AI Agent Frameworks Roadmap

The AI Agent Frameworks roadmap teaches the architecture behind coding agents and tool-using agent systems.

The first version is based on OpenHands-style internals because the imported content already contains detailed material about agents, controllers, runtimes, events, tools, memory, services, and conversation lifecycle.

Learning Arc

1. Core model. Understand the difference between a model call, an agent, an action, an observation, and a conversation state.
2. Control flow. Learn how the controller drives the loop, how state changes, and how events are emitted.
3. Runtime boundary. Understand where code execution happens, what the sandbox owns, and how the framework receives observations.
4. Tools and function calls. Study how tool schemas, function-call messages, and action execution interact.
5. Memory and specialization. Learn how memory, microagents, repository context, and service startup shape the agent behavior.

What the Learner Should Retain

By the end of the roadmap, a learner should be able to explain:

• what an agent framework owns beyond a model API call
• how an action becomes runtime work
• how observations return to the controller
• why event streams matter for debugging
• where memory and microagents enter the loop
• how a coding agent maintains conversation and execution state

Attached Flashcards

Agent-framework flashcards should focus on architecture boundaries and flows.

Examples:

• What is the difference between an action and an observation?
• What does the controller coordinate?
• Why does the runtime boundary exist?
• What information belongs in an event stream?
• How do microagents influence a task?

Roadmap Output

The output of this roadmap is the ability to inspect an agent framework and identify the moving parts instead of treating the agent as a black box.