상위: 00_00_MOC
00_01 — 용어 탄생과 계보
핵심 명제
“Context Engineering”은 논문에서 나온 말이 아니다. LLM을 프로덕션에 올려 쓰던 실무자들이 매일 겪던 고통에 이름이 붙은 순간이 출발점이었다. 2025년 6월 셋째 주, 엿새 간격으로 올라온 트윗 두 개가 그 기폭제가 됐다.
1. Tobi Lütke의 기폭제 (2025-06-19)
Shopify CEO Tobi Lütke가 X(트위터)에 다음 트윗을 올렸다. 조회 190만 회, 북마크 8,600을 기록했다.
“I really like the term ‘context engineering’ over prompt engineering. It describes the core skill better: the art of providing all the context for the task to be plausibly solvable by the LLM.” — Tobi Lütke (@tobi, 2025-06-19)
핵심 키워드는 “plausibly solvable(그럴듯하게 풀 수 있게)“이다. 모델에게 정답을 강요하는 대신, 풀어낼 만큼의 맥락을 공급하는 것이 엔지니어의 일이라는 프레임이다.
2. Andrej Karpathy의 확증과 확장 (2025-06-25)
전 OpenAI 연구자이자 AI 교육자인 Andrej Karpathy가 엿새 뒤 이에 동의하며 정의를 한층 다듬었다.
“+1 for context engineering over prompt engineering. People associate prompts with short task descriptions you’d give an LLM in your day-to-day use. When in every industrial-strength LLM app, context engineering is the delicate art and science of filling the context window with just the right information for the next step.” — Andrej Karpathy (@karpathy, 2025-06-25)
Karpathy는 여기에 균형 잡힌 경고를 덧붙였다.
“Too little or of the wrong form and the LLM doesn’t have the right context for optimal performance. Too much or of the wrong / irrelevant and the LLM costs might go up and performance might come down.”
부족해도 안 되고 과해도 안 된다는 뜻이다. “delicate(섬세한)“라는 표현이 컨텍스트 엔지니어링의 본질, 곧 최적화 문제를 정확히 짚는다.
Karpathy는 LLM을 CPU에, 컨텍스트 윈도우를 RAM에 빗대기도 했다. 이 비유는 이후 Lance Martin(LangChain)을 비롯해 여러 곳에서 널리 인용된다.
3. Simon Willison의 해설 — 왜 이 용어가 살아남았나 (2025-06-27)
Simon Willison은 이 용어가 지닌 지속력(sticking power)을 분석했다.
‘prompt engineering’은 많은 사람이 “그냥 챗봇에 잘 말 거는 법”으로 깎아내리며 허세 섞인 용어로 치부했다. 반면 ‘context engineering’은 단어에서 추론되는 의미(inferrable definition)가 실제 의도와 정확히 맞아떨어진다.
용어가 지식 공동체에서 살아남으려면 바로 이 조건이 가장 강력하게 작동한다. Willison은 이 언어적 전환을 “운 좋은 수정(fortunate correction)“이라고 불렀다.
4. 개념의 역사적 계보
용어는 신조어처럼 보여도 실천 자체는 새롭지 않다.
학술 논문 Context Engineering 2.0 (Qishuo Hua et al., arXiv:2510.26493, 2025-10-30)은 관련 실천의 뿌리를 1990년대 초 HCI(Human-Computer Interaction) 프레임워크까지 거슬러 짚으며, 그것이 오늘날의 HAI(Human-Agent Interaction) 패러다임으로 단계적으로 진화해왔다고 정리한다.
“Although context engineering is often regarded as a recent innovation of the agent era, related practices can be traced back to over 20 years ago.” — Qishuo Hua et al., 2025
LlamaIndex(Tuana Çelik, Logan Markewich, 2025-07-03)도 같은 결을 짚는다.
“The principles behind the term ‘context engineering’ are not new — [but] this wording is an effective abstraction that helps reason about the most pressing challenges.”
timeline title 컨텍스트 엔지니어링 개념의 역사적 진화 1990s : HCI 프레임워크 기반 컨텍스트 개념 2017 : Transformer 아키텍처 등장 (Vaswani et al.) 2020 : GPT-3 프롬프트 엔지니어링 부상 2022 : InstructGPT / RLHF — 시스템 프롬프트 최적화 2023 : RAG 확산 / Lost in the Middle (Liu et al.) 2024 : 에이전트·툴루프 / 1M 토큰 윈도우 Jun-2025 : Lütke(06-19) → Breunig(06-22) → Karpathy(06-25) → Willison(06-27) Jul-Sep-2025 : Chroma Context Rot / LangChain / Anthropic 공식 블로그 Oct-2025 : Context Engineering 2.0 논문 Apr-2026 : ContextBudget RL 프레임워크 (arXiv:2604.01664)
5. 2025년 6월 셋째 주가 의미하는 것
| 날짜 | 사건 |
|---|---|
| 2025-06-19 | Tobi Lütke 트윗 (1.9M 뷰) |
| 2025-06-22 | Drew Breunig — 4대 실패모드 정의 |
| 2025-06-25 | Andrej Karpathy — “delicate art and science” |
| 2025-06-27 | Simon Willison — 용어 지속력 분석 |
이 엿새는 컨텍스트 엔지니어링이라는 분과(discipline)가 명명된 주간으로 기록된다. 한 달 안에 Anthropic, LangChain, LlamaIndex 등 주요 기관이 잇따라 구현 가이드를 내놓았다.
참고문헌
- Tobi Lütke on X — “the art of providing all the context” — Tobi Lütke (Shopify CEO), 2025-06-19, X
- Andrej Karpathy on X — “delicate art and science of filling the context window” — Andrej Karpathy, 2025-06-25, X
- Context engineering — Simon Willison, 2025-06-27, 개인 블로그
- Context Engineering 2.0: The Context of Context Engineering — Qishuo Hua et al., 2025-10-30, arXiv:2510.26493
- Context Engineering — What it is, and techniques to consider — Tuana Çelik, Logan Markewich, 2025-07-03, LlamaIndex Blog
관련 노트: 00_02-정의-수렴과-프롬프트-비교 · 00_06-5대-실패모드-분류 · 00_핵심용어집