Back to Podcast Digest
Latent Space1h 24m

GitHub’s Agent Era: 14x Commits, 200M Developers, Copilot’s Next Act — Kyle Daigle

TL;DR

  • GitHub is running at a 14x commit growth rate: Daigle says GitHub went from about 1 billion commits in 2024 to roughly 275 million commits per week in 2025, with PRs and other activity accelerating fast enough to expose new infrastructure bottlenecks.

  • The main scaling pain is not old-fashioned traffic, it is agent-shaped traffic: More agents means more PRs, more builds, more CPU demand in Actions, larger monorepos, and permissioning/database layers like the internally infamous "MySQL 1" getting stressed in ways vertical or horizontal scaling alone cannot fix.

  • Copilot's next act is not just better autocomplete, it is software-wide agent orchestration: GitHub spent time on fine-tuning, then model progress changed the roadmap, and now Copilot is being rebuilt around a shared SDK across the CLI, desktop app, cloud agents, documentation, security remediation, and issue handling.

  • Micro-skills beat giant prompt contraptions: Internally, GitHub is moving away from giant all-in-one skills and toward tiny, Lego-like skills that do one thing well, because mega-skills rot quickly when tools, prompts, and workflows change.

  • AI is especially powerful for leaders who used to code: Daigle says his own commits are up because he can connect business context, notes, transcripts, PRs, Slack, Teams, and Obsidian into working tools, including a revenue-planning presentation he built with AI that nobody flagged as AI-generated.

  • GitHub's idea of a developer is expanding past traditional gatekeeping: Daigle says GitHub now has over 200 million developers, and he argues that if someone creates software with code, even with AI's help, they count, which helps explain why star counts and project adoption now move at a much faster, messier internet scale.

The Breakdown

GitHub is seeing commit volume jump from 1 billion in all of 2024 to a 14 billion annualized pace in 2025, and Kyle Daigle says the breakage is coming from genuinely new scaling patterns: bigger repos, more agents, more builds, and old permission systems being pushed past their design limits. He also makes a broader claim about the agent era: the winners may be former developers in leadership roles, because AI lets them turn years of business context into working apps, workflows, and even board-ready presentations without needing a full staff pipeline.

Was This Useful?

Share