After completing this lesson, you'll be confident doing the following three things.

• ▸✅ Monolith vs MSA decision checklist with 5 criteria
• ▸✅ Visualizing system diagrams with mermaid
• ▸✅ Separation of concerns (Presentation · Domain · Data)

Keep the learning objectives as a checklist, and close the lesson once you can answer all of them.

MSA (Microservices Architecture) vs Monolith:

Message Queue — asynchronous communication:

API Gateway — single entry point for MSA:

• ▸Routing (request → appropriate service)
• ▸Authentication·authorization (no duplication across services)
• ▸Rate limiting·caching·logging
• ▸Examples: Kong·AWS API Gateway·Envoy·Traefik

> 💡 When to use MSA?: 8+ team members · significantly different traffic patterns by domain. Before that: monolith → modularize → extract.

Event Sourcing — storing events instead of state

• ▸Traditional: update a row in the users table
• ▸ES: append UserCreated·EmailChanged to an events table (immutable)
• ▸Current state = replay of all events

Pros:

• ▸Complete audit log (who·when·what)
• ▸Time travel (reproducing past state)
• ▸Easy to add new read models

Cons:

• ▸Increased complexity (vs. simple CRUD)
• ▸Difficult to change event schema (events are immutable)
• ▸Steep learning curve

CQRS (Command Query Responsibility Segregation):

• ▸Command (write) — normalized DB, transactions
• ▸Query (read) — denormalized·cache·search engine

Example: e-commerce store

• ▸Command: PostgreSQL (order processing·payment)
• ▸Query: Elasticsearch (product search), Redis (cache)
• ▸Sync: change events delivered via Kafka

Good fit when:

• ▸🟢 Read/write ratio is highly skewed (e.g., 10000:1)
• ▸🟢 Complex search and aggregation
• ▸🔴 Simple CRUD → overkill

> 💡 ES + CQRS = often used together. Events → converted into various read models.

Modern use cases:

• ▸Banking: ES is the standard (audit·compliance)
• ▸E-commerce: ES applied to the order domain only
• ▸Gaming: recording player actions

Core takeaway

If you just tell an AI "Build it as MSA", it will simply split apart a monolith. What you get is not real MSA but a distributed monolith (the worst of both worlds).

Why AI alone can't handle architecture

LLMs have learned code patterns, not business domains. Where service boundaries lie — and who owns what data — are decisions humans must make.

❌ Bad prompt:
> "Split this monolith into MSA"

→ AI: User Service, Order Service, Payment Service... naive separation.
→ Result: circular calls between services + broken transactions — a real distributed monolith.

✅ Good prompt:
> "Design an MSA with the following business boundaries:
> - Auth Service: user authentication·sessions. Other services receive only a JWT
> - Order Service: order flow. Communicates with Payment via events (no tight coupling)
> - Notification Service: notifications. Other services publish to a Kafka topic
>
> ACID transactions required at: payment. Use the Saga pattern."

→ AI: implements real MSA with clear boundaries.

Decision checklist — Is my project a monolith or MSA?

5 questions scored (0–2 points each, 10 total):

• ▸0–3 pts → Monolith. Don't worry about anything else. Start with Next.js + one DB
• ▸4–6 pts → Monolith with clear module boundaries. Prepare for future extraction
• ▸7–10 pts → Consider MSA. But: recommended to start with monolith and extract only the parts that become painful

The truth about "start with a monolith"

> "If you can't build a monolith, what makes you think microservices is the answer?" — Simon Brown

95% of startups are well-served by a monolith + well-separated modules. MSA is warranted only when there are clear pain points.

The trap — hidden costs of MSA

At the startup stage, MSA = waste of time and money. Consider it when you reach Toss- or Coupang-level traffic.

The pattern for asking AI about architecture

❌ "Is MSA better or is monolith better?" — too abstract

✅ "My project has 3 team members, 5K daily users, 1 deploy per week, and a single Node.js stack. Does MSA make sense here? Analyze it based on the checklist."

→ AI gives an answer tailored to my situation.

Summary

• ▸AI only knows code patterns. Business boundaries are for humans to decide
• ▸Diagnose your situation with the 5-question checklist
• ▸Start with a monolith. MSA only when you truly need it
• ▸When asking AI, always specify your concrete situation

Knowing the concepts in this lesson lets you give AI specific instructions. Not a vague "fix it" but a request with vocabulary — that's the starting point for saving tokens.

• ▸"Before splitting this monolith into MSA, check it against the 5-point checklist"
• ▸"Draw this system diagram with mermaid"

Why this reduces tokens

Without the concepts, even after receiving AI's response you have to ask "What does that mean?" again. Those follow-up questions eat up tokens. Learn the concepts once and the conversation wraps up in a single exchange.