QnA

workflow-patterns

Q&A 정리: workflow-patterns

AI 에이전트의 sequential workflow란 무엇인가?

작업을 정해진 순서대로 하나씩 처리하는 방식이다. 공장 조립 라인처럼 1번 담당자가 끝내면 결과를 2번에게 넘기고, 2번이 끝나면 3번에게 넘기는 식으로 진행된다.

Sequential workflows execute tasks in a predetermined order. Agents at each stage process inputs, make decisions, make tool calls as needed, then pass results to the next stage. The result is a clear chain of operations where outputs flow linearly through the system.


AI 에이전트 sequential workflow의 트레이드오프는 무엇인가?

순서대로 처리하므로 전체 소요 시간은 늘어나지만, 각 단계의 담당자가 자기 역할에만 집중하기 때문에 정확도가 높아진다. 속도를 조금 희생하고 품질을 얻는 방식이다.

You're trading some latency for higher accuracy by focusing each agent on a specific subtask instead of trying to handle everything at once.


AI 에이전트의 parallel workflow란 무엇인가?

서로 영향을 주지 않는 독립적인 작업들을 여러 에이전트가 동시에 처리하는 방식이다. 시험에서 한 명이 모든 문제를 푸는 대신, 여러 명이 각자 다른 문제를 동시에 풀고 답안지를 합치는 것과 비슷하다.

Parallel workflows distribute independent tasks across multiple agents that execute simultaneously. Instead of waiting for one agent to finish before starting the next, you run multiple agents at once and merge their results. Agents don't hand off work to each other—they operate autonomously and produce results that contribute to the overall task.


AI 에이전트 parallel workflow의 트레이드오프는 무엇인가?

여러 에이전트를 동시에 돌리므로 비용이 더 들고, 각자의 결과를 하나로 합치는 전략이 필요하다. 대신 전체 완료 시간이 빨라지고, 복잡한 작업을 하나의 호출로 처리하는 것보다 품질이 높은 경우가 많다.

Costs more (multiple concurrent API calls) and requires an aggregation strategy. Can lead to faster completion and separation of concerns across engineering teams. For complex tasks, handling each consideration with a separate AI call often outperforms trying to juggle everything in one call.


AI 에이전트 parallel workflow 도입 시 주의할 점은?

결과를 합치는 방법(다수결, 점수 평균, 전문가 우선 등)을 먼저 정해야 한다. 또한 작업 간에 이전 결과가 필요하거나 서로 쌓아가야 하는 경우에는 병렬 처리가 적합하지 않다.

Design your aggregation strategy before implementing parallel agents. Will you take the majority vote? Average confidence scores? Defer to the most specialized agent? Having a clear plan for synthesizing results prevents you from collecting conflicting outputs with no way to resolve them. Don't use parallel workflows when agents need cumulative context or must build on each other's work. Skip this pattern when resource constraints like API quotas make concurrent processing inefficient, or when you lack clear strategies for handling contradictory results from different agents.


AI 에이전트의 evaluator-optimizer workflow란 무엇인가?

하나는 결과물을 만들고, 다른 하나는 그 결과물을 평가하여 피드백을 주는 두 에이전트가 짝을 이뤄 반복하는 방식이다. 작가가 글을 쓰고 편집자가 교정하면 작가가 다시 고치는 과정을 품질 기준에 도달할 때까지 되풀이하는 것과 같다.

Evaluator-optimizer workflows pair two agents in an iterative cycle: one generates content, another evaluates it against specific criteria, and the generator refines based on that feedback. This continues until the output meets your quality threshold or hits a maximum iteration count.


Evaluator-optimizer에서 생성과 평가를 왜 분리하는가?

만드는 것과 평가하는 것은 서로 다른 종류의 사고 작업이다. 역할을 분리하면 생성 에이전트는 콘텐츠 제작에, 평가 에이전트는 품질 기준 적용에 각각 집중할 수 있어 전체 결과물의 수준이 올라간다.

The key insight is that generation and evaluation are different cognitive tasks. Separating them lets each agent specialize—the generator focuses on producing content, the evaluator focuses on applying consistent quality criteria.


Evaluator-optimizer workflow는 어떤 조건에서 도입해야 하는가?

AI가 일관되게 적용할 수 있는 명확하고 측정 가능한 품질 기준이 있고, 첫 번째 시도와 최종 결과 사이의 품질 차이가 추가 비용을 정당화할 만큼 클 때 적합하다.

This pattern works when you have clear, measurable quality criteria that an AI evaluator can apply consistently, and when the gap between first-attempt and final quality is meaningful enough to justify the extra tokens and latency.


Evaluator-optimizer workflow의 대표적인 유스케이스는?

보안·성능 기준을 충족해야 하는 코드 생성, 톤과 정확성이 중요한 고객 커뮤니케이션 작성, API 문서 자동 생성, SQL 쿼리 작성 등 첫 초안의 품질이 요구 수준에 미치지 못하는 경우에 효과적이다.

Consider evaluator-optimizer workflows for:

  • Code generation with specific requirements (security standards, performance benchmarks, style guidelines)
  • Professional communications where tone and precision matter
  • Any scenario where first-draft quality consistently falls short of requirements

Evaluator-optimizer workflows work well for:

  • Generating API documentation (generator writes docs, evaluator checks for completeness, clarity, and accuracy against the codebase)
  • Creating customer communications (generator drafts email, evaluator assesses tone and policy compliance)
  • Producing SQL queries (generator writes query, evaluator checks for efficiency and security issues)

AI 에이전트 evaluator-optimizer workflow의 트레이드오프는 무엇인가?

첫 시도만으로 충분한 품질이 나온다면 반복은 비용 낭비다. 즉각 응답이 필요한 실시간 서비스, 단순 분류 작업, 평가 기준이 주관적인 경우에도 적합하지 않다. 코드 스타일 검사처럼 자동화 도구가 있다면 그쪽이 더 효율적이다.

Skip evaluator-optimizer workflows when first-attempt quality already meets your needs—you're burning tokens on unnecessary iterations. Don't use this pattern for real-time applications requiring immediate responses, simple routine tasks like basic classification, or when evaluation criteria are too subjective for an AI evaluator to apply consistently. If deterministic tools exist (like linters for code style), use those instead. Also avoid this pattern when resource constraints outweigh quality improvements.


AI 에이전트 evaluator-optimizer workflow 도입 시 주의할 점은?

반복을 시작하기 전에 최대 반복 횟수와 품질 기준선을 명확히 정해야 한다. 그렇지 않으면 평가 에이전트가 사소한 문제를 계속 지적하고 생성 에이전트가 끝없이 수정하면서, 품질은 이미 정체되었는데 비용만 쌓이는 악순환에 빠질 수 있다.

Set clear stopping criteria before you start iterating. Define maximum iteration counts and specific quality thresholds. Without these guardrails, you can end up in expensive loops where the evaluator keeps finding minor issues and the generator keeps tweaking, but quality plateaus well before you stop iterating. Know when good enough is good enough.