Building AI Agents with Agent Develeopment Kit Go @DevFest Cloud x Seoul 2025

DevFest Cloud x Seoul 2025
2025-11-30
Building AI Agents with Agent Develeopment Kit Go
박제창, jaichangpark

Transcript

1. Cloud x Seoul 박제창 GDE Dart-Flutter GDG Golang Korea, Flutter Seoul Building AI Agents with Agent Develeopment Kit Go 2025
2. 01 Hands-on ADK Go를 활용하여 Agents 만들기 ! gemini_cli 도 곁들인
3. Agents 개발 프레임워크 • LangChain, LangGraph • OpenAI Agents SDK • Claude Agent SDK • Agent Development Kit • etc…
5. Agent Development Kit Agent Development Kit (ADK)는 AI 에이전트 개발 및 배포를 위한 유연한 모듈식 프레임워크입니다. Gemini 및 Google 생태계에 최적화되어 있지만, ADK는 모델에 구애받지 않고(model-agnostic), 배포 환경에 제약이 없으며 (deployment-agnostic), 다른 프레임워크와의 호환성을 위해 구축되었습니다. ADK는 에이전트 개발이 일반적인 소프트웨어 개발처럼 느껴지도록 설계되었으며, 개발자가 간단한 작업부터 복잡한 워크플로에 이르는 에이전트 아키텍처를 쉽게 생성, 배포 및 오케스트레이션할 수 있도록 지원합니다.
6. Agents 개발 언어 • Python • (New!) Go NOV. 7, 2025 • Java
7. IDE • Visual Studio Code • Antigravity • GoLand
9. CLI • Gemini CLI • https://github.com/google-gemini/gemini-cli
10. Hands-on: Process Chart Structured Output 만들기 도구(Tools)를 활용한 검색 에이전트 만들기 커스텀 도구 (Custom Tools) 만들기 첫 번째 AI 에이전트 만들기 Router Agent Memory & Runner A2A(Agent-to- Agent) Parallel & Sequential Multi-Agents
11. Material https://github.com/JAICHANGPARK/devfest-cloud-seoul-2025
12. BASIC
13. model, err := gemini.NewModel(ctx, "gemini-3-pro-preview", &genai.ClientConfig{ APIKey: "GOOGLE_API_KEY", }) if err != nil { log.Fatalf("Failed to create model: %v", err) }
14. rootAgent, err := llmagent.New(llmagent.Config{ Name: "root_agent", Model: model, Description: "A helpful agent.", Instruction: "You are a helpful assistant. Answer the user's questions.", })
15. config := &launcher.Config{ AgentLoader: agent.NewSingleLoader(rootAgent), } // Package full provides easy way to play with ADK with all available options l := full.NewLauncher() if err = l.Execute(ctx, config, os.Args[1:]); err != nil { log.Fatalf("Run failed: %v\n\n%s", err, l.CommandLineSyntax()) }
16. https://google.github.io/adk-docs/runtime/
17. Tools
18. type getWeatherArgs struct { City string `json:"city" jsonschema:"The city to get weather for."` } func getWeather(ctx tool.Context, args getWeatherArgs) (string, error) { fmt.Printf("[Tool] Getting weather for %s...\n", args.City) return fmt.Sprintf("The weather in %s is Sunny, 25°C", args.City), nil }
19. weatherTool, _ := functiontool.New(functiontool.Config{ Name: "get_weather", Description: "Get weather for a city"}, getWeather, )
20. 구조화 출력, Structured Outputs
21. outputSchema := &genai.Schema{ Type: genai.TypeObject, Properties: map[string]*genai.Schema{ "summary": {Type: genai.TypeString}, "action_items": {Type: genai.TypeArray, Items: &genai.Schema{Type: genai.TypeString}}, }, } https://ai.google.dev/gemini-api/docs/structured-output?example=recip e#go_2
22. routerAgent, err := llmagent.New(llmagent.Config{ Name: "root_agent", Model: model, Description: "A helpful agent. Uses a router to route the user questions.", Instruction: "You are a helpful assistant. Answer the user's questions.", OutputSchema: outputSchema, })
23. Multi-Agents
24. restaurantScout, _ := llmagent.New(llmagent.Config{ Name: "RestaurantScout", Model: model, // 변경: 입력에서 도시를 추출하도록 명시 Instruction: `You are a Restaurant Scout. The user's request will contain a destination city (e.g., "Plan a trip to Tokyo"). 1. Extract the city name from the request. 2. IMMEDIATELY use Google Search to find the top 3 restaurants in that city. 3. Output ONLY a brief list of the restaurants found. Do not ask for clarification.`, Tools: []tool.Tool{geminitool.GoogleSearch{}}, OutputKey: "restaurant_list", })
25. activityScout, _ := llmagent.New(llmagent.Config{ Name: "ActivityScout", Model: model, // 변경: 입력에서 도시를 추출하도록 명시 Instruction: `You are an Activity Scout. The user's request will contain a destination city (e.g., "Plan a trip to Tokyo"). 1. Extract the city name from the request. 2. IMMEDIATELY use Google Search to find the top 3 tourist activities in that city. 3. Output ONLY a brief list of the activities found. Do not ask for clarification.`, Tools: []tool.Tool{geminitool.GoogleSearch{}}, OutputKey: "activity_list", })
26. // 3. Parallel Runner scouts, _ := parallelagent.New(parallelagent.Config{ AgentConfig: agent.Config{ Name: "CityScouts", Description: "Scouts for restaurants and activities in parallel.", SubAgents: []agent.Agent{restaurantScout, activityScout}, }, })
27. planner, _ := llmagent.New(llmagent.Config{ Name: "ItineraryPlanner", Model: model, Instruction: `You are a travel planner. Create a one-day itinerary based on the following research: Restaurants: {restaurant_list} Activities: {activity_list} Combine them into a logical schedule.`, })
28. // 5. Sequential Pipeline tripPlanner, _ := sequentialagent.New(sequentialagent.Config{ AgentConfig: agent.Config{ Name: "TripPlannerPipeline", Description: "Executes scouting and then planning.", SubAgents: []agent.Agent{scouts, planner}, }, })
29. Cloud x Seoul 박제창 GDE Dart-Flutter GDG Golang Korea, Flutter Seoul 감사합니다 . 2025