{"success":true,"count":1,"items":[{"videoId":"DGD9b8K42lk","chunkIndex":0,"totalChunks":1,"title":"How to Build an Internal AI Agent That Evolves Itself","thumbnail":"https://i.ytimg.com/vi/DGD9b8K42lk/maxresdefault.jpg","duration":334,"uploader":"YC Root Access","youtubeUrl":"https://www.youtube.com/watch?v=DGD9b8K42lk","keywords":["ai-agents","startup-automation","workflow-automation","llm-tools","internal-tools","customer-support","crm","developer-tools","sop","artificial-intelligence"],"normalizedKeywords":["엔지니어링","비즈니스·전략","프로덕트"],"targetAudience":[{"who":"초기 단계 창업자","why":"적은 인력으로 운영을 자동화해 founder time을 줄이는 설계가 핵심이다"},{"who":"엔지니어링 리더","why":"에이전트 하네스, CLI, 읽기 전용 DB/코드 연결 구조가 구체적이다"},{"who":"프로덕트 매니저","why":"고객지원과 운영을 지식화·자동화하는 내부 프로세스 설계에 유용하다"}],"normalizedAudience":["창업자·스타트업","엔지니어·개발자","프로덕트 매니저·기획자"],"summary":"이 영상은 작은 팀이 내부 AI 에이전트를 활용해 운영, 고객지원, CRM 업데이트, 비즈니스 질의응답까지 자동화한 방법을 설명한다. 화자는 단순히 정해진 작업을 수행하는 에이전트가 아니라, 반복 작업을 만나면 스스로 도구를 만들고 그 도구를 영구적으로 축적하는 'self-extending' 구조가 핵심이라고 강조한다.\n\n구현 관점에서는 Claude code 같은 코딩 가능한 CLI를 에이전트 하네스로 두고, Slack·이메일 같은 입력을 task queue로 받으며, 읽기 전용 코드베이스와 DB를 연결해 비즈니스 맥락을 이해하게 만든다. 또한 instructions.md 같은 편집 가능한 메모리 파일을 두어 행동 피드백을 누적시키고, coding sub-agent가 agent 자체의 기능을 확장하게 함으로써 에이전트가 시간이 갈수록 더 유능해지도록 설계한다.","insights":["운영 자동화의 진짜 가치는 반복 업무 제거보다 지식의 즉시조회성이다.","에이전트는 고정 기능보다 스스로 도구를 늘릴 때 비로소 유용해진다.","비즈니스 맥락은 코드와 DB를 읽게 해야 모델의 환각을 줄일 수 있다.","행동 피드백을 편집 가능한 메모리로 남겨야 에이전트가 학습한다.","사람에게 주는 피드백을 에이전트에도 그대로 적용해야 개선이 빨라진다."],"keyClips":[{"clipId":"DGD9b8K42lk:c0:4-13","startSegmentIndex":4,"endSegmentIndex":13,"startTime":13.6,"endTime":96.2,"durationSeconds":82.6,"preview":"소수 인력의 자동화","mustSee":true},{"clipId":"DGD9b8K42lk:c0:15-21","startSegmentIndex":15,"endSegmentIndex":21,"startTime":100,"endTime":161.12,"durationSeconds":61.1,"preview":"비즈니스 맥락 주입","mustSee":true},{"clipId":"DGD9b8K42lk:c0:22-29","startSegmentIndex":22,"endSegmentIndex":29,"startTime":161.12,"endTime":217.48,"durationSeconds":56.4,"preview":"스스로 도구를 만든다","mustSee":true},{"clipId":"DGD9b8K42lk:c0:30-38","startSegmentIndex":30,"endSegmentIndex":38,"startTime":217.48,"endTime":270.68,"durationSeconds":53.2,"preview":"수정 가능한 기억","mustSee":true},{"clipId":"DGD9b8K42lk:c0:39-49","startSegmentIndex":39,"endSegmentIndex":49,"startTime":270.68,"endTime":326.52,"durationSeconds":55.8,"preview":"세 가지 기억 설계","mustSee":true}],"curatedSegments":[{"segmentIndex":13,"text":"When it runs into a repeated task it cannot do yet, it asks a coding sub-agent to build a tool for it, and this tool becomes permanent and is available in future sessions.","startTime":84.03999999999999,"endTime":96.2,"durationSeconds":12,"level":"C1","overallScore":9,"rationale":"자가 확장 메커니즘을 구체적으로 설명."},{"segmentIndex":12,"text":"Now, the most important part of this is not that the agent can do a fixed set of tasks, is that the agent is self-extending.","startTime":74.72,"endTime":84.03999999999999,"durationSeconds":9,"level":"C1","overallScore":8,"rationale":"핵심 원리를 대비 구조로 강조함."},{"segmentIndex":20,"text":"A great way that we've found to be able to do this is by giving the agent a read-only copy of both our database as well as our code base, where it has a cron job that basically gives it an updated version every time we do a release.","startTime":131.76,"endTime":147.4,"durationSeconds":16,"level":"C1","overallScore":8.2,"rationale":"구현 팁과 구조 표현이 매우 풍부함."},{"segmentIndex":21,"text":"And whenever a customer support query comes now and they ask some question about the business, the agent can just read our code base to be able to figure that out, including things like what our subscription logic is and where things inside of our app are located.","startTime":147.4,"endTime":161.12,"durationSeconds":14,"level":"C1","overallScore":8,"rationale":"코드베이스 활용 원리를 실용적으로 설명."},{"segmentIndex":27,"text":"And what this means is that whenever we ask it to do a task it cannot do yet, it is simply able to just code that tool into existence and handle the task.","startTime":182.64,"endTime":192.72,"durationSeconds":10,"level":"C1","overallScore":8.2,"rationale":"의미와 결과를 선명하게 풀어 설명함."},{"segmentIndex":28,"text":"To us it's magical because we only ask it to do things, but it's able to self-author tools and has gone from just being a skeleton to being this full-blown tool with over 45 CLIs that it has made itself.","startTime":192.72,"endTime":206.12,"durationSeconds":13,"level":"C1","overallScore":8.2,"rationale":"진화 과정을 생생하게 보여주는 문장."},{"segmentIndex":33,"text":"This is how we are able to self-evolve the agent because we can give it feedback like we would give feedback to an employee and it just updates this instruction.","startTime":232.92000000000002,"endTime":240.04,"durationSeconds":7,"level":"C1","overallScore":8,"rationale":"자가 진화 원리를 비유와 함께 설명함."},{"segmentIndex":38,"text":"The agent updated its own instruction set and tool link and then that entire class of mistakes stopped happening again.","startTime":261,"endTime":270.68,"durationSeconds":10,"level":"C1","overallScore":8,"rationale":"피드백 루프의 효과가 선명하게 드러남."},{"segmentIndex":40,"text":"It needs factual memories, which is your code base and your database, things how they work operationally within your startup.","startTime":277.52,"endTime":285,"durationSeconds":7,"level":"C1","overallScore":8,"rationale":"메모리 유형을 개념적으로 정리함."},{"segmentIndex":4,"text":"So, we've been able to do over $2 million in ARR largely being two full-time employees, which is myself and my co-founder.","startTime":13.6,"endTime":22.36,"durationSeconds":9,"level":"B2","overallScore":6.8,"rationale":"성과와 운영 방식이 인상적으로 제시됨."},{"segmentIndex":5,"text":"Do have two or three contractors for things like design and outbound, but a large reason for why we've been able to do this is because we have an internal AI ops agent that handles a lot of the work that would normally consume founder time.","startTime":22.36,"endTime":36.96,"durationSeconds":15,"level":"C1","overallScore":7.2,"rationale":"핵심 원인과 효용을 구체적으로 설명."},{"segmentIndex":8,"text":"It collects user feedback across channels, helps with things like customer support, but more importantly, as Pete and Tom and Gary mentioned, it also lets us ask questions about our business at any time.","startTime":46.96,"endTime":60.56,"durationSeconds":14,"level":"C1","overallScore":7.2,"rationale":"기능 확장과 핵심 효용을 잘 드러냄."},{"segmentIndex":11,"text":"This has become instantly queryable for us, whereas previously, you know, we had to operate a bunch of different apps to get answers to our question.","startTime":65.08,"endTime":74.72,"durationSeconds":10,"level":"C1","overallScore":7.2,"rationale":"업무 방식 변화의 핵심 차이를 설명함."},{"segmentIndex":15,"text":"The first thing you should do is have a Claude code CLI wrapped in Python, where new messages from Slack, email, and other channels go into a task queue.","startTime":100,"endTime":109.47999999999999,"durationSeconds":9,"level":"C1","overallScore":7.2,"rationale":"실행 지침과 구조가 명확히 담김."},{"segmentIndex":26,"text":"But at the same time, it has a general coding agent also as a CLI, which can edit the agent code itself.","startTime":175.88,"endTime":182.64,"durationSeconds":7,"level":"C1","overallScore":7,"rationale":"자가 수정 가능성의 핵심을 짚음."},{"segmentIndex":29,"text":"A cool example for this is also when we just wanted to monitor our landing pages to make sure they're always up for ads, and we just sort of told it and it created a cron job into existence that does this.","startTime":206.12,"endTime":217.48,"durationSeconds":11,"level":"C1","overallScore":7.2,"rationale":"구체 사례라 이해와 표현 학습에 좋음."},{"segmentIndex":30,"text":"Now, moving slightly further, there's one very important thing that also this agent needs to have, which is an editable personality or memory.","startTime":217.48,"endTime":225.64,"durationSeconds":8,"level":"C1","overallScore":7.2,"rationale":"중요 요구사항을 강조하며 확장함."},{"segmentIndex":37,"text":"Instead of opening the code base or telling me or filing a ticket, he just messaged the agent in Slack and told it what was wrong.","startTime":256.12,"endTime":261,"durationSeconds":5,"level":"B2","overallScore":7.2,"rationale":"기존 방식 대비 새 흐름을 잘 보여줌."},{"segmentIndex":39,"text":"So, the broader lesson here is that uh an internal agent needs sort of three sorts of memories.","startTime":270.68,"endTime":277.52,"durationSeconds":7,"level":"B2","overallScore":7,"rationale":"핵심 교훈을 일반화해 제시함."},{"segmentIndex":41,"text":"It needs behavioral memories. This is what you teach the agent.","startTime":285,"endTime":287.2,"durationSeconds":2,"level":"B2","overallScore":6.8,"rationale":"두 번째 개념을 간결하게 정의함."}],"generatedAt":"2026-06-25T00:27:26.185Z","keyClipsTotalSec":309}]}