9.6 Agent frameworks
You can pick a framework and know what it is hiding from you.
Before:08. Large Language Models
Frameworks like LangGraph and CrewAI package the loop, state and orchestration — genuinely useful, provided you know what they are doing for you, which the from-scratch topic earlier made sure of. Checkpointed graph state is LangGraph's distinctive idea. It sits after the fundamentals deliberately. The investment advice is blunt: learn the concepts, rent the framework — abstractions here have short half-lives, and knowledge of the layer beneath them is what transfers.
Work through these
LangGraph: graphs, state, checkpoints
A framework built around explicit graphs of steps with persisted state, which makes long-running and resumable agents practical. The explicitness is its main advantage.
CrewAI, AutoGen, and role-based orchestration
Frameworks organised around several agents with assigned roles, which reads naturally and can be hard to debug. The abstraction fits some problems and obscures others.
Vendor SDKs and their lock-in
Provider-supplied toolkits are convenient and tie your code to one provider's shape. Knowing where that tie is helps you decide how much to accept.
When to drop the framework
Frameworks hide the loop, and when the loop is the thing you need to control, they cost more than they save. Recognising that moment is the point of this item.
Sign in to keep your progress.
Free resources
We haven't checked most of these for screen reader use yet.
Links last checked 29 Aug 2026.
Stuck here?
Ask a mentor. A real person answers, and they can see exactly which topic you're on. Usually within a couple of working days.
Checking your session…
Topics shown in module order.