GitHub's plan for Agents — Kyle Daigle, GitHub

Episode Summary

Executive Summary: GitHub’s Carvell describes how his role has expanded from developer/COO-type operator to Microsoft’s CMO for developers, while GitHub and Microsoft adapt to AI-driven shifts. The conversation centers on using AI for workflow automation, the rise of micro-skills over “mega skills,” GitHub’s scaling and reliability challenges, Copilot’s evolution into an agent platform, and the trust/security problems facing open source and enterprise software.

Main Topics: Expanded role across GitHub and Microsoft (Priority: 5/5): Carvell explains his new remit as both GitHub CEO and Microsoft’s CMO for developers, emphasizing a developer-first approach to product marketing, authenticity, and translating technical capabilities for enterprise buyers and broader audiences. AI as workflow automation and retrospective intelligence (Priority: 5/5): He argues the most valuable AI use cases are not just generating content, but stitching together context from tools like GitHub, Teams, Slack, email, and notes to summarize what happened, identify patterns, and plan what to do next. Skills, agents, and the shift from mega-tools to atomic tools (Priority: 4/5): The discussion contrasts oversized all-in-one AI skills with smaller, composable micro-skills that do one thing well and can be orchestrated together, with English-based editing seen as a key usability breakthrough. GitHub reliability and infrastructure scaling under AI growth (Priority: 5/5): Carvell details GitHub’s uptime issues as a consequence of explosive growth in commits, PRs, actions, and agents, stressing that the company is rewriting permissioning, queueing, and compute layers to support a much larger load. Copilot’s evolution into an agentic platform (Priority: 5/5): GitHub Copilot is positioned as more than code completion: it is becoming a shared SDK/harness for coding agents, CLI workflows, desktop apps, and cloud agents that can handle code, issues, security remediation, and documentation. Trust, security, and open source governance (Priority: 5/5): A major theme is how GitHub can increase security without breaking developer workflows, especially in npm and PR review. Carvell frames trust as fundamentally social, not purely technical, and says GitHub should provide flexible trust tools rather than impose one standard. OpenClaw, WorkIQ, and the future of ambient AI (Priority: 4/5): He argues the next leap is ambient AI that has access to work context across tools and can proactively help, not just react to prompts, and points to Microsoft’s work on context engines and sandboxing as part of that future.

Key Arguments: AI is most valuable when it helps people understand what happened across many systems and turn that into action, not just when it writes text or code. Non-technical leaders can now use AI effectively by creating reusable workflows and atomic skills instead of learning a new tool for every task. The era of giant, perfect AI skills is fading; smaller, specialized micro-skills are more maintainable and easier to adapt as work changes. GitHub’s outages are tied to a real, multi-dimensional scale problem: more users, more agents, more CI/build load, more permission checks, and bigger repositories. GitHub is moving infrastructure toward more compute, container isolation, and reworked database/permission systems to support AI-era usage. Copilot is becoming an orchestration layer for agents across the SDLC, not just a code-completion product. Trust in open source contributions cannot be reduced to stars or commit counts because those signals are easily gamed; social and contextual verification matters more. GitHub should adapt to community practices rather than force a single process, because maintainers and projects have different trust and contribution needs. The best AI experiences will be ambient and context-aware, with access to a user’s broader work graph and history, not isolated prompts. Microsoft’s developer strategy is becoming more system-level, spanning OS, sandboxing, compute, and context engines to support agents safely at work.

Data Points: Tenure at GitHub: 13 years - Carvell says he has worked at GitHub for 13 years, starting as a developer. Commit activity increase: Commits have gone up - He references a public callout that his commits increased again as he returned to coding with AI tools. GitHub growth pace: 1 billion commits in 2025; 275 million commits per week; on pace for 14 billion this year - Used to illustrate the scale and speed of GitHub’s growth and the resulting infrastructure strain. GitHub developer/account scale: Over 200 million - He says GitHub now has over 200 million accounts/users, emphasizing that the platform is broader than just professional developers. GitHub Pages / Actions history: Actions launched in October 2018 - He notes he helped launch the first version of GitHub Actions right before the Microsoft acquisition closed. Aging of Microsoft acquisition: About eight years - Mentioned when discussing GitHub’s continued use of Slack despite being part of Microsoft. Scale comparison of current growth: More in a month than in a year last year - He says recent monthly activity exceeds last year’s annual pace, showing how quickly usage has accelerated. Historical product period: 2013–2014 - He references this era when discussing his earlier coding work and the old GitHub infrastructure.

Pivotal Quotes: "The titles have always been like always felt a little strange to me." — Carvell: He explains that he still thinks of himself primarily as a developer and builder, despite senior executive titles. "We have to do this in a way where no one has to change how they work." — Carvell: He describes GitHub’s internal AI rollout philosophy: embed AI into existing workflows instead of forcing tool adoption. "We’re trying to codify trust. We’re trying to say, like, okay, if Sean reviews this, I’m going to trust it because you’re Sean or you’re the senior dev or you’re the whatever." — Carvell: He frames the core open-source review problem as a social trust issue, not just a technical verification issue.

Implications: GitHub and Microsoft are betting that AI’s real value will come from context-rich, work-integrated agents. For developers and enterprises, the winners will be tools that are trustworthy, composable, and safe at scale—not just impressive demos.