GitHub's Agent Era: 14x Commits, 200M Developers, and Copilot's Next Act

Kyle Daigle, CEO of GitHub and CMO of Developer for Microsoft, shares his insights on the evolving landscape of software development, the impact of AI on productivity, and GitHub's journey through unprecedented growth and innovation.

Why AI Got Kyle Coding Again

Kyle Daigle's personal coding activity has seen a significant resurgence, a trend he attributes directly to the advent of AI. He describes a newfound ability to leverage AI to connect disparate problems and data sources, enabling him to "crank up 15 agents on Saturday." This process, he explains, is less about writing traditional software and more about orchestrating AI to analyze vast amounts of information—from GitHub pull requests and online posts to personal notes and internal transcripts—to generate actionable insights and plans. This recursive, backward-looking analysis, identifying patterns and suggesting improvements, is something Large Language Models (LLMs) excel at, particularly for non-technical leaders. Daigle has built numerous internal tools and workflows using the GitHub Copilot desktop app, all of which are hosted on GitHub.

Running GitHub with AI: WorkIQ, MCP, Slack, Teams, and Skills

GitHub has embraced AI internally by focusing on integrating it into existing workflows without requiring employees to learn entirely new tools. The company has developed a set of "skills" that are distributed via a CLI and now a desktop app, granting AI access to read and process information from GitHub, Teams, email, and Slack. While GitHub has a long history of using Slack for "chat ops," they primarily use Microsoft Teams for video communication. This approach allows for powerful retrospective analysis, crucial for a globally distributed workforce that might miss information shared in real-time. AI-generated findings and industry reports are automatically posted into GitHub issues or discussions, fostering ongoing conversations and enabling teams to adapt quickly.

The Golden Age for Former Developers in Leadership

Daigle posits that we are currently in a "golden age for former developers who are now in leadership roles." This is because individuals with a development background possess a unique ability to identify patterns and solve problems, a skill that translates effectively across management, operations, and now, AI-driven development. With AI tools, these leaders can leverage their existing technical knowledge and extensive business expertise to build and deploy applications more efficiently. He highlights his own experience, where his decade-plus of leadership experience is now being applied as a developer with AI tools, creating a powerful synergy. This is particularly true for those who entered development after gaining experience in other fields, as they bring a broader skillset and a unique perspective to problem-solving.

GitHub's History: Actions, npm, Webhooks, and Open Source

GitHub's journey has been marked by significant milestones, including the launch of GitHub Actions in 2018, the acquisitions of npm, SEML, and Dependabot, and the ongoing evolution of its platform. Daigle recalls launching the first version of Actions at GitHub Universe in October 2018, a project he led as an engineering leader. He also touches on the acquisition of npm, emphasizing the shared goal of ensuring its continued scalability and security. Regarding security issues, Daigle clarifies that while Actions are a critical compute layer, the primary source of vulnerabilities often lies in the code within repositories themselves. He draws a parallel to early GitHub services like webhooks, where executing arbitrary Ruby code on behalf of users presented similar security challenges that have since been addressed through advancements in containerization and underlying infrastructure.

Slop Forks, Vendoring, and AI Dependency Management

The concept of "slop forks," where developers might vendor only the necessary parts of a dependency, is discussed as a potential approach to mitigating security vulnerabilities. Daigle acknowledges the historical practice of vendoring and suggests that while it might help to some degree, it won't entirely solve fundamental security problems. The challenge lies in convincing AI agents that code is secure, a task that requires continuous investment in static code analysis and runtime testing. He emphasizes GitHub's role in providing tools for maintainers to manage dependencies and AI-generated code, balancing security with the need for flexibility and community consensus.

Pull Requests, Prompt Requests, and Trust in Agent-Generated Code

The evolution of the pull request (PR) process is a key topic, with Daigle noting that GitHub standardized this once-messy workflow. He anticipates a future where a significant percentage of PRs might be generated by AI agents, raising questions about how trust will be established in this new paradigm. While verification flows and data analysis are improving, the core challenge remains building human trust in agent-generated code. He likens this to the development of self-driving cars, where a combination of verifiable proof and human perception of safety is crucial. For regulated industries, this trust must extend to governments and regulatory agencies, a complex hurdle that will likely be overcome as AI systems demonstrate consistent reliability and safety.

GitHub Stars, 200M+ Developers, and the New AI Builder Wave

GitHub has surpassed 200 million developers on its platform, a number that includes not just traditional coders but also individuals engaging with software development through AI tools. Daigle argues against segmenting developers, emphasizing that the barrier to entry is lowering, allowing more people to participate in creating and contributing code. He believes the rapid growth in stars for new projects reflects this influx of new builders who are actively exploring technologies and contributing to the vibrant open-source ecosystem. While acknowledging the existence of spam and gamification, he attributes much of the growth to this new cohort of AI-empowered creators.

GitHub Spark, Low-Code, and Why GitHub Still Shows the Code

GitHub's foray into low-code initiatives, like the Spark project, has reinforced a core tenet: always show the code. While Spark aimed to simplify app building and running, the team learned that the primary value for developers was in easy runtime, a need that can often be met by existing hosting solutions. GitHub's commitment to transparency means they will never hide the underlying code, even in simplified interfaces. This approach ensures that developers can always delve deeper, fostering a sense of empowerment and enabling them to build and modify software as easily as changing a light switch.

GitHub’s Hardest Era: 14x Growth, Reliability, and Scale

GitHub is currently navigating an unprecedented period of growth, experiencing what Daigle describes as both the "best and most exciting time" and a "hard time" due to the strain on its systems. The platform is seeing growth rates of up to 14x year-on-year, leading to novel scaling challenges that are breaking systems in new ways. While historical issues like webhook unreliability have been addressed, the current problems stem from complex permissioning layers and the sheer volume of compute required for actions, builds, and agentic workflows. GitHub is actively investing in expanding its compute capacity, moving to Azure, and re-architecting core services like its database infrastructure to handle this exponential growth.

Actions as the Compute Layer for CI/CD and Automation

GitHub Actions has become a critical compute layer for CI/CD, automation, and various processing tasks. While it offers significant value, its widespread use, including free tiers, contributes to the demand for compute resources. Daigle acknowledges that outages have sometimes been related to Actions, but ongoing improvements to underlying infrastructure and the expansion of compute capacity are expected to alleviate these issues. He anticipates a reduction in availability problems over the next three months as these material changes pay off, enabling GitHub to handle even greater exponential growth.

The State and Future of GitHub Copilot

GitHub Copilot, initially launched with a focus on code completion, has evolved significantly. The team initially concentrated on fine-tuning models for improved accuracy and performance. However, the rapid advancement of LLMs and the emergence of other AI coding tools prompted a shift. GitHub is now focused on building a unified SDK and harness for its coding agents, enabling them to be used across various applications like the Copilot CLI, desktop app, and cloud agents. The future of Copilot extends beyond code generation to encompass security remediation, issue management, and documentation analysis, aiming to provide a comprehensive AI coding agent experience that integrates seamlessly into the developer workflow.

Ambient AI, Background Agents, and the Future of the SDLC

The conversation turns to the future of AI in software development, with Daigle advocating for "ambient AI" that operates in the background, maintaining context across all developer activities, not just coding. He believes current tools are too focused on capturing and recalling information, whereas the ideal AI would possess a deep understanding of project specifications, emails, conversations, and business context to inform its decisions. This broader AI integration, he argues, is crucial for software development, which is inherently collaborative and influenced by business objectives and market trends. OpenClaw is highlighted as an interesting project that connects to various data sources, enabling AI agents to act with a more holistic understanding.

OpenClaw, Enterprise Security, and the New OS for Agents

Microsoft's significant investment in OpenClaw, including a dedicated CVP, underscores its importance as a potential "new operating system for AI." Daigle explains that OpenClaw's value lies in providing a standardized way for agents to access user resources and perform tasks, particularly in enterprise environments. The need for robust sandboxing at the OS level is paramount to ensure security and prevent malicious activity when agents operate on personal or work devices. Microsoft's contributions to OpenClaw, including developing platform components and enhancing security features, aim to empower developers to build and deploy AI agents more safely and effectively.

Build Announcements, WorkIQ, FoundryIQ, and Microsoft Context

Daigle expresses excitement about the announcements at Microsoft Build, emphasizing a significant shift in Microsoft's approach to developers, regardless of their preferred operating system. He highlights improvements for daily Windows users and the development of tools that enable developers to build and deploy applications at work with the same agility they experience in personal projects. WorkIQ and FoundryIQ are presented as powerful context engines that can access and process information across various work assets without compromising security. These tools aim to provide a seamless intelligence layer, bridging the gap between personal development workflows and the complex requirements of enterprise environments.

What Should swyx Ask Satya?

When asked what question he would pose to Microsoft CEO Satya Nadella, Daigle suggests asking about his vision for AI in two to three years. He believes this question can elicit insights into Microsoft's strategic direction regarding AI, inference, tokenization, and the future of work, potentially addressing current doubts and providing a clearer roadmap for the industry.