Agents and frameworks
Words made simple
A few useful meanings before you read the explanation.
- Agent
- An application that uses a model to choose its next action within defined limits. It can use tools and continue through several steps.
- AI
- Software techniques for tasks such as recognizing patterns, understanding language or generating content.
- Retrieval
- Finding and fetching information that is useful for the current question.
More words from this lesson
- Tool calling
- The model proposes using a named function and supplies its inputs. Application code checks permissions and runs the function.
- Orchestration
- Coordinating tasks: deciding their order, passing data between them, and handling waiting or failures.
- State
- The data describing where a workflow is now, what it already knows, and what still needs to happen.
- Checkpoint
- A saved snapshot of workflow progress that can help it continue later.
- Durable
- Able to retain progress and recover across interruptions. Durable workflows require appropriate storage and retry behavior.
| Concept | Practical explanation |
|---|---|
| Agent | Bounded decision/action/observation loop toward a goal |
| ReAct | Alternate choosing useful action and observing results |
| Router | Select a narrow permitted route |
| Planner/executor | Propose steps separately from controlled execution |
| Supervisor/handoff | Coordinate specialists / transfer task ownership |
| Reflection/critic | Check and revise a draft with bounded retries |
| Multi-agent | Specialized roles only when evaluations justify coordination |
| LangChain | Useful model, prompt, retrieval, and tool components |
| LangGraph | Explicit state, nodes, transitions, checkpointing, and review/resume |
| MAF | Model-driven agents plus explicit workflows in Microsoft ecosystem |
| OpenAI Agents SDK | Agent-focused tools, runners, and orchestration helpers |
| MCP | Standard integration protocol between AI hosts and external capabilities |
| MCP primitives | Tools act; resources provide context; prompts provide task templates |
LangGraph essentials: nodes return state updates; reducers define merge behavior; edges control next steps; model routes need validation; production needs durable checkpoint storage and authorized threads. An interrupted node can restart, so pre-pause side effects may repeat. Checkpointing does not make external writes atomic.
MCP essentials: host manages model interaction; client implements protocol; server exposes capabilities; business service enforces object/action rights. MCP is not a replacement for REST, model tool calling, or authorization. Review discovery and tool-definition changes.
Can you explain it without the notes?
Agents and frameworks
Use your own words. Explain an example and one limitation before checking the explanation.
Check your explanation
LangGraph essentials: nodes return state updates; reducers define merge behavior; edges control next steps; model routes need validation; production needs durable checkpoint storage and authorized threads. An interrupted node can restart, so pre-pause side…
Turn this concept into a clear answer.
Practice this conceptAdapted from 13_AI.MD · “Agents and frameworks”. Examples are study fragments; framework behavior can depend on your version.