← Back
Pragmatic Engineer

The Pulse: Anthropic builds its own source control system

•13 min read•
#deployment
The Pulse: Anthropic builds its own source control system
Level:Advanced
For:AI Engineers
✦TL;DR

Anthropic’s internal code‑merge rate exploded from ~2,500 PRs per day in May 2026 to >30,000 per day by September, a ten‑fold increase that exposed Git’s scalability limits at massive scale. To handle a 25 GB monorepo, 25 million daily Git checkouts, and thousands of always‑online Claude agents, the company built a custom source‑control system that removes offline‑client assumptions and eliminates the need for hundreds of Git mirrors. Engineers can now sustain near‑infinite merge throughput without the checkout‑and‑mirror bottlenecks that cripple traditional Git deployments.

⚡ Key Takeaways

  • PRs merged per day grew from ~2,500 (May 2026) to >30,000 (September 2026), a 10× increase.
  • Anthropic runs a 25 GB monorepo with ~25 million Git checkouts per day for CI.
  • The custom system discards Git’s offline‑client design, reducing unnecessary data transfer and enabling horizontal scaling.
  • Implementation replaces hundreds of Git mirrors with a unified, always‑online source‑control service.
  • Prerequisite: all developers and Claude agents must remain continuously connected to the corporate network.
💡 Why It Matters

By eliminating Git’s scaling bottlenecks, Anthropic can keep pace with AI‑driven code generation, ensuring that rapid PR turnover does not stall CI pipelines or deployment velocity.

✅ Practical Steps

  1. Deploy Anthropic’s custom source‑control service in place of existing Git mirrors across the organization.
  2. Configure CI pipelines to point to the new service for repository checkouts, eliminating the 25 million daily Git checkout overhead.

Want the full story? Read the original article.

Read on Pragmatic Engineer ↗

More like this

Built for Agents: The Case for Spark Declarative Pipelines

Databricks Blog•#agents

Multi-Turn Reinforcement Learning on Amazon SageMaker AI: A hands-on guide

AWS ML Blog•#agents

Orchestrating SageMaker AI endpoint auto scaling with AWS CDK

AWS ML Blog•#deployment

The science behind TorchNeuron, Amazon’s new PyTorch backend for AWS Trainium chips

Amazon Science•#deployment

EXPLORE AI NEWS

Daily hand-picked stories on LLMs, RAG, agents and production AI — curated for engineers who ship.

BROWSE NEWS

GET THE WEEKLY DIGEST

Join engineers getting the Monday signal-over-noise AI breakdown. No spam, unsubscribe anytime.

LEARN AI ENGINEERING

Curated courses, research papers, repos and tutorials built for engineers leveling up in AI.

START LEARNING