{"success":true,"count":3,"items":[{"videoId":"9DHZLw5653E","chunkIndex":0,"totalChunks":3,"title":"How Spotify runs agents across 20M+ lines of code, with Niklas Gustavsson — Part 1 of 3","thumbnail":"https://i.ytimg.com/vi/9DHZLw5653E/maxresdefault.jpg","duration":1571,"uploader":"Claude","youtubeUrl":"https://www.youtube.com/watch?v=9DHZLw5653E","keywords":["artificial-intelligence","llm","coding-assistant","software-engineering","developer-tools","automation","monorepo","agentic-workflow","code-maintenance"],"normalizedKeywords":["엔지니어링","기술 트렌드","리더십·매니지먼트"],"targetAudience":[{"who":"엔지니어 리더","why":"대규모 코드베이스에서 AI 에이전트를 어떻게 실무에 넣는지 볼 수 있음"},{"who":"개발자","why":"터미널 기반 에이전트 워크플로와 코드 수정 방식의 변화를 배울 수 있음"},{"who":"기술 의사결정자","why":"AI 도입이 단순 데모가 아니라 유지보수 비용 구조를 바꾸는지 이해할 수 있음"}],"normalizedAudience":["엔지니어·개발자"],"summary":"이 영상은 Spotify가 LLM과 에이전트를 이용해 거대한 코드베이스의 유지보수와 코드 변경을 자동화해 온 과정을 다룬다. 화자는 초기에는 LLM으로 코드 변경을 시도했지만 잘 되지 않았고, 모델 성능 향상과 함께 judge, 분해(decomposition), 반복 실험을 거치며 실용적인 시스템으로 발전시켰다고 설명한다. 특히 20M+ 라인 규모의 monorepo에서도 Claude가 잘 작동했고, 단순한 코드 생성기를 넘어 배경 작업과 Slack 기반 업무까지 처리하는 내부 도구 Honk으로 확장됐다고 말한다.\n\n개인적으로는 Opus 4/5 시점에 “스마트 자동완성”에서 “실제 문제를 맡길 수 있는 도구”로 체감이 바뀌었다고 회고한다. 예전에는 모델이 작성한 코드의 마지막 수정을 IDE에서 직접 해야 했지만, 이제는 그 마지막 10~20%까지 AI가 처리해 작업 방식 자체가 달라졌다는 점이 핵심이다. 이 영상은 AI 코딩의 미래를 낙관적으로 말하는 수준을 넘어, 대규모 조직에서 유지보수와 코드 변환을 어떻게 에이전틱하게 운영하는지 보여주는 사례다.","insights":["LLM의 진짜 변화는 코드 생성보다 마지막 수정을 없앤 데 있다.","대규모 코드베이스일수록 AI가 맥락 활용 능력으로 강해진다.","코드 변환은 한 번에 푸는 문제가 아니라 분해와 검증의 문제다.","유지보수 자동화는 결국 에이전트 운영체계로 진화한다.","AI 도입의 가치는 개발 속도보다 조직의 유지보수 병목 해소에 있다."],"keyClips":[{"clipId":"9DHZLw5653E:c0:1-8","startSegmentIndex":1,"endSegmentIndex":8,"startTime":12.23,"endTime":50.559,"durationSeconds":38.3,"preview":"업무 방식의 급변","mustSee":false},{"clipId":"9DHZLw5653E:c0:10-13","startSegmentIndex":10,"endSegmentIndex":13,"startTime":57.28,"endTime":86.88,"durationSeconds":29.6,"preview":"바이오에서 코딩으로","mustSee":false},{"clipId":"9DHZLw5653E:c0:17-24","startSegmentIndex":17,"endSegmentIndex":24,"startTime":103.92,"endTime":187.68,"durationSeconds":83.8,"preview":"진짜 전환점은 여기","mustSee":true},{"clipId":"9DHZLw5653E:c0:26-35","startSegmentIndex":26,"endSegmentIndex":35,"startTime":192.08,"endTime":269.759,"durationSeconds":77.7,"preview":"터미널 중심 워크플로","mustSee":false},{"clipId":"9DHZLw5653E:c0:40-52","startSegmentIndex":40,"endSegmentIndex":52,"startTime":327.6,"endTime":423.84,"durationSeconds":96.2,"preview":"유지보수 자동화의 이유","mustSee":false},{"clipId":"9DHZLw5653E:c0:53-68","startSegmentIndex":53,"endSegmentIndex":68,"startTime":423.84,"endTime":535.76,"durationSeconds":111.9,"preview":"한계 돌파의 방법","mustSee":true},{"clipId":"9DHZLw5653E:c0:69-74","startSegmentIndex":69,"endSegmentIndex":74,"startTime":535.76,"endTime":585.12,"durationSeconds":49.4,"preview":"도구가 플랫폼이 되다","mustSee":false}],"curatedSegments":[{"segmentIndex":14,"text":"In the back of my head, I was always a bit worried like we were talking about before of like how this was change completely changing the way we were working and I was pretty worried about that from just my personal point of view like am I going to miss that part of like the hard mental challenge of solving problems and now I find myself having you know five agents working in the background and my way of interacting with them is very different from the way that I was working a year or two ago and for me that's um turned out that I was wrong and I like the thing that I like to do is solving problems and the way that I solve those problems turn out to not be the most critical piece for me.","startTime":1290.32,"endTime":1332.559,"durationSeconds":42,"level":"C1","overallScore":7.6,"rationale":"개인적 고민과 깨달음이 길게 전개됨."},{"segmentIndex":22,"text":"It went from being this like smart autocomplete to something that I could actually throw real problems at and I didn't have to do all that much prompt engineering.","startTime":155.12,"endTime":164.8,"durationSeconds":10,"level":"B2","overallScore":6.8,"rationale":"도구의 변화와 체감 효과가 분명함."},{"segmentIndex":23,"text":"The biggest thing for me was also just not having to edit code anymore cuz my workflow up to then was I have the model write you know like maybe 80% of the code or 70% of the code depending on the model and then I always had to go into an IDE to do the last mile edits >> and I just stopped having to do that.","startTime":164.8,"endTime":182.56,"durationSeconds":18,"level":"C1","overallScore":7,"rationale":"워크플로 변화가 구체적으로 드러남."},{"segmentIndex":28,"text":">> Y >> and to me it feels kind of like a false dichotomy because if you want to go faster, the thing that you need to do is you need to automate your quality practices so that it's better encoded. It's not in someone's head.","startTime":840.16,"endTime":848.079,"durationSeconds":8,"level":"C1","overallScore":6.8,"rationale":"추상적 주장과 이유 제시가 좋음."},{"segmentIndex":31,"text":">> Y >> and that's ultimately what lets you go faster. And this is just another example of how in engineering productivity is always about investing in infrastructure. It's not about working more hours.","startTime":857.8389999999999,"endTime":865.279,"durationSeconds":7,"level":"C1","overallScore":6.6,"rationale":"성과와 원리를 연결해 설명함."},{"segmentIndex":61,"text":"So basically like the planned work that we have and then that connects to uh AB tests and rollouts and then we're able to from that see like basically attribute back to say this PR contributed to this uh DoD that we have and that contributed to this user value.","startTime":1066.48,"endTime":1084.64,"durationSeconds":18,"level":"C1","overallScore":6.8,"rationale":"인과관계와 측정 연결을 설명함."},{"segmentIndex":4,"text":"So I think if there's one advice I would give would be to not ignore those types of investments.","startTime":1228.72,"endTime":1235.52,"durationSeconds":7,"level":"B2","overallScore":6.4,"rationale":"조언이 담긴 일반화 문장이다."},{"segmentIndex":21,"text":"So that was certainly one for I have to say for my own personal coding the real breakthrough moment was probably Opus 4 or 5 back in November December.","startTime":144.4,"endTime":155.12,"durationSeconds":11,"level":"B2","overallScore":5.2,"rationale":"개인적 전환점이지만 표현은 단순함."},{"segmentIndex":35,"text":"Do you feel like one like monor repo or poly repo is a better fit for quad or >> I was a bit worried to be honest about the monor repo setup and agents originally because um I think with some of the prior tools we've been using we've been seeing issues with indexing and things like that um and this these are fairly large repositories that our backend monor repo is more than 20 million lines of code but turns out it cla works amazingly well in those repositories and Um, I think one of the things we found is how good Claude is looking at other code in the repository to get, I guess, inspiration for the problem you're trying to solve.","startTime":258.639,"endTime":269.759,"durationSeconds":11,"level":"C1","overallScore":5.8,"rationale":"대규모 저장소에서의 성능 통찰."},{"segmentIndex":61,"text":"So that inspired us as I mentioned before as pretty much as soon as the early LLMs came along of like hey these things can we apply them to this problem and early on it didn't work at all that well.","startTime":479.28,"endTime":493.44,"durationSeconds":14,"level":"B2","overallScore":5.6,"rationale":"시도 실패와 이유를 설명해 통찰 있음."},{"segmentIndex":65,"text":"So we started applying LMS as judge to make sure that the output was as intended.","startTime":514.8,"endTime":520.479,"durationSeconds":6,"level":"B2","overallScore":5.8,"rationale":"판단자를 활용한 방법론 설명."},{"segmentIndex":71,"text":"But it was the first sort of light in the tunnel of like yeah this is actually a problem that we can solve.","startTime":545.839,"endTime":551.68,"durationSeconds":6,"level":"B1","overallScore":5.4,"rationale":"문제 해결 가능성을 깨달은 전환점."},{"segmentIndex":1,"text":"verification step like a agentic verifier. Tell me more about that. >> So we used to have a judge in honk but we actually have removed that because we found that the uh agent and models just again going back to four or five got good enough that we don't didn't need judge anymore.","startTime":600,"endTime":614.64,"durationSeconds":15,"level":"B2","overallScore":5.8,"rationale":"판단 변화와 이유를 설명함."},{"segmentIndex":4,"text":"So >> so it's a big change but then again as we talked about the models caught up and the agent hardness caught up so we have now eliminated that judge from honk.","startTime":626.64,"endTime":638.48,"durationSeconds":12,"level":"B2","overallScore":5.4,"rationale":"변화 원인을 다시 정리함."},{"segmentIndex":16,"text":">> I think that's very true and I think it's true for us as well. One of the major changes that we did in our engineering practices as part of that was to strengthen our test automation.","startTime":749.519,"endTime":758.24,"durationSeconds":9,"level":"B2","overallScore":5.6,"rationale":"테스트 자동화 강화를 말함."},{"segmentIndex":25,"text":"So that meant then having to build out much better test automation to make sure that uh all our software could sort of survive those types of automated changes.","startTime":810.72,"endTime":821.04,"durationSeconds":10,"level":"B2","overallScore":5.6,"rationale":"자동 변경을 위한 교훈이 있음."},{"segmentIndex":35,"text":"Um I think we we're going to have to continue our investments into uh our reliability practices as well. I think we want to be able to basically have an idea and for a developer to have an idea and be able to ship that into production as quickly as possible.","startTime":887.92,"endTime":894.72,"durationSeconds":7,"level":"B2","overallScore":5.6,"rationale":"속도와 신뢰성 목표를 제시함."},{"segmentIndex":46,"text":"But in both of those cases, the faster we can iterate, we found that we um we both build better products and we're able to ship them faster to our users.","startTime":973.12,"endTime":982.32,"durationSeconds":9,"level":"B2","overallScore":5.4,"rationale":"빠른 반복의 효과를 말함."},{"segmentIndex":66,"text":"Yet, you know, as engineers, we still want to measure it. I'm going to say like the ROI discussion initially was fairly easy because we could see such large improvements and um but as the maturity is getting there and the costs have been improving.","startTime":1109.919,"endTime":1113.36,"durationSeconds":3,"level":"C1","overallScore":5.8,"rationale":"ROI 논의를 설명하며 문장도 복합적."},{"segmentIndex":71,"text":">> And now you know people are seeing these like many dozens or hundreds of percentage points of improvement and now you really want to attribute it to figure out like how many tokens did it take?","startTime":1146.96,"endTime":1155.36,"durationSeconds":8,"level":"C1","overallScore":5.6,"rationale":"성과를 수치로 귀속하는 질문."}],"generatedAt":"2026-06-30T13:58:05.510Z","keyClipsTotalSec":639},{"videoId":"9DHZLw5653E","chunkIndex":1,"totalChunks":3,"title":"How Spotify runs agents across 20M+ lines of code, with Niklas Gustavsson — Part 2 of 3","thumbnail":"https://i.ytimg.com/vi/9DHZLw5653E/maxresdefault.jpg","duration":1571,"uploader":"Claude","youtubeUrl":"https://www.youtube.com/watch?v=9DHZLw5653E","keywords":["ai-agents","software-engineering","devops","test-automation","code-review","ci-cd","developer-productivity","engineering-management","large-scale-systems"],"normalizedKeywords":["엔지니어링","리더십·매니지먼트","기술 트렌드"],"targetAudience":[{"who":"엔지니어링 리더","why":"AI 에이전트를 조직에 도입할 때 검증·품질·측정 체계를 어떻게 바꿔야 하는지 배울 수 있음"},{"who":"플랫폼/인프라 엔지니어","why":"대규모 코드베이스에서 CI, 테스트 자동화, 배포 파이프라인을 어떻게 설계하는지 힌트를 얻을 수 있음"},{"who":"개발자 생산성 담당자","why":"PR 빈도, 자동화 비율, ROI를 정량화하는 프레임을 참고할 수 있음"}],"normalizedAudience":["엔지니어·개발자","디자인 리더·CXO","지식노동자 일반"],"summary":"이 영상은 Spotify가 대규모 코드베이스에서 AI 에이전트를 실전 투입하면서, 가장 중요한 기반이 '모델 성능'이 아니라 검증과 테스트 자동화라는 점을 강조한다. 초반에는 사람 대신 판정하는 judge가 필요했지만, 모델과 에이전트가 충분히 좋아지면서 그 역할을 제거했고, 대신 CI 빌드·시뮬레이터·Mac/Linux 검증 루프를 더 강하게 만드는 쪽으로 진화했다.\n\n또한 빠른 배포와 높은 품질은 반대가 아니라, 품질을 코드와 인프라에 잘 인코딩할수록 동시에 개선될 수 있는 요소라고 주장한다. Spotify는 수천 개의 컴포넌트, 명확한 ownership, 강화된 테스트 자동화 덕분에 AI가 만들어내는 자동 PR과 자동 머지의 속도를 받아낼 수 있었고, PR 빈도와 AI 작성 비율 같은 내부 지표뿐 아니라 최종적으로는 사용자 가치와 매출까지 연결된 ROI 측정 체계를 만들고 있다.","insights":["AI 도입의 병목은 모델이 아니라 검증 루프다.","속도와 품질은 충돌이 아니라 자동화 수준의 함수다.","사람의 판단을 시스템에 박아 넣어야 대규모 자동화가 가능하다.","ROI는 생산성 지표에서 끝나지 않고 사용자 가치로 이어져야 한다.","배포를 빨리 하려면 더 강한 테스트와 신뢰성 투자가 필요하다."],"keyClips":[{"clipId":"9DHZLw5653E:c1:1-4","startSegmentIndex":1,"endSegmentIndex":4,"startTime":600,"endTime":638.48,"durationSeconds":38.5,"preview":"판정자 제거의 전환","mustSee":false},{"clipId":"9DHZLw5653E:c1:5-12","startSegmentIndex":5,"endSegmentIndex":12,"startTime":638.48,"endTime":721.519,"durationSeconds":83,"preview":"검증 루프의 깊이","mustSee":true},{"clipId":"9DHZLw5653E:c1:13-16","startSegmentIndex":13,"endSegmentIndex":16,"startTime":721.519,"endTime":758.24,"durationSeconds":36.7,"preview":"닫힌 고리의 핵심","mustSee":false},{"clipId":"9DHZLw5653E:c1:17-26","startSegmentIndex":17,"endSegmentIndex":26,"startTime":758.24,"endTime":832.56,"durationSeconds":74.3,"preview":"조직 구조의 재설계","mustSee":false},{"clipId":"9DHZLw5653E:c1:27-40","startSegmentIndex":27,"endSegmentIndex":40,"startTime":832.56,"endTime":921.92,"durationSeconds":89.4,"preview":"속도와 품질의 재정의","mustSee":false},{"clipId":"9DHZLw5653E:c1:41-49","startSegmentIndex":41,"endSegmentIndex":49,"startTime":921.92,"endTime":1000,"durationSeconds":78.1,"preview":"빠른 실험의 가치","mustSee":false},{"clipId":"9DHZLw5653E:c1:50-76","startSegmentIndex":50,"endSegmentIndex":76,"startTime":1000,"endTime":1198,"durationSeconds":198,"preview":"ROI와 리더의 과제","mustSee":true}],"curatedSegments":[{"segmentIndex":14,"text":"In the back of my head, I was always a bit worried like we were talking about before of like how this was change completely changing the way we were working and I was pretty worried about that from just my personal point of view like am I going to miss that part of like the hard mental challenge of solving problems and now I find myself having you know five agents working in the background and my way of interacting with them is very different from the way that I was working a year or two ago and for me that's um turned out that I was wrong and I like the thing that I like to do is solving problems and the way that I solve those problems turn out to not be the most critical piece for me.","startTime":1290.32,"endTime":1332.559,"durationSeconds":42,"level":"C1","overallScore":7.6,"rationale":"개인적 고민과 깨달음이 길게 전개됨."},{"segmentIndex":22,"text":"It went from being this like smart autocomplete to something that I could actually throw real problems at and I didn't have to do all that much prompt engineering.","startTime":155.12,"endTime":164.8,"durationSeconds":10,"level":"B2","overallScore":6.8,"rationale":"도구의 변화와 체감 효과가 분명함."},{"segmentIndex":23,"text":"The biggest thing for me was also just not having to edit code anymore cuz my workflow up to then was I have the model write you know like maybe 80% of the code or 70% of the code depending on the model and then I always had to go into an IDE to do the last mile edits >> and I just stopped having to do that.","startTime":164.8,"endTime":182.56,"durationSeconds":18,"level":"C1","overallScore":7,"rationale":"워크플로 변화가 구체적으로 드러남."},{"segmentIndex":28,"text":">> Y >> and to me it feels kind of like a false dichotomy because if you want to go faster, the thing that you need to do is you need to automate your quality practices so that it's better encoded. It's not in someone's head.","startTime":840.16,"endTime":848.079,"durationSeconds":8,"level":"C1","overallScore":6.8,"rationale":"추상적 주장과 이유 제시가 좋음."},{"segmentIndex":31,"text":">> Y >> and that's ultimately what lets you go faster. And this is just another example of how in engineering productivity is always about investing in infrastructure. It's not about working more hours.","startTime":857.8389999999999,"endTime":865.279,"durationSeconds":7,"level":"C1","overallScore":6.6,"rationale":"성과와 원리를 연결해 설명함."},{"segmentIndex":61,"text":"So basically like the planned work that we have and then that connects to uh AB tests and rollouts and then we're able to from that see like basically attribute back to say this PR contributed to this uh DoD that we have and that contributed to this user value.","startTime":1066.48,"endTime":1084.64,"durationSeconds":18,"level":"C1","overallScore":6.8,"rationale":"인과관계와 측정 연결을 설명함."},{"segmentIndex":4,"text":"So I think if there's one advice I would give would be to not ignore those types of investments.","startTime":1228.72,"endTime":1235.52,"durationSeconds":7,"level":"B2","overallScore":6.4,"rationale":"조언이 담긴 일반화 문장이다."},{"segmentIndex":21,"text":"So that was certainly one for I have to say for my own personal coding the real breakthrough moment was probably Opus 4 or 5 back in November December.","startTime":144.4,"endTime":155.12,"durationSeconds":11,"level":"B2","overallScore":5.2,"rationale":"개인적 전환점이지만 표현은 단순함."},{"segmentIndex":35,"text":"Do you feel like one like monor repo or poly repo is a better fit for quad or >> I was a bit worried to be honest about the monor repo setup and agents originally because um I think with some of the prior tools we've been using we've been seeing issues with indexing and things like that um and this these are fairly large repositories that our backend monor repo is more than 20 million lines of code but turns out it cla works amazingly well in those repositories and Um, I think one of the things we found is how good Claude is looking at other code in the repository to get, I guess, inspiration for the problem you're trying to solve.","startTime":258.639,"endTime":269.759,"durationSeconds":11,"level":"C1","overallScore":5.8,"rationale":"대규모 저장소에서의 성능 통찰."},{"segmentIndex":61,"text":"So that inspired us as I mentioned before as pretty much as soon as the early LLMs came along of like hey these things can we apply them to this problem and early on it didn't work at all that well.","startTime":479.28,"endTime":493.44,"durationSeconds":14,"level":"B2","overallScore":5.6,"rationale":"시도 실패와 이유를 설명해 통찰 있음."},{"segmentIndex":65,"text":"So we started applying LMS as judge to make sure that the output was as intended.","startTime":514.8,"endTime":520.479,"durationSeconds":6,"level":"B2","overallScore":5.8,"rationale":"판단자를 활용한 방법론 설명."},{"segmentIndex":71,"text":"But it was the first sort of light in the tunnel of like yeah this is actually a problem that we can solve.","startTime":545.839,"endTime":551.68,"durationSeconds":6,"level":"B1","overallScore":5.4,"rationale":"문제 해결 가능성을 깨달은 전환점."},{"segmentIndex":1,"text":"verification step like a agentic verifier. Tell me more about that. >> So we used to have a judge in honk but we actually have removed that because we found that the uh agent and models just again going back to four or five got good enough that we don't didn't need judge anymore.","startTime":600,"endTime":614.64,"durationSeconds":15,"level":"B2","overallScore":5.8,"rationale":"판단 변화와 이유를 설명함."},{"segmentIndex":4,"text":"So >> so it's a big change but then again as we talked about the models caught up and the agent hardness caught up so we have now eliminated that judge from honk.","startTime":626.64,"endTime":638.48,"durationSeconds":12,"level":"B2","overallScore":5.4,"rationale":"변화 원인을 다시 정리함."},{"segmentIndex":16,"text":">> I think that's very true and I think it's true for us as well. One of the major changes that we did in our engineering practices as part of that was to strengthen our test automation.","startTime":749.519,"endTime":758.24,"durationSeconds":9,"level":"B2","overallScore":5.6,"rationale":"테스트 자동화 강화를 말함."},{"segmentIndex":25,"text":"So that meant then having to build out much better test automation to make sure that uh all our software could sort of survive those types of automated changes.","startTime":810.72,"endTime":821.04,"durationSeconds":10,"level":"B2","overallScore":5.6,"rationale":"자동 변경을 위한 교훈이 있음."},{"segmentIndex":35,"text":"Um I think we we're going to have to continue our investments into uh our reliability practices as well. I think we want to be able to basically have an idea and for a developer to have an idea and be able to ship that into production as quickly as possible.","startTime":887.92,"endTime":894.72,"durationSeconds":7,"level":"B2","overallScore":5.6,"rationale":"속도와 신뢰성 목표를 제시함."},{"segmentIndex":46,"text":"But in both of those cases, the faster we can iterate, we found that we um we both build better products and we're able to ship them faster to our users.","startTime":973.12,"endTime":982.32,"durationSeconds":9,"level":"B2","overallScore":5.4,"rationale":"빠른 반복의 효과를 말함."},{"segmentIndex":66,"text":"Yet, you know, as engineers, we still want to measure it. I'm going to say like the ROI discussion initially was fairly easy because we could see such large improvements and um but as the maturity is getting there and the costs have been improving.","startTime":1109.919,"endTime":1113.36,"durationSeconds":3,"level":"C1","overallScore":5.8,"rationale":"ROI 논의를 설명하며 문장도 복합적."},{"segmentIndex":71,"text":">> And now you know people are seeing these like many dozens or hundreds of percentage points of improvement and now you really want to attribute it to figure out like how many tokens did it take?","startTime":1146.96,"endTime":1155.36,"durationSeconds":8,"level":"C1","overallScore":5.6,"rationale":"성과를 수치로 귀속하는 질문."}],"generatedAt":"2026-06-30T13:57:54.624Z","keyClipsTotalSec":639},{"videoId":"9DHZLw5653E","chunkIndex":2,"totalChunks":3,"title":"How Spotify runs agents across 20M+ lines of code, with Niklas Gustavsson — Part 3 of 3","thumbnail":"https://i.ytimg.com/vi/9DHZLw5653E/maxresdefault.jpg","duration":1571,"uploader":"Claude","youtubeUrl":"https://www.youtube.com/watch?v=9DHZLw5653E","keywords":["agentic-ai","software-engineering","codebase-consistency","prototyping","developer-tools","productivity","ai-coding","workflow-change"],"normalizedKeywords":["엔지니어링","프로덕트","기술 트렌드"],"targetAudience":[{"who":"엔지니어","why":"AI 에이전트가 실제 개발 방식과 코드 품질에 미치는 영향을 직접 다룸"},{"who":"프로덕트 매니저","why":"프로토타이핑과 아이디어 검증 속도가 어떻게 바뀌는지 배울 수 있음"},{"who":"스타트업 실무자","why":"적은 자원으로 빠르게 실험하고 공유하는 운영 방식을 참고할 수 있음"}],"normalizedAudience":["엔지니어·개발자","프로덕트 매니저·기획자","창업자·스타트업"],"summary":"이 구간은 AI 에이전트가 개발 현장을 어떻게 바꾸고 있는지, 그리고 그 변화 속에서 무엇이 여전히 중요해지는지를 이야기한다. 핵심 메시지는 '코드베이스의 일관성과 기본적인 엔지니어링 원칙'이 인간뿐 아니라 에이전트에게도 더 큰 생산성을 만든다는 점이다. 서로 다른 스타일의 코드가 많을수록 Claude 같은 도구가 혼란스러워지므로, 평소의 정돈된 엔지니어링 투자가 오히려 AI 시대에 더 중요해진다고 강조한다.\n\n또한 개발자의 시간은 구현 자체에서 기획, 고객 대화, 프로토타이핑으로 이동하고 있다고 말한다. Spotify는 자연어로 아이디어를 입력하면 모바일 앱과 백엔드에서 바로 프로토타입을 만들 수 있는 내부 인프라와 앱 스토어를 구축했고, 이제는 엔지니어뿐 아니라 디자이너, PM, 심지어 임원까지 아이디어를 빠르게 실험한다. 예전엔 몇 주~몇 달 걸리던 검증이 이제는 하루 만에도 가능해졌다는 점이 이 구간의 핵심이다.","insights":["AI 에이전트도 결국 깔끔한 코드베이스에서 가장 잘 작동한다.","일관성 없는 코드 스타일은 사람보다 에이전트를 더 혼란스럽게 만든다.","구현 시간은 줄고, 고객 대화·기획·실험의 비중은 커진다.","프로토타이핑은 이제 엔지니어만의 영역이 아니라 전사 역량이다.","아이디어는 자연어로 적고, 하루 안에 실물로 검증하는 시대다."],"keyClips":[{"clipId":"9DHZLw5653E:c2:1-7","startSegmentIndex":1,"endSegmentIndex":7,"startTime":1201.19,"endTime":1253.44,"durationSeconds":52.3,"preview":"일관성이 에이전트를 돕는다","mustSee":false},{"clipId":"9DHZLw5653E:c2:12-21","startSegmentIndex":12,"endSegmentIndex":21,"startTime":1274.4,"endTime":1378.08,"durationSeconds":103.7,"preview":"문제해결의 재정의","mustSee":false},{"clipId":"9DHZLw5653E:c2:22-25","startSegmentIndex":22,"endSegmentIndex":25,"startTime":1378.08,"endTime":1402.64,"durationSeconds":24.6,"preview":"시간의 재배치","mustSee":false},{"clipId":"9DHZLw5653E:c2:26-34","startSegmentIndex":26,"endSegmentIndex":34,"startTime":1402.64,"endTime":1514.799,"durationSeconds":112.2,"preview":"전사 프로토타입 문화","mustSee":true},{"clipId":"9DHZLw5653E:c2:35-40","startSegmentIndex":35,"endSegmentIndex":40,"startTime":1514.799,"endTime":1565.12,"durationSeconds":50.3,"preview":"하루짜리 검증","mustSee":true}],"curatedSegments":[{"segmentIndex":14,"text":"In the back of my head, I was always a bit worried like we were talking about before of like how this was change completely changing the way we were working and I was pretty worried about that from just my personal point of view like am I going to miss that part of like the hard mental challenge of solving problems and now I find myself having you know five agents working in the background and my way of interacting with them is very different from the way that I was working a year or two ago and for me that's um turned out that I was wrong and I like the thing that I like to do is solving problems and the way that I solve those problems turn out to not be the most critical piece for me.","startTime":1290.32,"endTime":1332.559,"durationSeconds":42,"level":"C1","overallScore":7.6,"rationale":"개인적 고민과 깨달음이 길게 전개됨."},{"segmentIndex":22,"text":"It went from being this like smart autocomplete to something that I could actually throw real problems at and I didn't have to do all that much prompt engineering.","startTime":155.12,"endTime":164.8,"durationSeconds":10,"level":"B2","overallScore":6.8,"rationale":"도구의 변화와 체감 효과가 분명함."},{"segmentIndex":23,"text":"The biggest thing for me was also just not having to edit code anymore cuz my workflow up to then was I have the model write you know like maybe 80% of the code or 70% of the code depending on the model and then I always had to go into an IDE to do the last mile edits >> and I just stopped having to do that.","startTime":164.8,"endTime":182.56,"durationSeconds":18,"level":"C1","overallScore":7,"rationale":"워크플로 변화가 구체적으로 드러남."},{"segmentIndex":28,"text":">> Y >> and to me it feels kind of like a false dichotomy because if you want to go faster, the thing that you need to do is you need to automate your quality practices so that it's better encoded. It's not in someone's head.","startTime":840.16,"endTime":848.079,"durationSeconds":8,"level":"C1","overallScore":6.8,"rationale":"추상적 주장과 이유 제시가 좋음."},{"segmentIndex":31,"text":">> Y >> and that's ultimately what lets you go faster. And this is just another example of how in engineering productivity is always about investing in infrastructure. It's not about working more hours.","startTime":857.8389999999999,"endTime":865.279,"durationSeconds":7,"level":"C1","overallScore":6.6,"rationale":"성과와 원리를 연결해 설명함."},{"segmentIndex":61,"text":"So basically like the planned work that we have and then that connects to uh AB tests and rollouts and then we're able to from that see like basically attribute back to say this PR contributed to this uh DoD that we have and that contributed to this user value.","startTime":1066.48,"endTime":1084.64,"durationSeconds":18,"level":"C1","overallScore":6.8,"rationale":"인과관계와 측정 연결을 설명함."},{"segmentIndex":4,"text":"So I think if there's one advice I would give would be to not ignore those types of investments.","startTime":1228.72,"endTime":1235.52,"durationSeconds":7,"level":"B2","overallScore":6.4,"rationale":"조언이 담긴 일반화 문장이다."},{"segmentIndex":21,"text":"So that was certainly one for I have to say for my own personal coding the real breakthrough moment was probably Opus 4 or 5 back in November December.","startTime":144.4,"endTime":155.12,"durationSeconds":11,"level":"B2","overallScore":5.2,"rationale":"개인적 전환점이지만 표현은 단순함."},{"segmentIndex":35,"text":"Do you feel like one like monor repo or poly repo is a better fit for quad or >> I was a bit worried to be honest about the monor repo setup and agents originally because um I think with some of the prior tools we've been using we've been seeing issues with indexing and things like that um and this these are fairly large repositories that our backend monor repo is more than 20 million lines of code but turns out it cla works amazingly well in those repositories and Um, I think one of the things we found is how good Claude is looking at other code in the repository to get, I guess, inspiration for the problem you're trying to solve.","startTime":258.639,"endTime":269.759,"durationSeconds":11,"level":"C1","overallScore":5.8,"rationale":"대규모 저장소에서의 성능 통찰."},{"segmentIndex":61,"text":"So that inspired us as I mentioned before as pretty much as soon as the early LLMs came along of like hey these things can we apply them to this problem and early on it didn't work at all that well.","startTime":479.28,"endTime":493.44,"durationSeconds":14,"level":"B2","overallScore":5.6,"rationale":"시도 실패와 이유를 설명해 통찰 있음."},{"segmentIndex":65,"text":"So we started applying LMS as judge to make sure that the output was as intended.","startTime":514.8,"endTime":520.479,"durationSeconds":6,"level":"B2","overallScore":5.8,"rationale":"판단자를 활용한 방법론 설명."},{"segmentIndex":71,"text":"But it was the first sort of light in the tunnel of like yeah this is actually a problem that we can solve.","startTime":545.839,"endTime":551.68,"durationSeconds":6,"level":"B1","overallScore":5.4,"rationale":"문제 해결 가능성을 깨달은 전환점."},{"segmentIndex":1,"text":"verification step like a agentic verifier. Tell me more about that. >> So we used to have a judge in honk but we actually have removed that because we found that the uh agent and models just again going back to four or five got good enough that we don't didn't need judge anymore.","startTime":600,"endTime":614.64,"durationSeconds":15,"level":"B2","overallScore":5.8,"rationale":"판단 변화와 이유를 설명함."},{"segmentIndex":4,"text":"So >> so it's a big change but then again as we talked about the models caught up and the agent hardness caught up so we have now eliminated that judge from honk.","startTime":626.64,"endTime":638.48,"durationSeconds":12,"level":"B2","overallScore":5.4,"rationale":"변화 원인을 다시 정리함."},{"segmentIndex":16,"text":">> I think that's very true and I think it's true for us as well. One of the major changes that we did in our engineering practices as part of that was to strengthen our test automation.","startTime":749.519,"endTime":758.24,"durationSeconds":9,"level":"B2","overallScore":5.6,"rationale":"테스트 자동화 강화를 말함."},{"segmentIndex":25,"text":"So that meant then having to build out much better test automation to make sure that uh all our software could sort of survive those types of automated changes.","startTime":810.72,"endTime":821.04,"durationSeconds":10,"level":"B2","overallScore":5.6,"rationale":"자동 변경을 위한 교훈이 있음."},{"segmentIndex":35,"text":"Um I think we we're going to have to continue our investments into uh our reliability practices as well. I think we want to be able to basically have an idea and for a developer to have an idea and be able to ship that into production as quickly as possible.","startTime":887.92,"endTime":894.72,"durationSeconds":7,"level":"B2","overallScore":5.6,"rationale":"속도와 신뢰성 목표를 제시함."},{"segmentIndex":46,"text":"But in both of those cases, the faster we can iterate, we found that we um we both build better products and we're able to ship them faster to our users.","startTime":973.12,"endTime":982.32,"durationSeconds":9,"level":"B2","overallScore":5.4,"rationale":"빠른 반복의 효과를 말함."},{"segmentIndex":66,"text":"Yet, you know, as engineers, we still want to measure it. I'm going to say like the ROI discussion initially was fairly easy because we could see such large improvements and um but as the maturity is getting there and the costs have been improving.","startTime":1109.919,"endTime":1113.36,"durationSeconds":3,"level":"C1","overallScore":5.8,"rationale":"ROI 논의를 설명하며 문장도 복합적."},{"segmentIndex":71,"text":">> And now you know people are seeing these like many dozens or hundreds of percentage points of improvement and now you really want to attribute it to figure out like how many tokens did it take?","startTime":1146.96,"endTime":1155.36,"durationSeconds":8,"level":"C1","overallScore":5.6,"rationale":"성과를 수치로 귀속하는 질문."}],"generatedAt":"2026-06-30T13:57:40.729Z","keyClipsTotalSec":639}]}