A legal-tech company is building an agent that drafts contract amendments. The task requires: (1) reviewing the original contract, (2) identifying clauses that need changes, (3) researching applicable regulations, (4) drafting amendments, and (5) verifying compliance. The team notices that when using a simple agent loop, the model often jumps to drafting before fully reviewing the contract, leading to incomplete amendments. They consider Plan-Execute instead. What fundamentally distinguishes Plan-Execute from simple agent loops?
Planning & Reasoning
Loading lesson content...
An agent creates a plan and executes it perfectly, but the result doesn't satisfy the user's actual need. What went wrong?
A Plan-Execute agent creates very detailed plans that take many tokens to generate and execute. For simple tasks, this overhead isn't justified. What architecture handles this?