# GitHub’s Growth: 225M Developers, 34.3B Commits — Jay Parikh, EVP of CoreAI, Microsoft

## Executive summary

GitHub is undergoing massive scaling, driven by its growth to 225 million developers and the increasing adoption of AI agents. The platform's next challenge is managing the 'agent workload.' The discussion highlights key architectural shifts, including the development of Copilot Fusion and the Copilot Harness—a meta-abstraction layer for model orchestration. For build engineers, the focus is on platform resilience, multi-model routing, and integrating AI-driven quality gates (like Copilot Code Review) to maintain quality as code generation scales.

## Key takeaways

- Massive Scale and Agent Adoption: GitHub now serves 225 million developers, handling 34.3 billion commits and 1.3 billion PRs last year. Agent usage is rapidly increasing, with 40% of pull requests being agent-created, and total agent growth having increased sixfold in six months.
- Platform Evolution and Infrastructure: The platform is evolving across infrastructure, model, and application layers, including a migration to Azure. The goal is to create a cohesive 'paved path' where context (memory, preferences) remains consistent whether the user is on mobile, desktop, or VS Code.
- AI Orchestration and Model Choice: Copilot Fusion and the Copilot Harness are key innovations. The Harness acts as a meta-abstraction layer, enabling multi-model routing to optimize for quality, speed, and cost by intelligently selecting the best model for a given task.
- Security and Code Quality: To manage increased code generation, GitHub is enhancing security with features like Copilot Code Review and AI scanning technology, which can block secrets from entering the repository. Ongoing work focuses on hardening the platform and limiting package manager exposures (e.g., npm).

## Technical details

- Copilot Fusion & Harness: Copilot Fusion is described as a meta-abstraction layer for model orchestration. The Copilot Harness is the underlying mechanism that supports multi-model routing, allowing developers to optimize tasks by selecting the best model based on performance, cost, and quality.
- Client Continuity: The platform aims to ensure that a single work session maintains context across multiple surfaces (phone, desktop, VS Code) using the same underlying harness, preventing context loss when switching tools or providers.
- Code Abstraction: The discussion questioned the definition of 'code' in the modern era, noting that repositories now contain much more than just executable files, including decades-old text, metadata, and various content types, requiring flexible tooling.
- Metrics and ROI: Jay Parikh advocates for measuring AI spending not just in tokens, but against completed work and business outcomes (e.g., cost savings, increased engagement, or retention), providing governance controls for large enterprises.

## Practical implications

- Build engineers must design systems to accommodate agent-driven workflows, treating agent output as a primary input source.
- Focus on platform abstraction layers (like the Copilot Harness) to ensure tool interoperability and context persistence across different client surfaces.
- Implement robust governance and cost-tracking mechanisms for AI services, moving beyond simple token counting to measure ROI and business outcomes.
- Prioritize security hardening for the entire software supply chain, especially concerning dependency management and deep code scanning.

## Topics

AI Engineering, Platform Engineering, DevOps, Open Source Security, Model Orchestration, Scaling Infrastructure, GitHub’s 2026 Octoverse report, Matt Pocock’s skills repo, Copilot Fusion documentation, Copilot session across clients demo

Source: https://www.youtube.com/watch?v=rSWiNE92xqM
