quiz · 4 min Planning & Reasoning Check What does decomposition mean in the context of an agent?Generating the list of subgoals before executing any of themRunning the agent loop faster by skipping tool callsBreaking the model prompt into shorter tokensReplacing the policy with a lookup tableWhy should the planner and executor be kept as separate roles?So they can each use a different programming languageTo reduce the total number of model calls to oneSo a bad plan is inspectable and replaceable without touching the executorBecause the executor cannot read plain textAccording to Lesson 22, when is adding a planning layer justified?Whenever a task involves any tool call at allAlways, because planning never adds latencyOnly when the model has fewer than 8k context tokens availableOnly when the task reliably needs multiple dependent steps and measurement shows the overhead pays offWhat is 'bounded reflection' and why does it matter?Limiting reflection to tasks under a fixed token countUsing a separate model for critique so the generator is not involvedSetting a hard iteration cap so the loop exits rather than burning tokens indefinitelyRunning reflection only on the final output, never on intermediate stepsWhen does reflection actively fail to improve output quality?When the output is longer than one paragraphWhen the critic and generator share the same blind spot, causing both to miss the same constraintWhenever more than two reflection passes are attemptedWhen the task has fewer than three subgoals in the planSubmit Mark complete ← Reflection & Its Limits When Multiple Agents? →