advanced⏱65 minLección 1 de 10

Introduction to AI Agents

Understand what AI agents are, their core capabilities, real-world use cases, and how they differ from traditional LLM interactions.

Introduction to AI Agents

What Are AI Agents?

AI agents are autonomous software systems that use large language models to perceive their environment, reason about goals, and take actions to accomplish tasks. Unlike simple chatbots that respond to individual prompts, agents maintain context, use tools, plan multi-step workflows, and learn from feedback.

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

ℹ️Note

An AI agent is not just an LLM with a system prompt. It's a loop: perceive → reason → act → observe → re-reason. This cycle is what distinguishes agents from simple chat completions.

Core Capabilities of AI Agents

AI agents possess several distinct capabilities that set them apart from traditional language model interactions:

1. Tool Use

Agents can call external tools and APIs to interact with the world. This includes file operations, web searches, code execution, database queries, and more.

python

import json

# Example: Agent tool registry
class ToolRegistry:
    def __init__(self):
        self.tools = {}

    def register(self, name, handler, description, parameters):
        self.tools[name] = {
            "handler": handler,
            "description": description,
            "parameters": parameters
        }

    def call(self, name, **kwargs):
        if name not in self.tools:
            raise ValueError(f"Tool '{name}' not found")
        return self.tools[name]["handler"](**kwargs)

registry = ToolRegistry()
registry.register(
    name="read_file",
    handler=lambda path: open(path).read(),
    description="Read contents of a file",
    parameters={"path": {"type": "string", "description": "File path"}}
)
registry.register(
    name="web_search",
    handler=lambda query: f"Results for: {query}",
    description="Search the web for information",
    parameters={"query": {"type": "string", "description": "Search query"}}
)

# Agent invokes tools dynamically
result = registry.call("read_file", path="config.json")
print(f"File contents: {result}")

2. Multi-Step Reasoning

Agents break complex goals into smaller sub-tasks, execute them sequentially, and adapt their plans based on intermediate results.

json

{
  "agent_plan": {
    "goal": "Deploy the application to production",
    "steps": [
      {
        "step": 1,
        "action": "Run tests",
        "tool": "bash",
        "command": "pytest tests/",
        "expected": "All tests pass"
      },
      {
        "step": 2,
        "action": "Build artifacts",
        "tool": "bash",
        "command": "npm run build",
        "expected": "Build succeeds"
      },
      {
        "step": 3,
        "action": "Deploy to server",
        "tool": "bash",
        "command": "rsync -avz dist/ user@server:/var/www/",
        "conditional": "Only if steps 1 and 2 succeed"
      }
    ]
  }
}

3. Memory and Context Management

Agents maintain conversational memory, store facts, retrieve relevant context, and summarize long histories to stay within context windows.

python

class AgentMemory:
    def __init__(self, max_tokens=8192):
        self.short_term = []
        self.long_term = {}
        self.max_tokens = max_tokens

    def add_interaction(self, role, content):
        self.short_term.append({"role": role, "content": content})
        if self._estimate_tokens() > self.max_tokens:
            self._summarize()

    def _estimate_tokens(self):
        return sum(len(m["content"].split()) for m in self.short_term)

    def _summarize(self):
        # Summarize oldest half of interactions
        mid = len(self.short_term) // 2
        old_content = " ".join(
            m["content"] for m in self.short_term[:mid]
        )
        summary = f"[Summary of earlier context: {old_content[:200]}...]"
        self.short_term = self.short_term[mid:]
        self.short_term.insert(0, {"role": "system", "content": summary})

    def recall(self, key):
        return self.long_term.get(key)

    def remember(self, key, value):
        self.long_term[key] = value

memory = AgentMemory()
memory.add_interaction("user", "My project is called 'Nova'. It's a web app.")
memory.add_interaction("assistant", "I'll help you build Nova.")
print(memory.recall("project_name"))  # None - not stored in long-term

# Store important facts long-term
memory.remember("project_name", "Nova")
memory.remember("tech_stack", ["Python", "React", "PostgreSQL"])
print(f"Project: {memory.recall('project_name')}")

💡Tip

Use long-term memory for facts that persist across sessions (user preferences, project constants) and short-term memory for conversational context within a single session.

Agent Autonomy Levels

Agents operate at different levels of autonomy depending on the use case and trust requirements:

Level	Name	Description	Example
0	No autonomy	LLM generates text only, no tools	Chat completion
1	Tool-assisted	LLM can suggest tools but user approves	Copilot suggestions
2	Semi-autonomous	Agent executes tools but asks for confirmation on critical actions	Code review with human approval
3	Conditional autonomous	Agent executes freely within defined guardrails	CI/CD pipeline automation
4	Fully autonomous	Agent operates independently with self-defined goals	Long-running research agents

yaml

# Agent autonomy configuration
agent_config:
  name: "semi-autonomous-coder"
  autonomy_level: 2
  guardrails:
    - action: "write"
      requires_approval: true
      paths: ["src/**"]
    - action: "bash"
      requires_approval: true
      patterns: ["rm *", "sudo *", "> /dev/*"]
    - action: "read"
      requires_approval: false
    - action: "grep"
      requires_approval: false
  fallback:
    on_uncertainty: "ask_user"
    on_error: "stop_and_report"

⚠️Warning

Higher autonomy levels introduce risk. Always start at level 2 (semi-autonomous) when deploying new agents in production environments. Gradually increase autonomy as you validate the agent's behavior through rigorous testing.

Real-World Use Cases

AI agents excel in scenarios that require multi-step reasoning, tool interaction, and adaptive planning:

Code Generation and Refactoring

Agents can generate boilerplate code, refactor existing codebases, and maintain consistency across files.

python

# Agent-driven code refactoring workflow
def refactor_function(source_code: str, target_style: str) -> str:
    """Agent analyzes code, plans refactoring, executes changes."""
    analysis = {
        "issues": [
            {"line": 15, "type": "long_function", "suggestion": "Split into helpers"},
            {"line": 42, "type": "magic_number", "suggestion": "Use constant"},
            {"line": 78, "type": "duplicate_code", "suggestion": "Extract method"}
        ],
        "complexity": "moderate",
        "estimated_effort": "15 minutes"
    }
    # Agent would iteratively apply changes
    refactored = source_code  # Transformation happens here
    return refactored

Automated Testing

Agents create test suites, generate edge cases, and validate behavior.

yaml

# Agent test generation plan
test_plan:
  source_file: "src/payment/processor.py"
  generated_tests:
    - file: "tests/test_payment_processor.py"
      test_cases:
        - name: "test_successful_payment"
          input: {"amount": 100, "currency": "USD"}
          expected: {"status": "success", "transaction_id": str}
        - name: "test_insufficient_funds"
          input: {"amount": 99999, "currency": "USD"}
          expected: {"status": "failed", "reason": "insufficient_funds"}
        - name: "test_invalid_currency"
          input: {"amount": 50, "currency": "XYZ"}
          expected: {"status": "error", "message": "Unsupported currency"}
      coverage_target: 85

Documentation and Knowledge Management

Agents maintain documentation, answer questions about codebases, and generate API references.

DevOps and Infrastructure

Agents manage deployments, monitor systems, and respond to incidents.

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

Agent vs Traditional LLM: Comparison

Aspect	Traditional LLM	AI Agent
Interaction model	Request-response	Goal-oriented loop
State	Stateless (per prompt)	Stateful (across steps)
Tool use	None (text only)	Dynamic tool invocation
Planning	None	Multi-step decomposition
Memory	Limited to context window	Structured memory systems
Error recovery	None	Retry, fallback, adaptation
Autonomy	None	Configurable autonomy levels
Observability	Prompt/response logs	Full action traces

✅Success

AI agents represent a paradigm shift from passive text generation to active task execution. By combining LLM reasoning with tool use, memory, and planning, agents can autonomously complete complex software engineering workflows that previously required human intervention.

Agent Communication Patterns

Agents communicate with users, tools, and other agents through structured patterns:

💡 Tip: Use Ctrl + Mouse Wheel to zoom, or click the fullscreen icon

💡Tip

| Design agents to communicate their reasoning process, not just results. This transparency builds trust and makes debugging easier when agents make mistakes.

Practice Exercises

Practice Question

What is the key difference between a traditional LLM interaction and an AI agent?

Practice Question

At which autonomy level does an agent execute tools freely within defined guardrails without requiring human approval for routine actions?

Practice Question

A development team wants an agent that can read files, search code, and suggest changes, but requires human approval before writing any file. Which autonomy level should they configure?

Practice Question

In the context of agent memory, what is the primary purpose of the summarization mechanism?

Practice Question

Which capability allows an AI agent to interact with external systems like databases, file systems, and web APIs?

Practice Question

When an agent encounters an unexpected error during a multi-step plan, what should it do according to best practices?

Practice Question

What information should an agent store in long-term memory vs short-term memory?

Practice Question

A user asks an agent to 'improve the performance of the data processing pipeline.' The agent reads the code, identifies a slow algorithm, replaces it, tests it, and reports the improvement. This is an example of which agent characteristic?

✅**Key Takeaways**

AI agents are autonomous systems that perceive, reason, act, and learn in a continuous loop
Core agent capabilities include tool use, multi-step reasoning, and structured memory management
Autonomy levels range from 0 (no autonomy) to 4 (fully autonomous), with most production agents operating at level 2 or 3
Tool registries allow agents to interact with files, commands, APIs, and databases dynamically
Memory systems combine short-term context with long-term persistent storage
Agents excel at complex workflows: code generation, testing, documentation, and DevOps
The perceive-reason-act loop fundamentally distinguishes agents from traditional chat-based LLMs
Start with lower autonomy levels and increase gradually as agent behavior is validated

Progreso10%

Agent Architecture

Contenido del Curso

1. Introduction to AI Agents 2. Agent Architecture 3. Skills & Capabilities 4. Context Management 5. Agent Pipelines 6. OpenCode as Development Tool 7. Building Custom Agent Skills 8. Prompt Engineering for Agents 9. Best Practices & Patterns 10. Real-World Agent Projects