools

The 4 Agent Communication Handoffs stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Expose tools with narrow schemas and explicit side-effect labels. Hosts need to know which calls mutate state before they auto-approve. The 4 Agent Communication Handoffs stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

```text
Research Agent
↓
Risk Agent
↓
Portfolio Agent
```

     ```text
      Supervisor
      /    |   \
 Research Risk Analytics
```

5. Tool Layer: Connecting Agents to Real Systems

For the 5 Tool Layer Connecting stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Authenticate at the gateway and re-authorize at the data plane. A bearer token alone is not a tenancy boundary.

6. The Critical Boundary: AI vs Blockchain Execution

For the 6 The Critical Boundary stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness.

```text
Agent Decision
     ↓
Schema Validation
     ↓
Policy Check
     ↓
Risk Check
     ↓
Transaction Simulation
     ↓
Authorization
     ↓
  Signing
     ↓
Blockchain
```

7. Guardrails and Failure Handling

For the 7 Guardrails and Failure stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Put human approval on edges that spend money or change production data. Compile-time wiring does not equal business completeness. For the 7 Guardrails and Failure stage, define the inputs, the owner of the step, and the exit criteria before changing code. Operators should be able to re-run the step from a known checkpoint without guessing hidden state. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

```text
Agent Output
     ↓
Schema Validation
     ↓
Business Validation
    ↓
Data Validation
    ↓
Tool Execution
    ↓
Result Verification
```

Why Idempotency Matters

When working through the Why Idempotency Matters stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

8. Observability: Understanding What the Agents Did

When working through the 8 Observability Understanding What stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

9. Production Multi-Agent Architecture for B2B Crypto

When working through the 9 Production Multi-Agent Architecture stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Treat this stage as a contract between inputs and validated outputs. Name the artifacts, define success checks, and refuse silent partial completion. Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node. When working through the 9 Production Multi-Agent Architecture stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest. Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Multi-Agent Systems Are Distributed Systems With LLMs Inside

The Multi-Agent Systems Are Distributed stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Document the happy path and the recovery path together. Retries, human gates, and dead-letter handling are part of the product, not later polish. Budget tokens per turn and per session. Agentic tools expand context aggressively; hard caps keep demos from becoming surprise invoices.

Final Takeaway

The Final Takeaway stage works best when treated as a measurable surface. Capture one golden transcript, one failure case, and the rollback note before expanding scope. Prefer small, testable units over sprawling scripts. When a step fails, the failure should point at a single responsibility rather than a tangled pipeline. Keep graph state flat and typed. Nested blobs hide which node wrote which field and break resume after interrupts.

Operational checklist

When working through the Operational checklist stage, write down the contract first: required inputs, success signal, and what happens on partial failure. That checklist keeps later code changes honest.

Record timings and token or query cost next to functional results. Cost visibility early prevents surprise bills when the path moves from demo to shared environments.

Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Pin dependency versions and record the image digest that ran the demo. Reproducibility beats tribal knowledge.

Keep configuration outside application code. Environment files, secret stores, and feature flags belong in one place operators can audit without reading the whole graph.

Checkpoint after expensive steps. Resume should not re-bill the same LLM call when an operator retries a later node.

Before promoting the stack, freeze versions, capture a golden transcript for the critical path, and confirm rollback steps. Shared environments need rate limits, tenancy checks, and a clear owner for secret rotation. Prefer boring reliability over clever one-off demos.

Batch note for d29901c09602: keep provider keys out of the repo, set a per-session token ceiling, and store transcripts next to the eval fixtures so later model swaps stay comparable.