What are agents?

Building Agentic Workflows with LlamaIndex

Laurie Voss

VP Developer Relations, LlamaIndex

LLMs have limitations

$$

  • LLMs are brilliant but have limitations

$$

  • Answer with text responses from training

$$

  • Can't fetch info or act unless we provide new data

single_llm.gif

Building Agentic Workflows with LlamaIndex

Agents: the bigger picture

$$

  • Leverage an AI model to interact

  • It can:

    • Reason
    • Plan
    • Execute actions via tools

$$

agent_capabilities.png

Building Agentic Workflows with LlamaIndex

Agents: the bigger picture

  • Scenario:
    • AI that can choose web search or step-by-step calculations
    • Handles complex or open-ended tasks
    • Breaks tasks into steps and uses tools

agents_choice.png

Building Agentic Workflows with LlamaIndex

LLMs vs. Agents

$$

  • LLM:
    • If the task is simple

"Summarize this text..."

  • Agent:
    • When task requires multi-step, external info, or conditional decisioning
    • When we want to add a tool

"Research this topic..."

agent_vs_llm.png

Building Agentic Workflows with LlamaIndex

Agentic workflows

Focus of the course:

  • Agentic workflows
  • From single agent + tool to deep research multi-agent systems

agent_to_agentic_workflow.gif

Building Agentic Workflows with LlamaIndex

Let's practice!

Building Agentic Workflows with LlamaIndex

Preparing Video For Download...