Claude Certified Architect – Foundations Practice Question 87: Programmatic Tool Calling
Claude Certified Architect, Foundations · Test 9 · question 7 of 10
An expense-audit agent must check 20 employees against budget limits. Each lookup tool call returns thousands of expense line items, and today every result passes through the model: 20 round trips, a bloated context window, and high latency. Only the few employees who exceeded their limits matter for the final answer. Which design addresses this most directly?
Show answer and explanation
Answer:
D is correct. Programmatic tool calling lets Claude write code that calls your tools inside a code execution container, instead of a model round trip per call. The script can loop over the 20 lookups, filter or aggregate, and return only the relevant lines; intermediate results are not loaded into Claude's context window. A tool opts in through allowed_callers (for example code_execution_20260120) alongside the code execution tool.
A is wrong because tool search reduces the cost of large tool definitions, not of large tool results. B is wrong because caching lowers the price of re-reading a prefix, but the thousands of line items still enter context and the round trips remain. C is wrong because context editing clears results after they have already consumed context once, and does not remove the 20 sequential round trips.
Reference: Programmatic tool calling - Claude API Docs