DEVOURED
GitHub's plan for Agents
GitHub COO Kyle Daigle discusses how the platform's infrastructure, designed for human speeds, is straining under a 1400% surge in agent-driven code contributions in 2026, leading to uptime issues.
What: Kyle Daigle, COO and CMO for Developer at Microsoft, reveals that GitHub is experiencing unprecedented load from AI agents, with commits growing from 1 billion in 2025 to a projected 14 billion in 2026. This 14x growth has exposed new, novel permissioning and database scaling challenges, leading to recent uptime problems, as traditional vertical and horizontal scaling methods prove insufficient.
Why it matters: The exponential growth of AI agent activity is fundamentally reshaping core developer platforms and infrastructure, forcing companies like GitHub to completely re-architect systems previously built around human development patterns.
Takeaway: Developers should anticipate ongoing infrastructure challenges as AI agent activity scales, and those building agents should be mindful of the load their tools place on shared platforms. GitHub is investing in new Azure compute and re-architecting databases; users experiencing issues may find more stability in coming months.
Deep dive
- GitHub COO Kyle Daigle notes a 1400% increase in agent-driven code contributions in 2026, pushing total commits from 1 billion in 2025 to an estimated 14 billion in 2026, causing significant infrastructure strain.
- The platform's original architecture was designed for human speeds, making it ill-suited for the exponential, agent-driven growth.
- Recent uptime issues are attributed to new, novel permissioning problems and database scaling challenges in core systems like "MySQL One," rather than simple legacy issues.
- Traditional vertical and horizontal scaling are no longer sufficient; GitHub is undertaking deep re-architecture of 10-15 year old services, including job queuing, and moving to Azure Dev Compute for actions.
- The growth also sees a shift back towards larger monorepos, posing unique performance challenges due to blob size.
- GitHub is investing heavily in increasing CPU capacity for Actions, which has become a general-purpose compute layer beyond CI/CD.
- Daigle emphasizes GitHub's commitment to transparency, sharing more technical details about ongoing fixes and scaling efforts.
- Copilot's evolution includes a new CLI, desktop app, and cloud agents using a unified SDK, moving beyond code completion to encompass broader AI coding agent automation throughout the SDLC.
- Daigle highlights the concept of "ambient AI" – an AI deeply integrated across all personal and work contexts, capable of understanding and acting based on comprehensive information, which he believes is the future of AI assistance.
- Microsoft is heavily invested in "OpenClaw"-like agents, with a CVP dedicated to it, focusing on sandboxing at the OS level (e.g., Windows) to enable secure agent interaction with work assets and user data.
- GitHub is projected to have over 200 million "developers" (GitHub account holders), redefining what constitutes a developer in the AI era.
Decoder
- Monorepo: A software development strategy where all code for many projects is stored in a single repository, often managed by a single team or organization.
- Git Infrastructure Layer: The underlying systems and protocols that manage how Git repositories are stored, accessed, and synchronized across servers, including handling data blobs, commits, and references.
- Job Queuing: A system that manages and prioritizes tasks or jobs (e.g., builds, tests, deployments) to be processed by available workers or resources, often used in CI/CD pipelines.
- Ambient AI: An artificial intelligence system that operates ubiquitously and subtly in the background, continuously processing context from a user's environment and actions to proactively assist them without explicit commands.
- OpenClaw: An open-source project and concept for an AI agent that can interact with a computer using its keyboard, mouse, and screen, essentially giving AI models the ability to "use" a computer like a human.
- CVP (Corporate Vice President): A senior leadership title at Microsoft, indicating a significant role with broad responsibilities across a business division or strategic initiative.
- Sandboxing: A security mechanism for running programs in an isolated environment, restricting their access to system resources to prevent malicious or buggy code from affecting the rest of the system.
- WorkIQ/FoundryIQ: Internal Microsoft context engines that aggregate information across various work contexts (e.g., M365, Slack, email) to help employees ask questions and gain insights without moving data to new tools, designed for enterprise security and compliance.
Original article
Full article content is not available for inline reading.
Explore all 53 articles in today's digest
Fresh dev news, AI-digested daily. Never miss a trend at devoured.fyi