Tier 1 Quiz — AI Agent Design Patterns

01When should you reach for ReAct instead of a one-shot prompt?

When the next step depends on what the last tool returnedWhen you already know every step before any tool runsWhen you need three specialists with separate contextsWhen a human must approve before any action

02You need research, drafting and critique, and one model keeps dropping context. What does the decision framework suggest first?

Stay single-agent and raise the temperatureMove to a multi-agent shape (e.g. orchestrator + specialists)Delete the critique stepReplace tools with a longer system prompt only

03A ReAct research agent keeps calling search forever. What is the sharpest fix?

Remove all toolsAdd maxSteps plus an explicit stop condition (e.g. enough citations)Ask the model to “try harder”Switch to Debate between two search agents

04Scenario: book a flight only after a human confirms the itinerary. Which family fits?

Foundations only — pure ReAct with more toolsOrchestration — fan-out every airline in parallel and auto-buyFeedback loops — Human-in-the-loop before the irreversible actionSkip agents; use a CSV