From Language Models to Agents

Lecture 2.3

• What LLMs can't do: no current info, no world interaction, unreliable math, no self-verification, no memory
• Tool calling: the protocol that bridges text prediction to real-world action
• The 5-step flow: describe tools → model requests → code executes → result returns → model continues
• System prompts: shaping agent identity, overriding training biases
• The complete picture: LLM + context + tools + system prompt + agent loop

Additional Resources

• Lecture slides