CCA Prep
CCAR-FUA

Lesson 2.3Lesson 10 of 30

Distributing tools and tool_choice

Distribute tools appropriately across agents and configure tool choice

Distributing tools and tool_choiceCoordinatorroutes complex casesSearch agentweb_search ·load_documentAnalysis agentread_document ·extract_dataSynthesis agentverify_fact (narrow)Report agentformat_citations4–5 tools per role · tool_choice auto | any | {type: tool, name}

Key points

  1. 1An agent with 18 tools picks worse than one with 4–5. Give each subagent only the tools its role needs.
  2. 2Tools outside an agent's specialisation get misused. A synthesis agent with web search starts searching instead of synthesising.
  3. 3Replace generic tools with constrained ones (fetch_url → load_document that validates the URL). For a frequent cross-role need, give a narrow tool like verify_fact and route complex cases through the coordinator.
  4. 4tool_choice: auto lets the model answer in text; any guarantees some tool is called; {type: tool, name} forces one specific tool, e.g. extract_metadata first, then continue in later turns.
What the exam guide checks here

Objective wording from the exam guide (English).

  • 2.3.K1Too many tools (e.g. 18 vs 4–5) degrades selection reliability
  • 2.3.K2Agents with tools outside their specialisation misuse them
  • 2.3.K3Scoped tool access with limited cross-role tools for high-frequency needs
  • 2.3.K4tool_choice options: auto, any, forced {type: tool, name}
  • 2.3.S1Restrict each subagent's tools to its role
  • 2.3.S2Replace generic tools with constrained ones (fetch_url → load_document that validates URLs)
  • 2.3.S3Give scoped cross-role tools for the common case (verify_fact) and route complex cases via coordinator
  • 2.3.S4Force a specific tool first (e.g. extract_metadata), continue in follow-up turns
  • 2.3.S5tool_choice any guarantees a tool call instead of text

Lock it in

Answer scenario questions on this topic. Every option comes with an explanation.

Practise 2.3 · 23 questions