Building Agentic Flows: A Hands-On Project Plan

I've been putting together a plan to build and teach a series of hands-on Agentic Flow projects — covering AI Agents, Agentic Flows, and Multi-Agent systems. Eight tiered projects, each one progressively more complex, designed to work both as a personal portfolio and as lecture material.

Here's the plan:

• Goal (Must Have): Design and implement hands-on Agentic Flow projects, each structured around Model, Tools, and Orchestration.
• Stretch Goal: Package into a full lecture curriculum and publish on Inflearn.
• Period: Roughly 6–10 weeks, one level at a time.
• How: Claude Pro for Level 1, Gemini 2.0 Flash (free API) from Level 2, with n8n, LangChain, and Claude Code / Cursor.
• Why: To build real Agentic Flow skills — and create lecture content grounded in actual implementation, not slides.

The Projects

Level 1.1 — Single Agent (Connector)

• Goal: Build an AI assistant that reads Google Calendar and Gmail with a single click
• Model: Claude Sonnet (claude.ai)
• Tools: Google Calendar, Gmail (Connector)
• Orchestration: claude.ai single Agent loop (automatic)
• Dev Tool: None

Level 1.2 — Single Agent (MCP)

• Goal: Rebuild the same AI assistant at the protocol level by writing the MCP config directly
• Model: Claude Sonnet (Claude Desktop)
• Tools: Google Calendar MCP, Gmail MCP, Filesystem MCP
• Orchestration: Claude Desktop + MCP server list
• Dev Tool: Cursor

Level 2 — Agentic Flow (External Orchestration)

• Goal: Build a workflow that receives an email, lets an AI decide the action, and auto-sends a Slack message
• Model: Gemini 2.0 Flash (free API)
• Tools: Webhook, HTTP Request, Slack/Gmail delivery
• Orchestration: n8n — Trigger → Agent decision → Action
• Dev Tool: Claude Code

Level 3 — Agentic Loop (ReAct)

• Goal: Build a research bot that autonomously searches, summarizes, and saves results by looping Think → Act → Observe
• Model: Gemini 2.0 Flash (free API)
• Tools: Web Search, file storage, text summarization
• Orchestration: n8n (external trigger) + ReAct loop (internal)
• Dev Tool: Cursor

Level 4.1 — Multi-Agent Flow (n8n)

• Goal: Build a multi-agent research system where Planner delegates to Researcher, Writer, and Reviewer — orchestrated visually in n8n
• Model: Gemini 2.0 Flash × 4
• Tools: Web Search, document parsing, markdown storage, quality check
• Orchestration: n8n — each node is one Agent; Planner → Researcher / Writer / Reviewer
• Dev Tool: Claude Code

Level 4.2 — Multi-Agent Flow (LangChain)

• Goal: Rebuild the same multi-agent system from 4.1 in LangChain code to compare the two approaches
• Model: Gemini 2.0 Flash × 4
• Tools: Web Search, document parsing, markdown storage, quality check
• Orchestration: LangChain — Agent handoff + state management in code
• Dev Tool: Claude Code

Level 5.1 — Autonomous Agentic Flow (n8n + Human-in-the-loop)

• Goal: Build an autonomous QA pipeline where three Agents test APIs end-to-end, with a human approval gate — orchestrated in n8n
• Model: Gemini 2.0 Flash × 3 (Planner / Executor / Reporter)
• Tools: API exploration, pytest execution, bug DB storage, result notifications
• Orchestration: n8n (outer pipeline + HITL approval gate) + ReAct loop (inner)
• Dev Tool: Claude Code + Cowork

Level 5.2 — Autonomous Agentic Flow (LangChain + Human-in-the-loop)

• Goal: Rebuild the same QA pipeline from 5.1 using LangChain Graph to complete a full Agentic Flow in code
• Model: Gemini 2.0 Flash × 3 (Planner / Executor / Reporter)
• Tools: API exploration, pytest execution, bug DB storage, result notifications
• Orchestration: LangChain Graph + interrupt node (HITL) + ReAct loop (inner)
• Dev Tool: Claude Code + Cowork

---

Agentic Flow 구현해보기 (한국어)

AI Agent와 Agentic Flow 프로젝트를 직접 구현해보기 위한 계획을 세웠다. 난이도별 8개 프로젝트로 구성했으며, 개인 포트폴리오와 강의 자료로 동시에 활용할 수 있도록 설계했다.

• 목표 (필수): Model, Tools, Orchestration 세 가지 구성 요소를 명확히 정의한 Agentic Flow 프로젝트를 구현한다
• 추가 목표: 인프런 강의로 출시한다
• 기간: 대략 6~10주
• 방법: Level 1은 Claude Pro, Level 2부터 Gemini 2.0 Flash (무료 API), n8n / LangChain / Claude Code / Cursor 활용
• 이유: 챗봇 래퍼가 아닌 진짜 Agentic Flow 역량을 쌓고, 실제 구현 경험 기반의 강의를 만들기 위해

프로젝트 목록

Level 1.1 — Single Agent (Connector)

• 목표: Google Calendar와 Gmail을 읽고 요약하는 AI 비서를 클릭 한 번으로 만든다
• Model: Claude Sonnet (claude.ai)
• Tools: Google Calendar, Gmail (Connector)
• Orchestration: claude.ai 단일 Agent 루프 (자동)
• 개발 도구: 없음

Level 1.2 — Single Agent (MCP)

• 목표: MCP 설정 파일을 직접 작성해서 같은 AI 비서를 프로토콜 레벨로 구현한다
• Model: Claude Sonnet (Claude Desktop)
• Tools: Google Calendar MCP, Gmail MCP, 파일시스템 MCP
• Orchestration: Claude Desktop + MCP 서버 목록
• 개발 도구: Cursor

Level 2 — Agentic Flow (External Orchestration)

• 목표: 이메일 수신 → AI 판단 → Slack 자동 발송까지 이어지는 업무 자동화 파이프라인을 만든다
• Model: Gemini 2.0 Flash (무료 API)
• Tools: Webhook, HTTP Request, Slack/Gmail 발송
• Orchestration: n8n — Trigger → Agent 판단 → Action
• 개발 도구: Claude Code

Level 3 — Agentic Loop (ReAct)

• 목표: 주제를 입력하면 스스로 검색·요약·저장을 반복하는 리서치 봇을 ReAct 패턴으로 구현한다
• Model: Gemini 2.0 Flash (무료 API)
• Tools: Web Search, 파일 저장, 텍스트 요약 함수
• Orchestration: n8n (외부 트리거) + ReAct 루프 (내부)
• 개발 도구: Cursor

Level 4.1 — Multi-Agent Flow (n8n)

• 목표: Planner가 태스크를 분해하고 Researcher/Writer/Reviewer가 협업하는 Multi-Agent 시스템을 n8n으로 시각적으로 구성한다
• Model: Gemini 2.0 Flash × 4
• Tools: Web Search, 문서 파싱, 마크다운 저장, 품질 체크 함수
• Orchestration: n8n — 노드 하나가 Agent 하나, Planner → Researcher / Writer / Reviewer
• 개발 도구: Claude Code

Level 4.2 — Multi-Agent Flow (LangChain)

• 목표: 4.1과 동일한 Multi-Agent 시스템을 LangChain 코드로 구현해 두 방식의 차이를 비교한다
• Model: Gemini 2.0 Flash × 4
• Tools: Web Search, 문서 파싱, 마크다운 저장, 품질 체크 함수
• Orchestration: LangChain — Agent 간 핸드오프 + 상태 관리 (코드)
• 개발 도구: Claude Code

Level 5.1 — Autonomous Agentic Flow (n8n + Human-in-the-loop)

• 목표: API를 자동 탐색·테스트·리포트하는 QA Agent 파이프라인을 n8n + Human-in-the-loop으로 구축한다
• Model: Gemini 2.0 Flash × 3 (Planner / Executor / Reporter)
• Tools: API 탐색, pytest 실행, 버그 DB 저장, 결과 알림
• Orchestration: n8n (외부 파이프라인 + HITL 승인 게이트) + ReAct 루프 (내부)
• 개발 도구: Claude Code + Cowork

Level 5.2 — Autonomous Agentic Flow (LangChain + Human-in-the-loop)

• 목표: 5.1과 동일한 QA Agent를 LangChain Graph로 구현해 완전한 Agentic Flow를 코드 레벨로 완성한다
• Model: Gemini 2.0 Flash × 3 (Planner / Executor / Reporter)
• Tools: API 탐색, pytest 실행, 버그 DB 저장, 결과 알림
• Orchestration: LangChain Graph + interrupt 노드 (HITL) + ReAct 루프 (내부)
• 개발 도구: Claude Code + Cowork