AIF-C01
Deep Dive
A foundation model on its own only generates text. An agent gives it the ability to act: query systems, execute actions, and chain steps together to complete a full task, not just answer a question.
Contents
An agent is a system that uses a foundation model to reason about a task, decide what actions to take, execute them using external tools, and chain multiple steps until the user's goal is completed — not just return a text response.
Unlike a simple prompt-response cycle, an agent can: interpret natural-language intent, plan the necessary steps, invoke one or more APIs or functions, observe the results, and decide the next step — iteratively, until the request is resolved.
Simple prompt-response
User asks → model generates text → end. There's no access to external systems or real-time data, and it can't execute any concrete action in the real world.
Agent
User asks for something → the agent decides which API or tool to call → executes the call → evaluates the result → possibly calls another tool → responds with the final result of the completed task.
AWS's managed service for building agents on top of Bedrock foundation models.
What it orchestrates
Key idea for the exam
Bedrock Agents translates natural language into calls to real systems, without you having to manually program the decision flow — the foundation model decides which action to take based on the user's request.
Create an agent in Bedrock capable of calling external APIs to complete multi-step tasks.
+1 more...
"Function calling" or "tool use" is the foundation model's ability to recognize, from natural language, that it needs to invoke a specific tool (a function or API) to complete the task — and to generate the correct parameters for that call.
1. Tool definition: The model is told which tools it has available: their name, purpose, and the parameters they need (e.g. "check_availability(date, service)").
2. Model decision: Given a user request, the model decides whether it needs to use a tool and which one, generating a structured call (not just free text).
3. External execution: The system (outside the model) actually executes that function and returns the result.
4. Final response: The model incorporates the tool's result into its response to the user, or decides to call another tool if it hasn't completed the task yet.
Model Context Protocol (MCP) is an open standard that defines how an agent discovers and calls external tools uniformly, without you having to integrate each API separately with a different connector. In practice, both Bedrock Agents and Bedrock AgentCore can connect to MCP servers to expose tools to the model, instead of manually defining each action group against each system.
The general theory of agents, MCP, and multi-agent architectures is covered in depth in the agentic AI section of the syllabus — here it's enough to recognize that this is the practical mechanism AWS services use to expose tools to an agent.
AWS's platform for taking AI agents to production securely and at scale.
What it solves
While Bedrock Agents focuses on building and orchestrating an agent, AgentCore focuses on deploying, operating, and scaling AI agents in production — regardless of the framework they were built with (Bedrock Agents, Strands Agents, or others).
Key idea for the exam
If the scenario talks about taking agents to production securely, with identity control and policies over what the agent can do, think of AgentCore — not just "building" the agent, but operating it securely at scale.
Strands Agents is an AWS open-source SDK for building AI agents with just a few lines of code, defining the model, the available tools, and the desired behavior simply and programmatically.
Unlike Bedrock Agents (a managed service inside the Bedrock console), Strands Agents is a code library you can use to define agents flexibly in your own applications, which you can then deploy and operate on AgentCore to take them to production.
Key idea for the exam
Bedrock Agents = managed service for building the agent. Strands Agents = open-source SDK for building agents programmatically. AgentCore = platform for deploying and operating either of them in production.
Appointment-booking agent
The user says "I need a dentist appointment next week in the afternoon". The agent queries the calendar API, finds available slots, confirms with the user, and creates the booking by calling the scheduling system.
Database-querying agent
An analyst asks "how many orders were canceled last month in the north region?". The agent translates the question into a structured query, runs it against the database, and returns the answer in natural language.
Technical support agent
It receives an incident, checks the customer's system status via API, and if it detects a known issue, automatically runs the resolution procedure or creates a ticket with the diagnosis already done.
Purchasing / inventory agent
Given "we need to restock product X", the agent checks the current inventory level, calculates the quantity to order based on business rules, and automatically generates the purchase order in the ERP system.
| Aspect |
Chatbot (Amazon Lex) | Agent (Bedrock Agents) |
|---|---|---|
| How it understands the request | Detects predefined "intents" with slots configured in advance | Reasons with a foundation model over open natural language, with no fixed intents |
| Flexibility | Limited to manually designed conversation flows | Can handle varied requests and chain steps not explicitly anticipated |
| Action execution | Can invoke Lambda when a specific intent is completed | Dynamically decides which tools to use and in what order, based on the model's reasoning |
| Ideal case | Structured FAQs, predictable conversational flows (e.g. booking a table with fixed steps) | Complex tasks requiring reasoning, multiple systems, or non-linear steps |
Common exam trap
Lex builds conversational interfaces based on intents/slots you define — it's deterministic and structured. A Bedrock agent reasons with the foundation model about what to do and can dynamically orchestrate multiple tools. If the scenario mentions "the system must decide on its own which API to call based on what the user asks in free-form language", the answer is an agent, not Lex.
Did you get this topic?
Test what you just learned
A company wants an assistant that, given a natural-language request like 'cancel my order #4521 and notify the warehouse', queries the order system, executes the cancellation, and then automatically sends the notification to the warehouse, all in a single interaction. Which AWS capability best describes this need?
Put what you learned to the test with original AIF-C01 exam questions.
Start AIF-C01 mock examUnlimited mock exams with the Pro plan
Log in to track your progress.