A 311 trash-day agent repeatedly fails on holiday weeks. Based on that feedback, how should it adapt its reasoning strategy?
Select an answer to reveal the explanation.
Short Explanation
After you burn dinner three nights in a row with the wrong oven setting, you check the recipe first. Same idea: repeated holiday trash misses mean the agent should hit the holiday calendar tool earlier next time.
Full Explanation
Experience-based adaptation updates tool preference from failure feedback. A 311 trash-day agent that repeatedly fails on holiday weeks should learn to prefer the holiday calendar tool earlier when answering collection-day questions, because holiday schedules are the recurring root cause of wrong-day answers in those weeks.
Ignoring holiday calendars permanently to reduce tool-call volume encodes the failure mode into policy. Answering from training-cutoff knowledge only and stopping city tool calls freezes outdated assumptions and drifts from published schedules. Escalating every trash question to a human before any tool use abandons automation without learning a better tool order.
Exam caveat: adaptation should change strategy (tool order, prompts, policies), not silently invent collection days. Operational check: after injecting holiday-week failures in evaluation, confirm the updated policy calls the holiday calendar before generic route tables, and that accuracy on holiday fixtures recovers without universal human escalation. Capture failure traces tagged by holiday week so the policy update is evidence-based, not anecdotal. Store the updated tool-preference policy in version control next to the 311 agent config so holiday-week regressions are reviewable in pull requests.