Harvey

Describe the typical agent execution loop.

When is a fixed LCEL chain preferable to an agent loop, and what is the decisive question to ask?

You change one word in the system prompt and prompt caching breaks. Why?

List SWE-bench, AgentBench, GAIA, and TAU-bench: what they measure, their shared blind spots, and how to interpret agent benchmark numbers.

What LangGraph feature allows the agent loop to cycle back to an earlier node rather than terminating?

Which practices reflect the principle of least privilege applied to an agent's tool set?

Compare CrewAI, AutoGen, and LangGraph coordination models and their reliability tradeoffs

Define activation quantization and explain how W8A8 differs from weight only quant.

Why might F1 score be a better choice than raw accuracy for evaluating entity extraction?

How does least privilege tool scoping reduce security risk in an agent system?