Skip to content
Home › Architect's Playbook Patterns

Forcing Execution Order

Architect's Playbook Patterns

Definition

Using the API's tool_choice parameter to force a specific tool on the first call, guaranteeing a deterministic pipeline order (e.g. metadata extraction before enrichment) rather than relying on prompt wording.

Key points

  • The problem: An agent needs to extract metadata before calling enrichment tools, but occasionally calls enrichment tools first, causing failures.
  • API config shown:
    • model: "claude-3-opus-20240229"
    • max_tokens: 1024
    • messages: [{role:"user", content:"Extract metadata from this paper and then look up its DOI."}]
    • tools: [{name:"extract_metadata", ...}, {name:"lookup_citations", ...}]
    • tool_choice: {"type": "tool", "name": "extract_metadata"}
  • The enforcement: Do not rely on prompt begging. Set tool_choice for the first API call to guarantee the pipeline executes in the required order — structured metadata extraction happens before any DOI lookup or enrichment.

Why it matters for the exam

  • This is the Multi-Agent × Compliance/Control cell in the Reference Matrix ("tool_choice Enforcement").
  • Tests knowing that tool_choice: {"type":"tool","name":...} forces a specific tool, and that this is the deterministic alternative to prompt instructions.

Common gotchas

  • "Prompt begging" (telling the model the order in the message) is the anti-pattern — it is unreliable.
  • tool_choice is set on the first call to seed the required first step; the rest of the pipeline follows from there.

See also

Sources

Referenced by