Anthropic’s Scion Shows Why AI Agents Are Forcing a Rethink of Source Control

Описание и применение

Anthropic разработал Scion, систему контроля версий, оптимизированную для работы с большим количеством изменений, генерируемых ИИ‑агентами, что позволило обрабатывать более 30 000 слияний в день, сократить время CI‑задач до секунд и уменьшить инфраструктурные затраты в 200‑размерном масштабе.

Задача

AI‑assisted development at Anthropic increased daily merge volume from ~2,500 in May 2026 to >30,000 in September 2026, overwhelming Git‑based source‑control, CI, and review workflows.

Результат

Scion enabled Anthropic to handle >30,000 daily merges, reduced CI job times to seconds, achieved 99%+ cache hit rates, cut infrastructure footprint 200×, and gained 25% CPU capacity, while maintaining safety controls.

Рабочий процесс

• 1. Agents generate code changes as changesets.
• 2. Changesets are submitted to Scion’s server‑side commit and merge system.
• 3. Server distributes changes across a distributed cache layer.
• 4. CI pipelines run on cached artifacts, reducing job time.
• 5. Code review via Antique surface, with IAM and semantic constraints.
• 6. Merges are performed server‑side without client pushes.

Доказательства

AI-assisted development pushed its daily merge volume from roughly 2,500 changes in May 2026 to more than 30,000 in September.

Источник

Scion changes several foundational assumptions of a conventional Git-centered workflow. Rather than modeling work around traditional Git commits, it uses changesets.

Источник

The company reports cache hit rates above 99%, as well as CI and job speed improvements from minutes to seconds in relevant workflows.

Источник

Scion’s infrastructure footprint is approximately 200 times smaller than its earlier Git mirroring approach in CPU and memory terms, contributing to an estimated 25% CPU capacity gain.

Источник

Anthropic says a seven-person team built Scion and Antique in five months.

Источник