Key ideas
An agent is a goal-directed tool loop
The talk adopts a minimal definition: an LLM agent repeatedly invokes tools until it reaches a g…Show moreShow less
The talk adopts a minimal definition: an LLM agent repeatedly invokes tools until it reaches a goal. This definition separates the agent's purpose from any particular framework or implementation language. Across all three implementations, the underlying behavior remains the same even as responsibility for the loop moves from user code to a framework and then to a hosted runtime.
Why it matters: This framing makes it easier to compare agent architectures by asking who owns the loop, tools, state, and execution environment.
Supporting evidence
An LLM agent runs tools uh in a loop until it achieves a goal.
Step-based histories fit agents better than chat turns
Traditional chat APIs model interaction as alternating user and model messages. Agent execution…Show moreShow less