Different abstractions, same goal
Each framework provides a powerful but distinct way to build complex agentic systems. The loan-processing use case is implemented in all three.
| Framework | Abstraction | Strength |
|---|---|---|
| Google ADK | Production agent / full lifecycle | Enterprise-grade building, testing, and deploying of robust, manageable agent services. |
| CrewAI | Team of specialists (hierarchical process) | Modeling collaborative, role-based tasks with delegation. |
| LangGraph | State machine | Fine-grained control, explicit state management, and robust error handling for cyclical flows. |
Match the abstraction to the problem
Key concept · Choose the right abstraction
- Use CrewAI's team metaphor for role-based, collaborative tasks — especially where agents delegate to one another.
- Use LangGraph's state machine for complex, cyclical processes that need explicit state and fine-grained control over flow and errors.
- Use ADK's production agent model for enterprise-grade, testable, and manageable services across the full lifecycle.
Whichever you pick, frameworks are accelerators: they supply the essential abstractions — agentic planners, state managers, tool dispatchers — so you can stop writing boilerplate and start applying the architectural patterns that create value. They are key enablers for reaching Level 5 (single-agent) and Level 6 (multi-agent) systems.
Observability and governance
Advanced agentic systems demand a mature approach to observability and governance. The chapter connects framework choice to traceability — a core component of responsible AI — using tooling built for the job.
- LangSmith
- Detailed tracing and debugging of agentic chains, for observability into complex flows.
- OpenTelemetry
- Standardized, vendor-neutral traceability across the system — the backbone of governance.
The lesson that closes the chapter: powerful frameworks accelerate building, but production-readiness still rests on the disciplines the book has stressed throughout — traceability, governance, and responsible AI.
Chapter 15 quiz
Thirteen questions on the three frameworks, choosing an abstraction, and observability/governance. Answer first, then expand Show answer.
Part A · Multiple choice
Which framework is best suited to complex, cyclical processes needing explicit state and fine-grained control?
- A CrewAI
- B LangGraph
- C ADK
- D None
Show answer
B — LangGraph. Its state-machine model gives explicit state and error control.
Which framework uses a "team of specialists" metaphor with a hierarchical process?
- A CrewAI
- B LangGraph
- C ADK
- D OpenTelemetry
Show answer
A — CrewAI. Strong for role-based, collaborative tasks with delegation.
Which framework is positioned as an enterprise-grade toolkit for the full build/test/deploy lifecycle?
- A CrewAI
- B LangGraph
- C Google ADK
- D LangSmith
Show answer
C — Google ADK. Focused on robust, manageable agent services.
Why are frameworks called "accelerators"?
- A They make the LLM run faster
- B They supply planners, state managers, and tool dispatchers so you apply patterns instead of writing boilerplate
- C They eliminate the need for governance
- D They remove the LLM
Show answer
B. They provide the essential abstractions for Level 4–5 systems.
These frameworks are identified as key enablers for which maturity levels?
- A Levels 0–1
- B Levels 5 and 6 (single- and multi-agent)
- C Level 2 only
- D No levels
Show answer
B. Single-agent (L5) and multi-agent (L6) systems.
Which tool provides standardized, vendor-neutral traceability?
- A OpenTelemetry
- B FunctionTool
- C ThinkingConfig
- D GGUF
Show answer
A — OpenTelemetry. LangSmith provides framework-level tracing/debugging.
Part B · True or false
You must build agentic planners and state managers from scratch for every project.
Show answer
False. Frameworks provide these abstractions so you can focus on patterns and value.
The framework you choose should match the shape of your problem.
Show answer
True. Team metaphor (CrewAI), state machine (LangGraph), or production service (ADK).
Advanced agentic systems require mature observability and governance.
Show answer
True. Traceability is a core component of responsible AI.
All three frameworks in this chapter run on Google Gemini as the LLM.
Show answer
True. The loan use case is implemented on Gemini across ADK, CrewAI, and LangGraph.
Part C · Short answer
Name the three frameworks compared in this chapter.
Show answer
Google ADK, CrewAI, and LangGraph.
Match each framework to its core abstraction.
Show answer
ADK = production agent/lifecycle; CrewAI = team of specialists (hierarchical); LangGraph = state machine.
Name two tools used for observability/traceability in production agentic systems.
Show answer
LangSmith (tracing/debugging) and OpenTelemetry (standardized traceability).