PRODUCTION SCENARIO
A logistics client asks for an autonomous multi-tool agent to classify 5,000 inbound shipping exceptions a day into 12 stable categories. Each exception fits in one message and the categories have not changed in two years.
Which approach should the delivery team ship first?
Answering here is anonymous. Nothing is saved unless you sign in.
Show answer and explanation
Answer: A single LLM call with retrieval and worked examples, evaluated on a labeled set
Anthropic's guidance is to start with simple prompts, optimize them with comprehensive evaluation, and add multi-step agentic systems only when simpler solutions fall short. A classification task with stable categories and single-message inputs is where one optimized call with retrieval and in-context examples is usually enough, and an agent trades latency and cost for flexibility this workload never uses.
Official Source: https://www.anthropic.com/engineering/building-effective-agents ↗