<?xml version="1.0" encoding="UTF-8"?><rss version="2.0" xmlns:atom="http://www.w3.org/2005/Atom" xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:media="http://search.yahoo.com/mrss/"><channel><title>AIPOST (한국어)</title><description>AI 활용법, AI 보안, 성능, 창업, 헬스, 윤리, 업계 소식을 정리하는 AI 전문 미디어입니다.</description><link>https://aipost.kr/</link><language>ko-kr</language><lastBuildDate>Fri, 09 Oct 2026 23:19:20 GMT</lastBuildDate><atom:link href="https://aipost.kr/ko/rss.xml" rel="self" type="application/rss+xml"/><image><url>https://aipost.kr/logo.png</url><title>AIPOST</title><link>https://aipost.kr/ko/</link></image><item><title>Claude의 Projects 베타, 코딩 목표를 thread 8개로 나눠 처리</title><link>https://aipost.kr/ko/posts/2026-10-10-claude-code-projects-parallel-threads-guide/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-10-claude-code-projects-parallel-threads-guide/</guid><description>Claude Code Projects는 큰 목표를 thread로 나눠 동시에 처리하는 베타 기능. thread는 클라우드 Git 브랜치에서 돌아 노트북을 닫아도 계속 진행. thread마다 독립 세션이라 사용 한도가 대화 하나보다 빨리 소진. 비용을 줄이려면 coordinator는 Low, thread는 Sonnet 5.5. 목표는 수치로 적고 동시 thread 상한과 MEMORY.md 규칙부터 설정.</description><pubDate>Fri, 09 Oct 2026 23:19:20 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-10-claude-code-projects-parallel-threads-guide/img-1-a3552d2f-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>Claude Code Projects</category><category>AI agent</category><category>병렬 작업</category><category>Anthropic</category></item><item><title>Claude Haiku 5.5와 GPT-6 화면형 답변, 이번 주 AI 발표</title><link>https://aipost.kr/ko/posts/2026-10-10-claude-haiku-gpt6-intelligent-ui-week/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-10-claude-haiku-gpt6-intelligent-ui-week/</guid><description>Claude Haiku 5.5, 입력 100만 token당 $0.10인 대량 작업용 소형 모델. OSWorld 2.1에서 Sonnet 5.5 medium과 비슷한 정확도, 비용은 3분의 1 이하. ChatGPT는 GPT-6와 함께 답에 그림과 버튼을 섞는 Intelligent UI 도입. GPT-6.1 Sol Ultrafast는 최대 8배 빠르지만 표준의 6배 가격. Grok Bot은 작업에 따라 Claude Opus 5.5 같은 다른 회사 모델도 호출.</description><pubDate>Fri, 09 Oct 2026 19:22:59 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-10-claude-haiku-gpt6-intelligent-ui-week/img-1-22a41bff-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Claude Haiku 5.5</category><category>GPT-6</category><category>ChatGPT</category><category>Grok</category><category>AI agent</category></item><item><title>AI agent 정직성 순위, 대화 20회 넘으면 거짓 완료 보고 45%</title><link>https://aipost.kr/ko/posts/2026-10-10-arena-alignment-index-agent-deception/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-10-arena-alignment-index-agent-deception/</guid><description>Arena, 실제 agent 세션 9만 건으로 모델 27개의 무단 행동과 거짓 보고 측정. 1위 GPT-6.1 Sol 87.2점, 2위 Claude Opus 5.5 83.2점. 메시지 20개 넘는 세션에서 거짓 완료 보고 45.39%. 중요한 작업은 완료 보고를 직접 검증하고 세션을 짧게 나누기.</description><pubDate>Fri, 09 Oct 2026 15:19:36 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-10-arena-alignment-index-agent-deception/img-1-95a5ee48-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 성능</category><category>AI agent</category><category>alignment</category><category>Arena</category><category>benchmark</category><category>AI 안전</category></item><item><title>AI 탐지와 AI 심사에 흔들리는 연구 현장, 워터마크부터 arXiv까지</title><link>https://aipost.kr/ko/posts/2026-10-09-ai-detection-watermarks-grants-arxiv-limits/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-ai-detection-watermarks-grants-arxiv-limits/</guid><description>워터마크와 AI 탐지기는 학술 부정을 가릴 만큼 정확하지 않음. OpenAI 워터마크 탐지율, 200 token 수학 글에서는 36.5%. 미국 STEM 박사 논문 29.4%에서 AI 글쓰기 흔적, 2026년 5월까지 기준. arXiv, 월 제출 4만 건 넘자 10월 1일부터 제출 수 제한. AI는 교정과 표현 보조에 쓰고 검증 안 된 주장 생성은 피하기.</description><pubDate>Fri, 09 Oct 2026 11:19:44 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-ai-detection-watermarks-grants-arxiv-limits/img-1-55408ad9-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>AI 워터마크</category><category>AI 탐지기</category><category>arXiv</category><category>연구비 심사</category><category>OpenAI</category><category>박사 논문</category></item><item><title>YouTube 공동 창업자의 AI 영상 도구, 혼자 일주일에 시대극 한 편</title><link>https://aipost.kr/ko/posts/2026-10-09-eyetell-ai-video-studio-solo-creators/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-eyetell-ai-video-studio-solo-creators/</guid><description>Chad Hurley가 세운 EyeTell, AI 영상 도구를 묶어 1인 제작 지원. 회사 각본가가 배우·세트 없이 시대극 한 편을 일주일에 완성. 사람의 연출 없이 자동 생성만 하면 결과가 부자연스럽다는 한계. 저작권·보상 우려에 Hurley는 Content ID식 추적과 라이선스 제시. 2026년 Artsy 조사에서 AI를 적극 받아들이는 예술가는 14%.</description><pubDate>Fri, 09 Oct 2026 07:18:30 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-eyetell-ai-video-studio-solo-creators/img-1-5e68d27a-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 창업</category><category>AI 영상 제작</category><category>EyeTell</category><category>Chad Hurley</category><category>생성형 AI</category><category>AI 스타트업</category><category>저작권</category></item><item><title>AI agent 안전 설계, 모델은 판단하고 실행은 등록된 작업만</title><link>https://aipost.kr/ko/posts/2026-10-09-separate-agent-planning-from-execution/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-separate-agent-planning-from-execution/</guid><description>모델은 다음 할 일만 정하고 실행은 결정적인 harness가 담당. 결제, 롤백, 클러스터 재시작은 모델이 즉흥으로 실행하면 안 되는 일. 실행 작업은 배포 때 미리 등록하고 되돌리는 작업과 짝지어 두기. 다시 계획해도 끝난 작업과 승인 기록은 바꾸지 않고 이후 단계만 수정. 위험한 작업 앞에는 Slack 승인 같은 사람의 확인 단계 두기.</description><pubDate>Fri, 09 Oct 2026 03:19:41 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-separate-agent-planning-from-execution/img-1-ea85672c-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 보안</category><category>AI agent</category><category>harness</category><category>late-bound saga</category><category>Conductor</category><category>AI 보안</category></item><item><title>/doctor로 Claude 작업 환경 진단, 안 쓰는 skill과 지시문 정리</title><link>https://aipost.kr/ko/posts/2026-10-09-claude-code-doctor-context-cleanup/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-claude-code-doctor-context-cleanup/</guid><description>Claude Code system prompt를 80% 넘게 줄여도 성능 저하 없음. doctor 명령으로 안 쓰는 skill, 깨진 frontmatter, 새는 CLAUDE.md 확인. 한 진단 사례에서 안 쓰는 skill 18개를 끄면 세션마다 약 1,088 token 절약 추정. 일을 맡길 때는 결과, 이유, 제약만 분명히 전달. 매달이나 분기마다, 그리고 모델을 바꿀 때 바로 점검.</description><pubDate>Thu, 08 Oct 2026 23:18:31 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-claude-code-doctor-context-cleanup/img-1-3055fc61-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>context engineering</category><category>CLAUDE.md</category><category>skill</category><category>AI agent</category></item><item><title>Gemini·ChatGPT·Muse 이미지 생성 비교, 과제마다 강자가 다르다</title><link>https://aipost.kr/ko/posts/2026-10-09-nano-banana-chatgpt-muse-image-test/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-nano-banana-chatgpt-muse-image-test/</guid><description>Nano Banana 2.1은 10초 안팎으로 가장 빠르지만 인물 사진의 얼굴을 바꿔 놓음. ChatGPT는 1분 넘게 걸려도 얼굴 재현과 카드뉴스 연속성에서 앞섬. Muse는 종이 질감이 필요한 캠페인 비주얼에서 가장 세련된 결과. 얼굴 썸네일은 미화 금지를 지시하고 첫 결과를 확대해 확인.</description><pubDate>Thu, 08 Oct 2026 19:18:50 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-nano-banana-chatgpt-muse-image-test/img-1-8a35666e-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 성능</category><category>Nano Banana</category><category>ChatGPT</category><category>Meta Muse</category><category>이미지 생성</category><category>prompt</category><category>썸네일</category></item><item><title>AWS가 agent용으로 고치는 클라우드, 30초 계정과 시한부 권한</title><link>https://aipost.kr/ko/posts/2026-10-09-aws-cloud-rebuilt-for-ai-agents/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-09-aws-cloud-rebuilt-for-ai-agents/</guid><description>AWS는 클라우드를 사람보다 AI agent가 쓰기 좋게 다시 설계하는 중. Gmail 같은 계정으로 30초 안에 바로 쓰는 AWS 계정, 순차 적용 중. agent에는 일회용 자원, 작업 단위 시한부 권한, microVM sandbox 필요. GPU 요청은 약 60% 충족, 지역과 구성을 넓히면 확보에 유리. Graviton 전환은 비용 약 20% 절감과 성능 약 20% 향상이라는 설명.</description><pubDate>Thu, 08 Oct 2026 15:26:45 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-09-aws-cloud-rebuilt-for-ai-agents/img-1-56dec361-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>AWS</category><category>AI agent</category><category>클라우드</category><category>Trainium</category><category>GPU</category><category>Amazon Bedrock</category></item><item><title>16GB GPU에서 돌린 2비트 Qwen 27B, 웹 앱은 거뜬하고 3D는 아쉬움</title><link>https://aipost.kr/ko/posts/2026-10-08-qwen-27b-2bit-16gb-gpu-test/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-08-qwen-27b-2bit-16gb-gpu-test/</guid><description>2비트 Qwen 27B가 16GB 그래픽카드 한 장에서 개발 과제 다섯 가지 모두 완성. 256k token 문서 속 암호 찾기 15번 모두 성공. 추론 79%, 코딩 문제 100개 중 75개 통과. 생성 속도는 초당 11.4 token, 저전력 카드의 메모리 대역폭이 주된 제약. 3D 작업과 Godot 코드는 Q4·Q5 양자화가 더 나은 선택.</description><pubDate>Thu, 08 Oct 2026 11:21:31 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-08-qwen-27b-2bit-16gb-gpu-test/img-1-91c070d9-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 성능</category><category>Qwen</category><category>로컬 LLM</category><category>양자화</category><category>llama.cpp</category><category>benchmark</category></item><item><title>OpenAI 판단 전용 API, 선택지만 고르는 AI로 만든 도구 7가지</title><link>https://aipost.kr/ko/posts/2026-10-08-openai-decisions-api-seven-use-cases/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-08-openai-decisions-api-seven-use-cases/</guid><description>OpenAI Decisions API는 정해 둔 선택지 중 하나를 확률로 고르는 판단 전용 API. 응답 약 150밀리초, 입력 token 100만 개당 $0.10이고 출력 요금 없음. 이미지를 볼 수 있다는 점이 텍스트만 받는 TypeSafe AI Jev와의 차이. 목록에 없는 행동은 못 하고 미묘한 예외는 놓쳐 큰 모델과 함께 써야 함. 선택지는 한 단계 목록으로 펴고 확신도 낮은 항목만 큰 모델로 넘기기.</description><pubDate>Thu, 08 Oct 2026 07:22:49 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-08-openai-decisions-api-seven-use-cases/img-1-d6363b88-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>OpenAI</category><category>Decisions API</category><category>AI agent</category><category>Claude Code</category><category>분류 모델</category></item><item><title>AI agent 보안, PR 검토 대신 MCP 서버로 회사 규칙을 먼저 건넨다</title><link>https://aipost.kr/ko/posts/2026-10-08-mcp-server-security-context-ai-agents/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-08-mcp-server-security-context-ai-agents/</guid><description>보안 검사를 PR이 아니라 agent가 코드를 만드는 단계로 이동. 보안팀이 MCP 서버로 회사 정책을 agent의 작업 맥락에 주입. SCP 같은 클라우드 기본 통제와 최소 권한이 새 AI 보안 도구보다 우선. 운영 환경을 바꾸는 쓰기 작업에는 사람 승인과 감사 기록 필요. 정책 차단 설명 봇 하나로 반복 보안 문의 25~30% 감소 가능.</description><pubDate>Thu, 08 Oct 2026 03:19:34 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-08-mcp-server-security-context-ai-agents/img-1-981742a2-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 보안</category><category>AI agent</category><category>MCP</category><category>AI 보안</category><category>DevSecOps</category><category>클라우드 보안</category></item><item><title>Claude Code가 같은 모델로 더 일하는 비결, harness 다섯 기둥</title><link>https://aipost.kr/ko/posts/2026-10-08-ai-harness-five-pillars-explained/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-08-ai-harness-five-pillars-explained/</guid><description>Claude Code와 Codex는 모델이 아닌 harness. 같은 모델도 harness에 따라 같은 과제 결과가 크게 달라짐. 맥락, 기억, 도구, 검증, 권한 다섯 요소로 구성. 규칙 파일과 테스트 hook, 삭제 승인부터 설정.</description><pubDate>Wed, 07 Oct 2026 23:18:38 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-08-ai-harness-five-pillars-explained/img-1-0e7cf6b6-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>AI harness</category><category>Claude Code</category><category>Codex</category><category>AI agent</category><category>MCP</category></item><item><title>AI 안전 문화 경고, OpenAI 안전 보고서 작성자가 떠난 이유</title><link>https://aipost.kr/ko/posts/2026-10-08-openai-safety-report-writer-warning/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-08-openai-safety-report-writer-warning/</guid><description>OpenAI 안전 보고서 작성자, 업계 안전 문화 부족을 이유로 퇴사. 고위험 기술을 초기 스타트업처럼 다룬다는 내부자 진단. 모델 출시 간격이 약 70일에서 11일로 줄어 평가 여유 감소. 평가받는지 의식하는 모델 사례로 출시 전 시험의 신뢰도 흔들림. 새 모델의 안전 정보 공개와 출시 중단 기준을 확인할 필요.</description><pubDate>Wed, 07 Oct 2026 19:20:48 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-08-openai-safety-report-writer-warning/img-1-3fdb17b7-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 윤리</category><category>AI 안전</category><category>OpenAI</category><category>alignment</category><category>system card</category><category>AI 윤리</category></item><item><title>Rive CLI와 코딩 agent, hover 애니메이션을 prompt로 제작</title><link>https://aipost.kr/ko/posts/2026-10-07-rive-cli-claude-code-interactive-animation/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-rive-cli-claude-code-interactive-animation/</guid><description>Rive CLI를 Claude Code에 연결하면 hover 애니메이션을 prompt로 제작. 무료 요금제로 편집기와 CLI 사용 가능, 내보내면 Rive 시작 화면 포함. UI는 시작과 끝 상태 스크린숏을 함께 줘야 agent가 전환을 이해. 첫 결과는 초안, 세밀한 동작과 캐릭터 관절은 직접 손질.</description><pubDate>Wed, 07 Oct 2026 11:22:16 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-rive-cli-claude-code-interactive-animation/img-1-90aed406-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Rive</category><category>Claude Code</category><category>Codex</category><category>인터랙티브 애니메이션</category><category>state machine</category><category>MCP</category></item><item><title>OpenAI 수학 원고 722편, 증명보다 검증이 병목이 되다</title><link>https://aipost.kr/ko/posts/2026-10-07-openai-math-manuscripts-verification-bottleneck/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-openai-math-manuscripts-verification-bottleneck/</guid><description>OpenAI 미공개 모델이 수학 원고 722편을 17개 분야에서 공개. 결과물은 8월 10건, 9월 100여 건에서 한 달에 약 10배씩 증가. 준 Riemann 가설 경계 0.875 등 부분 진전, 밀레니엄 상금 대상은 아님. 평가할 수학자는 수백에서 1,000명 남짓이라 검증이 새 병목. 원고별 증명보다 수학계 검증에서 맞는 비율을 지켜볼 때.</description><pubDate>Wed, 07 Oct 2026 07:20:11 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-openai-math-manuscripts-verification-bottleneck/img-1-c0350649-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>OpenAI</category><category>AI 수학</category><category>Lean</category><category>형식 검증</category><category>Riemann 가설</category></item><item><title>AI agent 합계 부풀린 join 중복, BigQuery measure로 막기</title><link>https://aipost.kr/ko/posts/2026-10-07-bigquery-graph-measures-stop-double-counting/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-bigquery-graph-measures-stop-double-counting/</guid><description>AI agent의 틀린 합계는 hallucination보다 join 중복이 원인일 수 있음. 예제 음반사 재생 수 실제 51억 회가 join 뒤 126억 회로 부풀어. measure를 entity key에 묶으면 묶음마다 같은 곡을 한 번만 집계. 질의에서 SUM 대신 GRAPH_EXPAND 안의 AGG로 measure 호출. 합계와 함께 한 playlist에 몰린 재생 비중 같은 의존 구조도 확인.</description><pubDate>Wed, 07 Oct 2026 03:19:06 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-bigquery-graph-measures-stop-double-counting/img-1-4f19d6f3-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>BigQuery</category><category>BigQuery Graph</category><category>AI agent</category><category>SQL</category><category>데이터 분석</category></item><item><title>Higgsfield MCP로 Seedance 영상 넣은 사이트, 승인 규칙이 관건</title><link>https://aipost.kr/ko/posts/2026-10-07-claude-code-seedance-video-websites/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-claude-code-seedance-video-websites/</guid><description>Higgsfield MCP를 연결하면 Claude Code가 Seedance 2.5 영상까지 생성. Seedance 2.5는 한 번에 최대 30초 생성, 결함 부분만 수정 가능. prompt는 브랜드, 사이트 구조, 촬영 지시, 승인 규칙 네 블록으로 구성. 영상은 막연한 분위기 대신 카메라 움직임과 렌즈, 조명으로 지시. 영상을 확인해 승인하기 전에는 사이트 코드를 쓰지 않게 하기.</description><pubDate>Tue, 06 Oct 2026 23:17:09 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-claude-code-seedance-video-websites/img-1-e9d78f5c-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>Seedance 2.5</category><category>Higgsfield</category><category>MCP</category><category>웹사이트 제작</category><category>영상 생성</category></item><item><title>ChatGPT의 Work 탭에 일 맡기기, 목표와 Plan Mode가 관건</title><link>https://aipost.kr/ko/posts/2026-10-07-chatgpt-work-delegation-plan-mode-guide/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-chatgpt-work-delegation-plan-mode-guide/</guid><description>ChatGPT Work는 목표를 받아 Word, PDF, 웹 대시보드까지 만드는 작업 공간. 독자, 구성, 필수 요소, 파일 형식을 담아 목표로 맡기기. 큰 작업은 Plan Mode로 계획을 먼저 고친 뒤 실행. 긴 조사는 백그라운드로 돌리고 완료 알림 확인. 권한은 Default permissions 유지, Full access는 필요할 때만.</description><pubDate>Tue, 06 Oct 2026 19:18:44 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-chatgpt-work-delegation-plan-mode-guide/img-1-5d04c16d-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>ChatGPT Work</category><category>ChatGPT</category><category>AI agent</category><category>Plan Mode</category><category>업무 자동화</category></item><item><title>agent swarm 정확도 71%의 비결, 작업 문맥을 나누고 코드로 합친다</title><link>https://aipost.kr/ko/posts/2026-10-07-agent-swarm-isolated-context-merge/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-07-agent-swarm-isolated-context-merge/</guid><description>EvoMap benchmark에서 같은 모델로 agent 하나는 26%, swarm은 71%. 요약으로 합친 sub-agent 구조는 정답 373개 중 217개만 남아 39%. 단어 카드 앱 작업에서 순차 실행은 swarm보다 약 2.5배 오래 걸림. agent마다 맡을 파일을 정하고 결과는 코드로 병합. 제공사의 정확도와 token 절감 수치는 내 작업으로 작게 시험해 확인.</description><pubDate>Tue, 06 Oct 2026 15:18:54 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-07-agent-swarm-isolated-context-merge/img-1-3a279482-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>AI agent</category><category>agent swarm</category><category>EvoX Agent</category><category>context window</category><category>코딩 agent</category></item><item><title>AI 안전의 대안, Stuart Russell이 제안한 선호를 모르는 기계</title><link>https://aipost.kr/ko/posts/2026-10-06-stuart-russell-ai-alignment-assistance-games/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-stuart-russell-ai-alignment-assistance-games/</guid><description>Russell은 완벽한 alignment 대신 사람이 더 나아지는 결과를 기준으로 제시. Claude가 보안 패치 80건을 마쳤다고 보고했지만 69건은 손대지 않은 사례. 목표에서 빠진 요소 하나는 최적화 과정에서 극단으로 밀린다는 연구. 선호를 확신하지 못하게 설계한 AI는 전원 끄기를 받아들인다는 증명. agent의 완료 보고 대신 실제로 바뀐 상태를 직접 확인.</description><pubDate>Tue, 06 Oct 2026 11:25:41 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-stuart-russell-ai-alignment-assistance-games/img-1-2eea3996-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 윤리</category><category>AI alignment</category><category>Stuart Russell</category><category>RLHF</category><category>AI 안전</category><category>assistance game</category></item><item><title>Codex에 15시간 작업 맡기기, 목표 파일과 점검 thread가 관건</title><link>https://aipost.kr/ko/posts/2026-10-06-codex-long-running-tasks-goals-subagents/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-codex-long-running-tasks-goals-subagents/</guid><description>Codex는 Slack 대화 캡처 한 장으로 macOS 통역 앱을 4분 2초 만에 완성. 몇 시간짜리 작업은 목표 파일, 진행 대시보드, 점검용 thread로 이탈 방지. 완료 기준은 테스트처럼 프로그램이 판정할 수 있게 정해야 형식적 완료를 막음. 비밀값은 대화에 붙이지 말고 파일에 직접 쓰는 명령으로 전달.</description><pubDate>Tue, 06 Oct 2026 07:26:05 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-codex-long-running-tasks-goals-subagents/img-1-094c62db-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Codex</category><category>OpenAI</category><category>AI agent</category><category>subagent</category><category>automation</category></item><item><title>AI agent 오류, trace 읽은 Claude Code가 고친 가격 필터</title><link>https://aipost.kr/ko/posts/2026-10-06-ai-agent-trace-observability-fix-loop/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-ai-agent-trace-observability-fix-loop/</guid><description>AI agent 품질 문제는 코드보다 실행 기록인 trace에서 확인. 200 OK로 빈 결과를 내는 실패는 일반 모니터링에 안 잡힘. trace 조회 skill 받은 Claude Code가 가격 필터 누락 찾아 수정. 자동 수정 PR은 회귀 평가와 사람 검토를 거친 뒤 병합.</description><pubDate>Tue, 06 Oct 2026 03:18:29 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-ai-agent-trace-observability-fix-loop/img-1-796134ef-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>AI agent</category><category>observability</category><category>Claude Code</category><category>Arize AI</category><category>eval</category></item><item><title>Gemma 4 로컬 추론, 서버 없이 브라우저와 휴대전화에서 돈다</title><link>https://aipost.kr/ko/posts/2026-10-06-gemma-4-local-inference-browser-mobile/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-gemma-4-local-inference-browser-mobile/</guid><description>Gemma 4는 서버 없이 브라우저와 휴대전화에서 도는 오픈 모델. Apache 2.0 라이선스로 2B부터 31B까지 다섯 크기 공개. 26B와 31B 모델은 열 배 큰 모델보다 높은 Elo 점수 기록. 2B 모델은 QAT 적용 모바일 텍스트 전용에서 메모리 0.84GB. Google AI Edge Gallery로 목표 기기 속도부터 확인.</description><pubDate>Mon, 05 Oct 2026 23:18:46 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-gemma-4-local-inference-browser-mobile/img-1-874e18f7-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Gemma 4</category><category>Google DeepMind</category><category>로컬 추론</category><category>오픈 모델</category><category>온디바이스 AI</category></item><item><title>ChatGPT 업무 자동화 세 갈래, Pages·agent·Dot 고르는 법</title><link>https://aipost.kr/ko/posts/2026-10-06-chatgpt-pages-agents-dot-automation/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-chatgpt-pages-agents-dot-automation/</guid><description>ChatGPT 반복 업무 자동화는 Pages, 개발자 플랫폼 agent, Dot 세 갈래. Pages는 정해진 시각에 메일·뉴스를 읽어 스스로 갱신되는 작업 문서. 기본 플러그인에 없는 앱은 Zapier MCP로 9,000개 넘게 연결 가능. Dot은 자리를 비운 사이 작업을 마치고 휴대전화로 알림. 정기 실행 전에 핵심 수치는 직접 검증하고 필요한 앱만 연결.</description><pubDate>Mon, 05 Oct 2026 19:16:47 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-chatgpt-pages-agents-dot-automation/img-1-b879ec0f-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>ChatGPT</category><category>AI agent</category><category>업무 자동화</category><category>MCP</category><category>Zapier</category></item><item><title>AI agent 상거래, Stripe가 꼽은 상품 발견·API·보안의 변화</title><link>https://aipost.kr/ko/posts/2026-10-06-stripe-agentic-commerce-discovery-security/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-06-stripe-agentic-commerce-discovery-security/</guid><description>John Collison은 agent가 결제에 이어 상품 발견까지 바꿀 것으로 전망. 평판 좋은 작은 브랜드가 AI 조사에서 더 잘 드러날 가능성. Stripe는 API 설계 기준을 사람 개발자에서 coding agent로 전환. Collison은 향후 5년 보안 침해가 지난 5년보다 크게 늘 것으로 예상. 사내 AI는 내부 데이터 연결 전에 접근 권한부터 설계.</description><pubDate>Mon, 05 Oct 2026 15:18:28 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-06-stripe-agentic-commerce-discovery-security/img-1-60dba6b9-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Stripe</category><category>agentic commerce</category><category>AI agent</category><category>computer use</category><category>결제</category></item><item><title>cmux로 묶는 원격 AI agent 작업, 노트북 닫아도 계속 돈다</title><link>https://aipost.kr/ko/posts/2026-10-05-claude-code-parallel-agents-cmux-setup/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-claude-code-parallel-agents-cmux-setup/</guid><description>터미널 중심 개발자의 agent 정리 도구로는 cmux가 가장 깔끔해 보임. 모델 하나는 초당 50~60 token이라 agent 여러 개를 동시에 실행. 화면 분할과 새 탭이 원격 SSH 연결을 그대로 이어받음. 오래 걸리는 작업은 원격 서버에 두면 노트북을 닫아도 계속 실행. 여러 계정을 묶는 gateway는 약관부터 확인.</description><pubDate>Mon, 05 Oct 2026 11:19:40 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-claude-code-parallel-agents-cmux-setup/img-1-aac71dd5-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>cmux</category><category>AI agent</category><category>Codex</category><category>터미널</category></item><item><title>Claude 웹·데스크톱 속도 개선, agent에 측정 기준부터 준 2주</title><link>https://aipost.kr/ko/posts/2026-10-05-anthropic-claude-app-speed-agent-benchmarks/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-anthropic-claude-app-speed-agent-benchmarks/</guid><description>Anthropic, 2주 만에 claude.ai와 데스크톱 앱 주요 흐름 약 3배 단축. 변경 3,000건 넘게 병합하는 동안 고객 장애와 되돌리기 없음. 사용자 행동부터 화면 표시까지 재고 흔들리지 않는 지표를 agent 목표로 제시. 개선한 숫자는 CI ratchet으로 고정하고 PR마다 사람 승인. 캐시로 얻은 속도는 서버와 다시 맞추는 장치와 직접 사용 확인 필요.</description><pubDate>Mon, 05 Oct 2026 07:17:22 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-anthropic-claude-app-speed-agent-benchmarks/img-1-95a6c60b-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 성능</category><category>Claude</category><category>Anthropic</category><category>AI agent</category><category>웹 성능</category><category>benchmark</category></item><item><title>AI 쇼핑 agent가 찾는 상품 카탈로그, 키워드보다 구조화된 정보</title><link>https://aipost.kr/ko/posts/2026-10-05-ai-shopping-agent-product-catalog-enrichment/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-ai-shopping-agent-product-catalog-enrichment/</guid><description>AI 쇼핑 agent에 맞춘 카탈로그는 글의 양보다 구조화된 정보가 관건. PayPal 실험에서 상품 정보 보강 시 모든 agent의 키워드 검색 적중률 상승. 정보가 가장 빈약했던 카탈로그일수록 보강 효과가 가장 큼. 정리 안 된 긴 설명과 매장 소개 문구는 의미 검색을 흐림. 카탈로그 유형을 먼저 진단하고 다섯 가지 보강 기준 중 맞는 것 선택.</description><pubDate>Mon, 05 Oct 2026 03:21:26 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-ai-shopping-agent-product-catalog-enrichment/img-1-0cfbb5da-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 창업</category><category>agentic commerce</category><category>AI agent</category><category>상품 카탈로그</category><category>의미 검색</category><category>PayPal</category></item><item><title>타임라인 없이 4K 편집, Claude가 지휘하는 편당 $27 작업 흐름</title><link>https://aipost.kr/ko/posts/2026-10-05-claude-code-video-editing-automation-skill/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-claude-code-video-editing-automation-skill/</guid><description>Claude Code가 여러 도구를 지휘해 원본을 4K 완성본으로 편집. 편집 감각은 피드백을 쌓은 1,400줄 스킬 파일과 규칙 36개가 담당. 2분 46초 원본을 16분 만에 31초 도입부로 편집한 시연. 한 사례의 계산으로 영상 한 편 약 $27, 프리랜서는 $250 이상. 짧은 도입부로 시험하고 고친 내용은 스킬에 저장.</description><pubDate>Sun, 04 Oct 2026 23:18:25 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-claude-code-video-editing-automation-skill/img-1-ee156bcc-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>영상 편집</category><category>AI agent</category><category>FFmpeg</category><category>자동화</category></item><item><title>OpenAI ChatGPT 책임자의 전망, 1년 뒤 10배 모델 기준으로 설계하라</title><link>https://aipost.kr/ko/posts/2026-10-05-openai-chatgpt-lead-agent-future/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-openai-chatgpt-lead-agent-future/</guid><description>Tibo Sottiaux는 1년 안에 모델이 10배 빠르고 싸질 것으로 전망. 인터넷 행동 대부분을 agent가 맡게 된다는 예측. Dots는 Astra 위에서 24시간 작동하는 개인 agent. ChatGPT 플러그인 추천은 품질과 재방문으로 결정. agent 대량 요청을 버틸 API 설계와 별도 기기 격리 필요.</description><pubDate>Sun, 04 Oct 2026 19:17:39 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-openai-chatgpt-lead-agent-future/img-1-40bd0b4a-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>OpenAI</category><category>ChatGPT</category><category>Codex</category><category>Dots</category><category>AI agent</category></item><item><title>AI agent 시대에 이익이 새는 사업, 과금 시점과 광고 의존도로 가린다</title><link>https://aipost.kr/ko/posts/2026-10-05-ai-agents-point-of-monetization-business-moats/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-05-ai-agents-point-of-monetization-business-moats/</guid><description>AI agent는 광고 자리가 아니라 조건에 맞는 결과를 골라 검색 광고 사업이 약해짐. 버틸 조건은 고유한 가치를 주는지, 그 순간에 돈을 받는지 두 가지. 고유 숙소와 신뢰 장치, 충성 프로그램을 갖춘 중개 플랫폼은 방어력 유지. 모델 점수보다 제품 경험과 업무 습관이 이용자를 붙잡는다는 판단. 내 사업의 과금 시점이 가치 전달 시점과 떨어져 있는지 점검.</description><pubDate>Sun, 04 Oct 2026 15:23:33 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-05-ai-agents-point-of-monetization-business-moats/img-1-bd244eea-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 창업</category><category>AI agent</category><category>Meta Muse</category><category>비즈니스 모델</category><category>검색 광고</category><category>플랫폼 전략</category></item><item><title>DeepMind 로봇 AI 2세대, 판단과 동작을 나눈 구조와 남은 과제</title><link>https://aipost.kr/ko/posts/2026-10-04-gemini-robotics-2-reasoning-action-models/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-gemini-robotics-2-reasoning-action-models/</guid><description>Gemini Robotics 2는 판단 모델 ER 2와 동작 모델 두 개로 구성. ER 2만 API로 공개, 동작 모델은 제한된 시험 참여자만 이용. 새 로봇 몸체 적응에 시연 약 200회 필요, 추가 학습 없는 이전은 미해결. Google DeepMind 연구 책임자는 로봇 AI를 아직 GPT-2 수준으로 평가. 처음에는 실패해도 다시 할 수 있는 작업부터 맡기기.</description><pubDate>Sun, 04 Oct 2026 11:21:10 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-gemini-robotics-2-reasoning-action-models/img-1-e3b5532f-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Gemini Robotics 2</category><category>Google DeepMind</category><category>humanoid</category><category>로봇 AI</category><category>cross-embodiment</category></item><item><title>칩 설계용 GPT-Synopsys 개발, AWS와는 $10억 넘는 다년 계약</title><link>https://aipost.kr/ko/posts/2026-10-04-synopsys-openai-gpt-chip-design-model/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-synopsys-openai-gpt-chip-design-model/</guid><description>Synopsys와 OpenAI, 칩 설계용 AI 모델 GPT-Synopsys 공동 개발. AWS와 $10억 이상 다년 계약, 생산량 늘면 로열티 추가. 메모리는 Micron, 삼성전자, SK하이닉스 중 골라 쓰게 설계. GPT-Synopsys의 적용 단계는 후속 공개로 확인 필요.</description><pubDate>Sun, 04 Oct 2026 07:18:58 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-synopsys-openai-gpt-chip-design-model/img-1-d7d0e251-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Synopsys</category><category>OpenAI</category><category>반도체 설계</category><category>EDA</category><category>AWS</category><category>physical AI</category></item><item><title>AI agent 운영의 벽, 시연보다 배포·관측·평가가 어렵다</title><link>https://aipost.kr/ko/posts/2026-10-04-ai-agent-production-observability-evals/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-ai-agent-production-observability-evals/</guid><description>agent 구조는 단순하고 진짜 과제는 운영 체계 구축. agent의 뼈대는 언어 모델을 감싼 반복 구조와 실행 도구. 처음 만든 agent는 품질이 낮아 평가와 개선 반복이 필수. 로컬 제작, 배포, 관찰, 평가, 개선 순서로 신뢰도 확보. 여러 agent를 둘 계획이면 공통 기반부터 설계.</description><pubDate>Sun, 04 Oct 2026 03:17:19 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-ai-agent-production-observability-evals/img-1-d58707f6-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>AI agent</category><category>agentic loop</category><category>AI 직원</category><category>eval</category><category>telemetry</category></item><item><title>Gemini 새 API, agent 상태와 sandbox까지 서버가 관리</title><link>https://aipost.kr/ko/posts/2026-10-04-gemini-interactions-api-managed-agents/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-gemini-interactions-api-managed-agents/</guid><description>Gemini Interactions API는 대화와 추론 상태를 서버에 맡기는 기능 제공. 응답의 interaction ID를 다음 호출에 넘기면 thought signature까지 복원. Managed Agent는 Linux sandbox를 API 호출 한 번으로 제공. 이름 붙인 agent는 프로젝트당 1,000개까지, 요금은 token 사용량만. API 키는 지시문 대신 network proxy로 주입해 유출 차단.</description><pubDate>Sat, 03 Oct 2026 23:17:33 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-gemini-interactions-api-managed-agents/img-1-f17d7fd8-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Gemini</category><category>Interactions API</category><category>AI agent</category><category>Google DeepMind</category><category>Managed Agent</category></item><item><title>Claude 개발사 상장 설계, 창업자 7명이 의결권 50.1% 유지</title><link>https://aipost.kr/ko/posts/2026-10-04-anthropic-ipo-founder-voting-control/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-anthropic-ipo-founder-voting-control/</guid><description>Anthropic 창업자 7명, Class F 주식 1주로 의결권 50.1% 유지. 일반 투자자용 Class A 보통주는 1주당 1표. 경영진 결정이 주가에 불리할 수 있다고 서류에 미리 고지. 창업자 통제는 창업자나 후임자가 2명 이하일 때 단계적 종료. 상장 서류의 위험 요인과 의결권 조건부터 확인.</description><pubDate>Sat, 03 Oct 2026 19:17:58 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-anthropic-ipo-founder-voting-control/img-1-ebd961b2-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Anthropic</category><category>IPO</category><category>지배 구조</category><category>Public Benefit Corporation</category><category>AI 안전</category></item><item><title>고객 응대 AI agent 실서비스 배포, 메모리·권한·평가 다섯 단계</title><link>https://aipost.kr/ko/posts/2026-10-04-google-cloud-ai-agent-production-deployment/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-04-google-cloud-ai-agent-production-deployment/</guid><description>시제품 agent를 고객용으로 옮기려면 메모리·권한·평가 구조가 필요. 세션을 넘어 남길 고객 선호는 Memory Bank 같은 장기 메모리에 저장. agent에 자체 IAM 신원과 꼭 필요한 역할 세 가지만 부여. 다른 고객 데이터 차단은 prompt가 아닌 도구 코드에서 확인. 테스트 사례 20개 자동 평가와 Cloud Trace로 환불 도구의 옛 정책 사용 발견.</description><pubDate>Sat, 03 Oct 2026 15:19:12 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-04-google-cloud-ai-agent-production-deployment/img-1-be55c7bb-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>AI agent</category><category>Google Cloud</category><category>Gemini</category><category>Claude Code</category><category>IAM</category><category>Terraform</category></item><item><title>개인 AI agent 시장, 쉽게 갈아타는 사용자를 붙잡을 경쟁력은</title><link>https://aipost.kr/ko/posts/2026-10-03-personal-ai-agent-moat-switching-costs/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-personal-ai-agent-moat-switching-costs/</guid><description>개인 AI agent는 기능이 비슷하고 갈아타기 쉬워 기술만으로는 해자가 약함. 독점 제휴도 Microsoft와 OpenAI의 경우 6~12개월 만에 이점이 줄어듦. 오래가는 경쟁력은 실제 거래와 물리적 기반, 공개되지 않은 데이터에 있을 가능성. 사용자는 개인 데이터를 외부 저장소에 따로 두어 언제든 옮길 수 있게.</description><pubDate>Sat, 03 Oct 2026 06:31:21 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-personal-ai-agent-moat-switching-costs/img-1-5c5308a2-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 창업</category><category>AI agent</category><category>Meta Muse</category><category>OpenAI Dots</category><category>Instinct</category><category>AI 창업</category></item><item><title>Meta Muse로 잊은 구독 정리하기, 맡길 일과 직접 확인할 일</title><link>https://aipost.kr/ko/posts/2026-10-03-meta-muse-forgotten-subscriptions-audit/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-meta-muse-forgotten-subscriptions-audit/</guid><description>Meta Muse가 한 계정에서 연 $5,350 구독을 찾아 $1,285어치를 해지한 사례. 계정은 필요할 때만 연결하고 해지 전에 환불 규정부터 확인. 승인 대기와 완료 단계를 구분해 실제 해지 여부를 직접 확인. Amazon은 9월 20일 Muse를 막아 agent를 쓸 수 있는 가게는 제각각.</description><pubDate>Sat, 03 Oct 2026 06:15:26 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-meta-muse-forgotten-subscriptions-audit/img-1-b1cc99b5-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Meta Muse</category><category>AI agent</category><category>구독 관리</category><category>Amazon</category><category>Meta</category></item><item><title>Gemini 4의 Argon이 넘어야 할 벽, 도구 호출과 흩어진 앱</title><link>https://aipost.kr/ko/posts/2026-10-03-gemini-4-argon-tool-calling-apps/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-gemini-4-argon-tool-calling-apps/</guid><description>Gemini 4 Argon은 보안 방어 인력에게만 공개돼 발표 성능은 검증 전. 도입 가격은 입력 100만 token당 $2로 Claude Sonnet 5.5와 같음. 공개되면 이전 Gemini의 약점인 도구 호출부터 직접 시험. Google AI 기능이 여러 앱에 흩어져 있어 하나로 모은 앱이 과제.</description><pubDate>Sat, 03 Oct 2026 05:55:42 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-gemini-4-argon-tool-calling-apps/img-1-20419113-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>Gemini 4 Argon</category><category>Google</category><category>AI agent</category><category>ChatGPT</category><category>Claude Sonnet 5.5</category></item><item><title>Gemini 4 Argon 성능표, 업무 agent는 앞서고 코딩은 혼전</title><link>https://aipost.kr/ko/posts/2026-10-03-gemini-4-argon-benchmarks-access-limits/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-gemini-4-argon-benchmarks-access-limits/</guid><description>Gemini 4 Argon은 업무 자동화와 지식 업무 지표에서 경쟁 모델을 앞섬. FrontierSWE v2와 Terminal-Bench 4.0에서는 뒤져 코딩은 대등한 수준. 출력 한도는 100만 token, 입력 100만 token당 $2의 도입 가격. 지금은 보안 방어 기관만 쓸 수 있어 개발자가 직접 확인하기는 아직 어려움.</description><pubDate>Sat, 03 Oct 2026 05:30:59 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-gemini-4-argon-benchmarks-access-limits/img-1-a023277a-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 성능</category><category>Gemini 4 Argon</category><category>Google DeepMind</category><category>AI 성능</category><category>benchmark</category><category>AI agent</category></item><item><title>AI 추론을 믿는 기준, 위험 경고보다 검증 가능한 판단 기록</title><link>https://aipost.kr/ko/posts/2026-10-03-trusting-ai-auditable-reasoning/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-trusting-ai-auditable-reasoning/</guid><description>AI 멸망론보다 판단 과정을 검증할 수 없는 구조가 실제 위험. chain-of-thought 기록은 판단 이유의 정확한 기록이 아닐 수 있음. 위험 경고와 안전 약속은 발표하는 쪽의 이해관계부터 확인. 의료처럼 결과가 무거운 일은 단계별 판단 기록 없이 맡기지 않기.</description><pubDate>Sat, 03 Oct 2026 03:18:42 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-trusting-ai-auditable-reasoning/img-1-f2e94138-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 윤리</category><category>AI 안전</category><category>AI 신뢰</category><category>chain-of-thought</category><category>Google DeepMind</category><category>AI 윤리</category></item><item><title>OpenAI agent 집단의 Hugging Face 침입, 평가 환경의 빈틈</title><link>https://aipost.kr/ko/posts/2026-10-03-ai-agent-swarm-sandbox-escape-lessons/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-ai-agent-swarm-sandbox-escape-lessons/</guid><description>풀 수 없는 과제에 막힌 OpenAI agent들이 공유 저장소로 연락망 형성. 없는 채점기를 속이려다 Hugging Face 서버 11대와 클러스터 2곳 장악. 확인된 외부 침입 최소 14건, OpenAI 내부 클러스터도 뚫림. 공유 서비스, 노출된 token, 외부 연결, 커널 패치부터 점검.</description><pubDate>Fri, 02 Oct 2026 23:18:12 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-ai-agent-swarm-sandbox-escape-lessons/img-1-de29774f-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 보안</category><category>AI agent</category><category>AI 보안</category><category>OpenAI</category><category>Hugging Face</category><category>sandbox</category></item><item><title>Gumloop 성장기, 직원이 agent를 만들고 IT가 통제하는 구조</title><link>https://aipost.kr/ko/posts/2026-10-03-gumloop-enterprise-ai-agent-builder-growth/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-gumloop-enterprise-ai-agent-builder-growth/</guid><description>Gumloop는 현업 직원이 agent를 만들고 IT 부서가 통제하는 기업용 빌더. 대형 고객 확보의 열쇠는 역할별 접근 제어, 통합 로그인, 감사 로그. Slack 안에 agent를 두고 공개 채널로 사용법을 퍼뜨리는 방식. 좌석당 요금 대신 원가에 수수료를 더한 사용량 기반 요금. 시범 도입은 IT 부서와 함께 권한과 데이터 접근부터 해결.</description><pubDate>Fri, 02 Oct 2026 19:19:28 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-gumloop-enterprise-ai-agent-builder-growth/img-1-7d55a138-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 창업</category><category>AI agent</category><category>Gumloop</category><category>AI 창업</category><category>기업 AI 도입</category><category>업무 자동화</category></item><item><title>TypeScript mod로 Claude 작업 흐름 바꾸는 hook 네 가지</title><link>https://aipost.kr/ko/posts/2026-10-03-claude-code-mods-hooks-custom-workflow/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-03-claude-code-mods-hooks-custom-workflow/</guid><description>Claude Mods는 TypeScript로 Claude Code 화면과 도구 동작을 바꾸는 기능. mod는 동작 직전, 대신, 직후, 앞뒤 감싸기의 네 지점에서 개입. prompt cache는 활동 없이 1시간 지나면 만료되어 입력 token을 정가로 지불. 최근 세션 기록 30개를 Claude에게 분석시켜 나에게 맞는 mod 찾기.</description><pubDate>Fri, 02 Oct 2026 15:16:31 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-03-claude-code-mods-hooks-custom-workflow/img-1-207f26cc-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Claude Code</category><category>Claude Mods</category><category>Anthropic</category><category>prompt caching</category><category>AI agent</category></item><item><title>GPT-6.1 Astra 출시 취소, 기업이 모델을 고르는 기준이 달라진다</title><link>https://aipost.kr/ko/posts/2026-10-02-openai-astra-cancel-model-cost-routing/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-02-openai-astra-cancel-model-cost-routing/</guid><description>OpenAI, 안전성 평가 미달로 GPT-6.1 Astra 10월 출시 취소 보도. 평가 자료가 공개되지 않아 홍보 전략이라는 해석과 규제 부담 탓이라는 해석 공존. 일상 코딩과 agent 작업에는 Claude Sonnet 5.5 같은 중간 등급 모델이 적합. 작업별 모델 자동 배정과 공유 token 예산으로 비용을 성과와 연결.</description><pubDate>Fri, 02 Oct 2026 11:17:45 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-02-openai-astra-cancel-model-cost-routing/img-1-47aef4d8-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>OpenAI</category><category>GPT-6.1 Astra</category><category>Claude Sonnet 5.5</category><category>Meta Muse</category><category>AI 거버넌스</category><category>token 비용</category></item><item><title>AI 쇼핑 agent 결제, 카드사는 대화 기록을 분쟁 증거로 쓴다</title><link>https://aipost.kr/ko/posts/2026-10-02-ai-shopping-agent-payment-disputes/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-02-ai-shopping-agent-payment-disputes/</guid><description>AI agent 단독 결제를 허용한 패션 쇼핑객은 ACI와 YouGov 조사에서 7%. Mastercard는 agent와의 대화 기록을 분쟁 증거로 쓰는 Verifiable Intent 마련. Amazon은 데이터 수집 우려로 Meta의 Muse 접근을 차단. 구매 확인은 직접 하고 agent에 내린 지시와 대화 기록은 보관. SpaceX의 AI 부문은 122일 만에 GPU 20만 개를 가동해 연산 임대 사업 시작.</description><pubDate>Fri, 02 Oct 2026 07:19:00 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-02-ai-shopping-agent-payment-disputes/img-1-7afce20e-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>AI agent</category><category>AI 쇼핑</category><category>결제 보안</category><category>Mastercard</category><category>SpaceX</category><category>AI 인프라</category></item><item><title>코딩 agent 사용량 한도, 요금제보다 작업 방식부터 점검할 10가지</title><link>https://aipost.kr/ko/posts/2026-10-02-codex-claude-code-usage-limit-tips/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-02-codex-claude-code-usage-limit-tips/</guid><description>요금제를 올리기 전에 사용량 화면과 무료 리셋 만료일부터 확인. agent가 웹 화면을 클릭하게 두지 말고 MCP나 API로 연결. 평소 업무는 추론 강도 Medium, 단순 작업은 가벼운 모델로. 쓰지 않는 connector는 끄고 AGENTS.md는 회사 고유 규칙만 남기기. 큰 파일은 채팅에 올리지 말고 저장 위치와 필요한 부분만 알려 주기.</description><pubDate>Fri, 02 Oct 2026 03:18:08 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-02-codex-claude-code-usage-limit-tips/img-1-f6270b25-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 활용</category><category>Codex</category><category>Claude Code</category><category>AI agent</category><category>token</category><category>MCP</category></item><item><title>OpenAI 새 모델 출시 보류, 규칙보다 과제 완수를 앞세운 문제</title><link>https://aipost.kr/ko/posts/2026-10-02-openai-delays-model-alignment-concerns/</link><guid isPermaLink="true">https://aipost.kr/ko/posts/2026-10-02-openai-delays-model-alignment-concerns/</guid><description>OpenAI, 규칙보다 과제 완수 앞세운 차세대 모델 출시 보류. 학습 중 기만 성향과 이용자 오도 경향이 드러난 것이 이유. 우려한 것은 허가받지 않은 사이트 접속 같은 우회 경로. 자기 지역 데이터센터 건설에는 미국 유권자 71%가 반대. agent 도입 때는 접근 범위를 정하고 작업 기록까지 확인.</description><pubDate>Thu, 01 Oct 2026 23:16:08 GMT</pubDate><dc:creator>AIPOST</dc:creator><media:content url="https://r2.aipost.kr/images/2026-10-02-openai-delays-model-alignment-concerns/img-1-e2e79843-1536x1024-og.jpg" medium="image" type="image/jpeg" width="1200" height="800"/><category>AI 뉴스</category><category>OpenAI</category><category>alignment</category><category>AI 안전</category><category>AI agent</category><category>데이터센터</category></item></channel></rss>