생성형 AI 학습 보조 도구 V2 for 강의 노트

--- title: "AI Agents Fundamentals" type: note program: "IBM-GEN_AI_ENGINEERING" course: null module: null lecture: null start_date: null end_date: null tags: [ai_agents, artificial_intelligence, automation] source: "Coursera" duration: null status: TODO aliases: [] --- ## [[SOURCE INFORMATION]] This content comes from an educational video on AI agents fundamentals, covering what AI agents are, how they work, and their key components - a foundational lesson for understanding agentic AI systems. ## [[LEARNING FOCUS]] > [!tip] Learning Objectives > After this material, you will be able to: > 1. Define what an AI agent is and distinguish it from a standalone large language model > 2. Identify the core components of an AI agent's architecture > 3. Explain how agents plan, use tools, and maintain memory > 4. Describe how different agent types specialize in different tasks ## [[NOTES]] ### What is an AI Agent? - An AI agent is a system that uses a large language model to perform actions in a computational environment, such as retrieving information, answering questions, or executing code. - The model acts as the core decision-making engine, adapting to various scenarios based on available resources. ### Key Distinction: Agent vs. Standalone LLM - A standalone LLM is constrained to text-based interactions** - it can generate and receive text but cannot take actions or interact with external data. - An AI agent goes beyond text generation by actually performing tasks and interacting with the environment. - The consequence of this difference: **the agent expands the capabilities of the LLM from text-only operations to real-world actions**. ### Core Agent Components > [!note] The four key components of an AI agent: > 1. **Environment** - the space where the agent operates > 2. **Tools** - utilities the agent can invoke to take actions (e.g., web search, API calls, code execution) > 3. **Agent Core** - the decision engine, typically the LLM, that orchestrates and decides what to do next > 4. **Memory** - allows the agent to store and recall past interactions and learned information from the environment ### The Role of Tools and Memory - Tools are essential for the agent to perform useful actions, enabling it to search, compute, and interact beyond its internal knowledge. - Memory is critical for self-improvement** - it stores information about past actions and their outcomes, which the agent can use to make better decisions over time. ### How Agents Work - The agent uses the **plan-execute-reflect loop** - determine (plan), run an action (execute), and then assess whether the process worked and could be improved (reflect). - The reflection phase drives the agent to become more effective over time. ### Video Series Overview - The video is the first in a short series on AI agents. - The series covers the foundations of AI agents, types of agentic architectures, how to self-host an agent library, and future career roles in the AI agent space. ## [[EXAMPLES, PATTERNS, OR DEMONSTRATIONS]] ### Agent - Environment Interaction Loop ```mermaid flowchart LR A[Agent Core - LLM] -->|Generation| B[Tools & Actions] B -->|Execution| C[Environment] C -->|Observation| D[Memory] D -->|Context & Feedback| A A -->|Recap & Adapt| A ``` ### Visualization of an AI Agent in Action 1. The user prompts the agent with a task. 2. The agent core (LLM) plans which tool to invoke. 3. The tools execute the action (e.g., a web search). 4. The memory stores the results and experience. 5. The agent core processes the observation and plans the next step. ## [[KEY TAKEAWAYS]] > [!important] Make it Memorable > 1. AI agents leverage LLMs to act, perform actions with tools, and use memory to continually improve. > 2. An agent unlocks capabilities beyond text dumps - file reading, website searches, and tool calls like calculations. > 3. The major components of an agent are the environment, tools, agent core, and memory - each play a distinct role. > 4. Reflection is key to the agent's self-learning process, making it effective over time. ## [[EXAM-READY SUMMARY]] ### Core Questions **What is this?** An AI agent is an LLM-powered system that interacts with its environment through tools and memory to accomplish tasks. **Why is this important?** Agents represent the shift from text generation to action-taking AI, enabling real-world application and workflow automation. **Why should I learn this?** This is foundational for working with modern AI systems - understanding agents is required for building scalable, automated AI solutions. **When will I need this?** When designing systems that need reasoning, using external tools, or adapting based on experience, use the agent architecture. **How does this work?** The LLM acts as a decision core, which plans and uses tools. Observed outcomes and past context are stored in memory to learn from experience. **What problem does this solve?** It bridges AI from confined text-based QA to dynamic, autonomous task execution. ### Patterns & Mental Models > [!example] Pattern Template > When you see the need for an AI to take action, think AGENT, this usually means a core model + tools + memory architecture with an iterative loop. ### Simplified Re-Explanation > [!note] The Coffee Shop Test > Think of an LLM as a cook who can only give you recipes. An AI agent is a full kitchen team - the agent picks the tools, fetches ingredients (tools and memory), adjusts based on what's available, and improves over time, making the meal itself. ### Rapid Review Checklist - [ ] Can you explain what an AI agent is in one sentence? - [ ] Can you list the 4 core components of an agent? - [ ] Can you describe the plan-execute-reflect loop? - [ ] Can you explain how an agent differs from a standard LLM? ### FAQ > [!question] Q: Is the LLM the agent itself? > A: Not exactly - the LLM is the core (the decision-maker), but the overall agent includes tools and memory. > [!question] Q: Why do we need memory if tools are available? > A: Memory supports iteration and learning - it stores context from earlier steps that is not inherently available to the tools. ### Career & Real-World Connections - **Industry Applications:** Used in chatbots, autonomous coding assistants, customer service automation, and data analysis pipelines. - **Interview Prep:** > [!important] Interview Alert > Questions often touch on where tools fit into the pipeline, and the value of memory. Practice describing the agent flow as a looped architecture. - **Portfolio & Project Ideas:** Build a simple agent that uses a search tool, and note how it refines a response between loops. - **Pro Tips:** > [!tip] Pro Tip 1. Start with agents that use tools/action, then gradually add memory to make it iterative. 2. Do not over-engineer - clear abstraction between core, tools, memory prevents chaos later on. ### Connections & Related Topics - Related to agentic AI chapter, extends approaches like RAG (Retrieval-Augmented Generation) which learners may have encountered in prior modules. - Next topics in the series include types of agentic approaches and self-hosted agents. ### Motivational Anchor > [!success] You've Got This > Understanding agents moves you from just using models to building automated systems that do real work in the real world. This is where the industry is heading, and you're at the forefront. **END OF NOTES**