AI 코딩 도구 9건, 새로 만든 곳은 몇 곳일까요

본문의 각주 번호는 아래 항목과 1:1 로 대응합니다. 정량 주장은 가능한 경우 둘 이상의 출처로 대조했고, 인용문은 원문에서 직접 확인한 것만 실었습니다.

1. 앤트로픽의 하루 코드 병합은 2026년 5월 평균 약 2,500건에서 9월 3만 건 이상으로 늘었고, 앤트로픽은 깃 대신 자체 버전 관리 Scion 을 만들어 모노레포를 다섯 달에 걸쳐 옮겼다고 밝혔다In May 2026, Anthropic averaged around 2,500 code changes merged per day; as of September, we merged over 30,000 per day.https://anthropic.com/research/redesigning-source-control확인 2026-10-26In May 2026, Anthropic averaged around 2,500 code changes merged per day; as of September, we merged over 30,000 per day,https://www.theregister.com/devops/2026/10/13/anthropic-reinvents-source-control-as-claude-floods-the-world-with-code/5302943확인 2026-10-26we built Scion, an in-house version control system, and fully migrated our monorepo from git in five months.https://anthropic.com/research/redesigning-source-control확인 2026-10-26After several months of model prompting and five months of migration, the result is a homemade version control system called Scionhttps://www.theregister.com/devops/2026/10/13/anthropic-reinvents-source-control-as-claude-floods-the-world-with-code/5302943확인 2026-10-26
   ↩
2. 앤트로픽이 든 이유는 깃이 사용자가 오프라인에서 저장소 전체 사본을 갖는다고 전제하는데 앤트로픽의 사용자는 거의 다 늘 연결된 에이전트와 작업한다는 것이다Git assumes that clients work offline and need a complete copy of the repository.https://anthropic.com/research/redesigning-source-control확인 2026-10-26we were using git as a centralized system with one authoritative copy, but nearly all of our clients were agents and jobs that are online by definition.https://anthropic.com/research/redesigning-source-control확인 2026-10-26Why did battle-tested git fail here? Because from the day git was created, its core user was a human.https://yage.ai/share/anthropic-scion-agent-scale-infra-en-20261018.html확인 2026-10-26
   ↩
3. 앤트로픽이 밝힌 효과는 기반 시설이 이전 깃 미러 대비 CPU·메모리 기준 200배 작아졌고, 체크아웃이 분 단위에서 초 단위가 됐고, CI 에서 CPU 용량 25% 를 아꼈다는 것이다(회사 자체 측정)Our Scion infrastructure footprint is now 200x smaller than our previous git mirroring infrastructure in terms of CPU and memory requirements.https://anthropic.com/research/redesigning-source-control확인 2026-10-26CI alone realized a 25% CPU capacity win.https://anthropic.com/research/redesigning-source-control확인 2026-10-26source control infrastructure shrank by 200x compared to the old git mirror cluster.https://yage.ai/share/anthropic-scion-agent-scale-infra-en-20261018.html확인 2026-10-26CI and research jobs check out code two orders of magnitude faster (in seconds rather than minutes), which frees up a sizable amount of capacity across the fleet; CI alone realized a 25% CPU capacity win.https://anthropic.com/research/redesigning-source-control확인 2026-10-26
   ↩
4. 대규모 저장소 때문에 깃 밖의 중앙 집중식 버전 관리를 쓴 선례가 있다(구글 Piper, 메타 Sapling)Source control for huge monorepos has prior art, notably Google’s Piper and Meta’s Saplinghttps://anthropic.com/research/redesigning-source-control확인 2026-10-26Google uses Piper, a monorepo systemhttps://www.theregister.com/devops/2026/10/13/anthropic-reinvents-source-control-as-claude-floods-the-world-with-code/5302943확인 2026-10-26Meta has Sapling.https://www.theregister.com/devops/2026/10/13/anthropic-reinvents-source-control-as-claude-floods-the-world-with-code/5302943확인 2026-10-26
   ↩
5. GitHub 는 에이전트를 덧붙이지 않고 쓰던 GitHub 흐름 안에 넣겠다며 Agent HQ 를 냈고, 에이전트의 변경은 초안 PR 로 올라와 동료 작업처럼 검토된다Agents shouldn’t be bolted on. They should work the way you already work. That’s why we’re making agents native to the GitHub flow.https://github.blog/news-insights/company-news/welcome-home-agents/확인 2026-10-26Agent-generated changes show up as draft pull requests and comments, reviewed the same way you'd review a teammate's work.https://www.i-programmer.info/news/105-artificial-intelligence/18651-github-adds-claude-and-codex-to-agent-hq.html확인 2026-10-26
   ↩
6. Cursor 2.0(2025년 10월)은 파일 대신 에이전트를 중심에 둔 화면으로 바꿨고, SD Times 는 에이전트 최대 8개가 깃 워크트리로 서로 간섭 없이 일할 수 있다고 전했다It’s more focused and designed from the ground up to be centered around agents rather than files.https://cursor.com/blog/2-0확인 2026-10-26With this new interface, up to eight agents can work in parallel, using git worktrees and remote trees to prevent them from interfering with each other.https://sdtimes.com/ai/cursor-2-0-enables-eight-agents-to-work-in-parallel-without-interfering-with-each-other확인 2026-10-26October 30, 2025https://sdtimes.com/ai/cursor-2-0-enables-eight-agents-to-work-in-parallel-without-interfering-with-each-other확인 2026-10-26
   ↩
7. 구글 Antigravity 는 단순한 편집기에 그치지 않는 에이전트 우선 개발 플랫폼으로 나왔고, 젯브레인스 Air 는 이미 쓰는 에이전트를 붙여 쓰는 별도 시스템이다Antigravity isn't just an editor—it's a development platform that combines a familiar, AI-powered coding experience with a new agent-first interface.https://developers.googleblog.com/build-with-google-antigravity-our-new-agentic-development-platform/확인 2026-10-26The company sees Antigravity as an evolution of the IDE into an agent-first future, with capabilities like browser control and asynchronous interaction patterns.https://sdtimes.com/softwaredev/google-announces-agentic-development-platform-google-antigravity/확인 2026-10-26Works with the agents you already usehttps://www.jetbrains.com/air/확인 2026-10-26Air was created to enable developers to build tools around agents, guide them and fine-tune their outputs.https://sdtimes.com/ai/jetbrains-previews-air-an-agentic-development-environment/확인 2026-10-26
   ↩
8. 클로드 코드 웹판은 작업마다 격리된 샌드박스에서 돌고, 끝나면 저장소에 브랜치를 열고 PR 을 낼 수 있다Every Claude Code task runs in an isolated sandbox environment with network and filesystem restrictions.https://claude.com/resources/articles/claude-code-on-the-web확인 2026-10-26Once it’s done it opens a branch on your repo with its work and can optionally open a pull request.https://simonwillison.net/2025/Oct/20/claude-code-for-web/확인 2026-10-26
   ↩
9. Entire 의 Checkpoints 는 에이전트 작업 기록을 깃 객체 저장소 안에 두고 브랜치 이력에는 남기지 않으며, 창업자 돔케는 에이전트 코딩에서 막히는 곳이 코드를 쓰는 일에서 검토로 옮겨 간다고 했다Checkpoints live in your repository's own git object store, never in your branch's history.https://github.com/entireio/cli확인 2026-10-26Dohmke also said that the bottleneck in agentic coding is shifting from authoring to review. "If you keep that going through the software life cycle, the next thing you do after writing code is reviewing code," he said.https://rcpmag.com/articles/2026/02/12/ex-github-ceo-thomas-dohmke-unveils-entire.aspx확인 2026-10-26Thomas Dohmke, the former CEO of GitHub, has officially launched a new venture named Entire,https://rcpmag.com/articles/2026/02/12/ex-github-ceo-thomas-dohmke-unveils-entire.aspx확인 2026-10-26
   ↩
10. MCP 는 2025년 12월 발표 기준 활성 공개 서버가 1만 개를 넘었고, 같은 날 리눅스 재단은 AGENTS.md 를 6만 개 이상의 오픈소스 프로젝트·에이전트 프레임워크가 쓴다고 밝혔다(채택 수 · 발표 당시 자체 집계)There are now more than 10,000 active public MCP servers, covering everything from developer tools to Fortune 500 deploymentshttps://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation확인 2026-10-26AGENTS.md has already been adopted by more than 60,000 open source projects and agent frameworkshttps://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation확인 2026-10-26MCP has rapidly become the universal standard protocol for connecting AI models to tools, data and applications, with more than 10,000 published MCP servers now covering everything from developer tools to Fortune 500 deployments.https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation확인 2026-10-26A simple, open format for guiding coding agents, used by over 60k open-source projects.https://agents.md확인 2026-10-26THE LINUX FOUNDATION | 09 DECEMBER 2025https://www.linuxfoundation.org/press/linux-foundation-announces-the-formation-of-the-agentic-ai-foundation확인 2026-10-26Dec 9, 2025https://www.anthropic.com/news/donating-the-model-context-protocol-and-establishing-of-the-agentic-ai-foundation확인 2026-10-26
    ↩
11. METR 실험에서 숙련 오픈소스 개발자가 AI 도구를 쓰면 19% 더 오래 걸렸다(2025년 초 도구 · 16명 · 과제 246건)Surprisingly, we find that when developers use AI tools, they take 19% longer than without—AI makes them slower.https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/확인 2026-10-26we recruited 16 experienced developers from large open-source repositories (averaging 22k+ stars and 1M+ lines of code) that they’ve contributed to for multiple years. Developers provide lists of real issues (246 total)https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/확인 2026-10-26Despite thinking they would work 24% quicker, in reality, it took them 19% longer than without AI assistance.https://leaddev.com/velocity/ai-doesnt-make-devs-as-productive-as-they-think-study-finds확인 2026-10-26We view this result as a snapshot of early-2025 AI capabilities in one relevant setting;https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/확인 2026-10-26We have released results that are current as of early 2026, in a continuation of this study. We believe these historical results no longer reflect the current impact of AI models on open-source developer productivity.https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/확인 2026-10-26
    ↩
12. DORA 2025 에서 응답자 90% 가 일에 AI 를 쓰고 80% 넘게 생산성이 올랐다고 믿지만 AI 도입은 배포 안정성과 음의 관계를 보였고, 스택오버플로 2025 조사에서 개발자 46% 가 AI 도구의 답이 맞는지 믿지 않는다AI adoption is near-universal: 90% of survey respondents report using AI at work. More than 80% believe it has increased their productivity.https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report확인 2026-10-26However, AI adoption does continue to have a negative relationship with software delivery stability.https://cloud.google.com/blog/products/ai-machine-learning/announcing-the-2025-dora-report확인 2026-10-26Nearly 90% of technology professionals now use artificial intelligence in their work.https://www.infoq.com/news/2025/09/dora-state-of-ai-in-dev-2025/확인 2026-10-26More developers actively distrust the accuracy of AI tools (46%) than trust it (33%)https://survey.stackoverflow.co/2025/ai확인 2026-10-2646% of the developers surveyed said they do not trust the accuracy of the output from AI toolshttps://www.infoworld.com/article/4031673/ai-use-among-software-developers-grows-but-trust-remains-an-issue-stack-overflow-survey.html확인 2026-10-26
    ↩
13. Faros 가 2025년 6월 기준 1만 명 넘는 개발자·1,255개 팀의 텔레메트리를 분석한 결과(자체 조사) AI 도입이 높은 팀은 PR 병합이 98% 늘었지만 PR 리뷰 시간은 91% 늘었고, 인포큐에 따르면 아고다는 코딩이 원래 병목이 아니었다는 글을 냈다Developers on teams with high AI adoption complete 21% more tasks and merge 98% more pull requests, but PR review time increases 91%, revealing a critical bottleneck: human approval.https://www.faros.ai/blog/ai-software-engineering확인 2026-10-26research by Faros AI analyzing telemetry from over 10,000 developers across 1,255 teams found that teams with high AI adoption completed 21% more tasks and merged 98% more pull requests, yet PR review time increased by 91%.https://www.infoq.com/news/2026/03/agoda-ai-code-bottleneck/확인 2026-10-26Agoda recently published an observation arguing that while AI coding tools have measurably raised individual developer output, the resulting velocity gains at the project level have been surprisingly modest, because coding was never the real bottleneck.https://www.infoq.com/news/2026/03/agoda-ai-code-bottleneck/확인 2026-10-26Versioning note: This version of the report reflects analysis as of June 2025.https://www.faros.ai/blog/ai-software-engineering확인 2026-10-26For each team, we calculated the percent change in metric values between the two quarters with the lowest AI adoption and the two quarters with the highesthttps://www.faros.ai/blog/ai-software-engineering확인 2026-10-26
    ↩