GitLab is building next-generation source control for agentic workloads, combining machine-scale performance with context, security, and governance to deliver trusted software.
AI agents are fundamentally changing the economics of software development. As the cost of producing code falls, the volume of changes moving through the software factory is accelerating beyond what infrastructure designed for human developers was built to handle.
WHAT PROBLEM THIS SOLVES FOR LARGE-SCALE AGENT ENVIRONMENTS
GitLab’s next-gen SCM is designed to support thousands of concurrent agents working against large repositories. Traditional Git workflows require clients to download substantial repository data, creating bottlenecks when agents operate at scale. TechStaged has also covered GitLab arms developers with cost controls for AI agents via Duo Agent Platform and Orbit.
- Preserves Git compatibility
- Introduces stateless compute that scales horizontally against shared object storage
- Offers agent-native interfaces so agents fetch only the needed files and history without full repository clones
ARCHITECTURE AND EARLY PERFORMANCE SIGNALS
The architecture preserves Git compatibility while modernizing the backend. It relies on stateless compute and shared object storage, with agent-native interfaces to minimize data transfer.
- Workloads in early internal tests showed up to 50x faster completion
- Up to 2x fewer tokens
- Up to 1,000x less network traffic under specific test conditions
ANTHROPIC COLLABORATION AND SCION DEMONSTRATION
Anthropic demonstrated the importance of agent-scale source control through Scion, a system designed for enormous monorepos where most clients are agents that are always online. Scion is not a general-purpose replacement for Git and requires its own supporting CI, review, and search infrastructure.
GitLab’s approach aims to scale for broader enterprise use while maintaining Git compatibility and integrating with existing software lifecycles.
- June announcement at GitLab Transcend with Anthropic’s Nick G.
- August updates on architecture and early performance results
- Scion illustrates why specialized, agent-focused SCM matters in practice
DATA SOVEREIGNTY, DEPLOYMENT OPTIONS, AND THE ROAD AHEAD
The same architecture is designed to run on GitLab.com and on S3-compatible storage, including in air-gapped data centers, keeping repository data within customer environments while enabling agent-scale workloads.
Next-gen SCM is in private beta as GitLab validates performance against customers’ repositories and workloads.
- Data stays within customer infrastructure
- Support for air-gapped environments
- Early access available for select customers
RELATED COVERAGE
- GitLab arms developers with cost controls for AI agents via Duo Agent Platform and Orbit
- GPT-6 Astra on GitLab Duo Agent Platform delivers faster runs and fewer tokens
- GitLab Credits Bring Usage-Based Pricing to Duo Agent Platform
- GitLab Duo Self-Hosted gains Bring-Your-Own-Model support with Microsoft Foundry
- Developer Tools articles
SOURCES
- GitLab: Scaling Git for the agentic software factory Published · Primary source
