AIF-C01 ยท Generative AI concepts
Context engineering and agentic AI
What context engineering is, plus agent building blocks: tool usage, memory management, workflow orchestration, multi-agent patterns and the Model Context Protocol.
Key points
- Context engineering: choosing what the model sees (instructions, documents, conversation history, tool results) in a limited token window.
- Agent: a loop: plans, calls tools, uses memory (short-term for the conversation, long-term across sessions).
- Multi-agent: agents cooperate.
- Model Context Protocol (MCP): open standard connecting agents to tools and data.
- Amazon Bedrock AgentCore: managed services to run agents.
Source: https://docs.aws.amazon.com/bedrock-agentcore/latest/devguide/what-is-bedrock-agentcore.html
Practice questions
What is the Model Context Protocol (MCP) mainly used for?
- A standard way for agents to connect to tools and data sources
- A method for pre-training foundation models
- A pricing plan for token usage
- An encryption scheme for model weights
Show answer
Answer: A. A standard way for agents to connect to tools and data sources
MCP is an open protocol that standardizes how AI applications and agents connect to external tools and data sources. It is not a training method, a pricing plan or an encryption scheme.
Which items are typically part of the context an application assembles for a model call?
- System instructions describing the task and tone
- Documents retrieved for the current question
- The model's full original training dataset
- Recent turns of the conversation
- The monthly cloud billing invoice
Show answer
Answer: A. System instructions describing the task and tone, B. Documents retrieved for the current question, D. Recent turns of the conversation
Instructions, retrieved documents and recent conversation history are classic context. The model's training data is already baked into its weights, and the billing invoice is not something the model needs.
True or false: the more text you put into the context window, the better the answer always gets.
Show answer
Answer: False
False. The window is limited, every token adds cost and latency, and irrelevant text can distract the model. Context engineering is about including the right information, not the most.
Practice 8 questions on this topic in Bitzu
Short lessons, instant explanations and spaced review of what you get wrong. Free on iOS and Android.
