Agent Tool Use

Tool calling은 model이 tool과 arguments를 제안하는 interface일 뿐이다. Tool use는 그보다 넓어서 selection, authorization, execution, observation validation, retry, stopping까지 포함한다. ReAct는 inference 중 observation을 reasoning에 되먹이는 실행 패턴이고, Toolformer는 API call example을 self-supervised하게 학습하는 training method다 — 둘은 같은 layer에 있지 않다.

실행 흐름: 제안과 실행은 분리된다

sequenceDiagram
    participant O as Orchestrator
    participant M as Model
    participant X as Executor
    O->>M: Context and tool schema
    M-->>O: Tool and arguments
    O->>O: Validate and authorize
    O->>X: Execute
    X-->>O: Raw result
    O->>O: Validate and record
    O->>M: Observation

Whitepaper의 도구 분류

유형 역할 control
Extensions API와 agent의 standardized bridge agent-side infrastructure
Functions model이 function과 arguments 제안 client가 실제 실행
Data Stores indexed data retrieval context observation 반환

ReAct와 Toolformer는 같은 층위가 아니다

기준 ReAct Toolformer
시점 inference prompting training data와 fine-tuning
단위 Thought, Action, Observation API call과 result
feedback environment observation future-token loss filtering
아닌 것 architecture 전체 runtime planner

Production에서 신뢰성을 좌우하는 것

이 네 가지는 위 실행 흐름과 도구 분류로부터 이 페이지가 도출한 운영 원칙이며, 세 source가 직접 명시한 production checklist는 아니다.

흔히 겪는 실패

테스트 질문

출처


  1. 22365_19_Agents_v8.pdf (Figure 8, Figure 9) — 이 PDF는 이번 변환 세션에서 poppler 부재로 재추출하지 못했고, 2026-07-11 conformance 기록 시점에 검증된 기존 claim을 그대로 유지한다. ↩︎ ↩︎

  2. ReAct_2210.03629.pdf — 이 PDF는 이번 변환 세션에서 poppler 부재로 재추출하지 못했고, 2026-07-11 conformance 기록 시점에 검증된 기존 claim을 그대로 유지한다. 상세 mechanism은 ReAct 참고. ↩︎

  3. Toolformer_2302.04761.pdf — 이 PDF는 이번 변환 세션에서 poppler 부재로 재추출하지 못했고, 2026-07-11 conformance 기록 시점에 검증된 기존 claim을 그대로 유지한다. 상세 mechanism은 Toolformer 참고. ↩︎