Key Moments

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

Latent Space PodcastLatent Space Podcast
Science & Technology5 min read85 min video
Jun 2, 2026|783 views|19
Save to Pod

Want to know something specific about what's covered?

We've already dissected every moment. Ask and we will deliver (with timestamps).

TL;DR

GitHub's COO/CMO Kyle Daigle reveals AI is enabling him to write 14x more code and build complex workflows previously impossible, but this surge is also breaking GitHub's systems.

Key Insights

1

Kyle Daigle's personal commit count has increased by 14x due to his use of AI agents for tasks like analyzing PRs, online posts, Obsidian notes, and work transcripts.

2

GitHub now has over 200 million developers on its platform, experiencing exponential growth that strains its infrastructure, leading to recent system outages.

3

The 'agent era' allows non-technical leaders to craft complex workflows, enabling them to analyze past performance and generate future strategies, a task previously impossible.

4

The shift towards AI in software development is causing a "diagonal scaling" challenge, requiring rewrites of 10-15 year old foundational systems rather than simple vertical or horizontal scaling.

5

GitHub Copilot has evolved beyond code completion to an SDK and harness for coding agents across various tasks, including security remediation and issue management.

6

The most significant scaling challenges for GitHub are stemming from Actions (requiring more CPUs) and permissioning layers (still heavily reliant on outdated databases like MySQL 1).

AI supercharges productivity and broadens developer roles

Speaking with swyx, GitHub COO and CMO Kyle Daigle details a profound shift in productivity driven by AI, particularly for leaders. Daigle himself has seen a 14x increase in his personal code commits, attributing it to the ability to build and deploy complex AI agents. These agents automate tasks previously impossible for a single person, such as synthesizing information from PRs, online posts, personal notes (Obsidian), and work transcripts (Teams, Slack, Work IQ). This empowers individuals, including non-technical leaders, to perform retrospection on past performance and derive actionable insights for future strategy. Daigle emphasizes that AI is not just for basic code completion but for creating sophisticated workflows, enabling a return to a 'feeling of creation' previously hard to replicate in other professional capacities. He also highlights that this empowerment extends to individuals with diverse backgrounds and previous careers, who can now leverage AI with their accumulated knowledge and skills.

GitHub's exponential growth strains infrastructure

The platform is experiencing unprecedented growth, now boasting over 200 million developers. This surge, significantly driven by the 'agent era' and new cohorts of developers, is placing immense pressure on GitHub's systems. Daigle notes that while growth has always been a factor, the current pace is astronomical, leading to system breakages that are both novel and challenging. This rapid expansion necessitates fundamental changes to the platform's underlying architecture, moving beyond traditional scaling methods.

The "diagonal scaling" challenge: Rewriting foundational systems

GitHub is facing a unique 'diagonal scaling' problem. Traditional vertical (more powerful servers) and horizontal (more servers) scaling methods are insufficient due to global CPU/GPU constraints and the sheer complexity of modern workloads. This necessitates rewriting foundational systems that have been in place for 10-15 years. Daigle identifies two primary areas of pain: Actions, which require significantly more CPUs due to increased builds from tools, agents, and PRs, and permissioning layers, still heavily dependent on legacy databases like 'MySQL 1.' The move to Azure and additional cloud compute is underway to address these needs. Furthermore, a trend towards larger monorepos, contrary to previous industry deflation, also presents performance challenges that require infrastructure improvements benefiting all repositories.

Evolving GitHub Copilot: From code completion to an agent harness

GitHub Copilot, initially known for code completion, has evolved into a comprehensive SDK and harness for coding agents. It now supports a broader range of tasks, including security remediation, issue management, and agent orchestration across various platforms like the new CLI and desktop app. Daigle explains that the focus shifted from fine-tuning a single model to developing a flexible system that allows agents to leverage diverse data sources and context. The goal is to enable users to interact with agents more naturally, moving beyond transactional commands to more integrated workflows across development, security, and operational tasks.

Trust, security, and the future of open source collaboration

The conversation delves into the complexities of trust and security in open source, especially with the rise of AI-generated contributions. Daigle discusses the challenges of managing vulnerabilities and potential 'slot forks' in package management like npm. He highlights GitHub's role in balancing community needs with security, acknowledging that solutions must be malleable to diverse maintainer preferences. The platform is exploring ways to enhance trust signals beyond passive metrics like stars and commit counts, potentially through more active social signals and agentic workflows that codify trusted heuristics. The idea of a universal digital identity for developers is also floated as a way to build trust.

Ambient AI and the dream of deep context integration

Daigle expresses a vision for 'ambient AI' that goes beyond current assistant models. He imagines AI agents with a deep, pervasive understanding of all available context – spec docs, emails, conversations, and dependencies – to inform their decisions. This contrasts with tools focused on capturing and recalling information. The ultimate goal is to bridge the gap between weekend projects and enterprise development, enabling the same agility and power in work environments. This includes enhanced sandboxing for agents and leveraging AI for tasks beyond just coding, such as security remediation and documentation analysis.

The evolving definition of "developer" and the democratization of creation

The platform's growth extends to a broadening definition of a 'developer,' now encompassing over 200 million users. Daigle pushes back against gatekeeping, emphasizing that anyone who creates code or an application, regardless of their background or how they learned, is a developer. This inclusive approach aligns with GitHub's ethos of being a home for all creators, allowing individuals to pursue ideas and build applications with the same passion that fueled early tech communities. The focus is on removing barriers to entry and fostering a sense of empowerment, where individuals can modify and build upon existing tools and applications.

The future of operating systems and AI as a platform component

Microsoft's investment in projects like Open Claw signifies a shift towards a new 'operating system for AI.' This involves building foundational platform components rather than just end-user products. The goal is to provide developers with the tools and infrastructure needed to build advanced AI applications, including better sandboxing for OS-level agents and cloud versions for wider accessibility. This approach aims to empower developers by offering reusable components and enabling them to build at work with the same agility they exhibit in personal projects, addressing complex security and compliance needs and moving towards physics-level problems in compute.

Leveraging AI for Developer Productivity and Company Operations

Practical takeaways from this episode

Do This

Embrace AI tools to build workflows and connect disparate data sources for retrospection and planning.
Focus on developing micro-skills that perform one task well, rather than complex mega-skills.
Utilize AI for tasks previously done by staff, like presentation building or data analysis, to improve efficiency.
Persist with AI tool adoption, even if initial results are not perfect; iterate by adjusting skills.
Leverage AI's pattern-finding capabilities for retrospection and planning, especially for non-technical leaders.
Consider AI as a new operating system component that needs sandboxing and security at the OS level.
Explore context engines like Work IQ and Foundry IQ for accessing and utilizing work-related data.
Look for AI agents with ambient capabilities that understand context beyond specific coding tasks.

Avoid This

Avoid trying to teach everyone new tools; build AI integrations that fit existing workflows.
Don't rely solely on mega-skills that become difficult to update as requirements change.
Do not be afraid to experiment and iterate on AI skills, even if they aren't initially perfect.
Avoid hiding AI usage; be transparent, but also craft outputs to appear natural or professional.
Do not abandon older, effective tools if they are purpose-built and deeply integrated.
Be cautious of AI tools that only capture, codify, and recall information without proactive application.
Do not assume AI will replace human connection needs, especially in roles like Chief of Staff.

Common Questions

Kyle Daigle is leveraging AI by building agents and workflows that connect various data sources to automate complex tasks. He specifically mentions using AI to process PRs, online posts, personal notes, and transcripts from work tools to generate weekly messaging plans, demonstrating a significant increase in his coding activity.

Topics

Mentioned in this video

Software & Apps
Dependabot

Mentioned as one of the acquisitions GitHub made.

ChatGPT

Mentioned as a common AI tool used by non-technical employers for basic tasks.

GitHub Copilot

Discussed as a key AI tool for developers, its evolution, and its role in increasing commit activity and enabling new workflows.

Claude

Mentioned as a common AI tool used by non-technical employers for basic tasks.

Obsidian

Mentioned as a tool used for storing notes, which AI agents can process for workflow analysis.

Microsoft Teams

Used by GitHub for video communication, with a note that Slack is still used for chat operations.

Work IQ

A context engine used at Microsoft/GitHub to ask questions across work data, particularly useful for geographically distributed or remote teams.

Slack

Used by GitHub for chat operations and integrated with commands and workflows, despite the Microsoft acquisition.

GitHub Actions

Discussed as a feature launched in 2018, and later identified as a significant source of scaling issues due to increased CPUs needed for builds.

Ruby

Mentioned in the context of early GitHub services where arbitrary Ruby code could be executed, highlighting the evolution of security practices.

Azure

Mentioned as the cloud platform used for the new Dev Compute service that GitHub is leveraging.

Windows

Discussed in the context of Microsoft's efforts to improve its operating system for developers and AI agents, including sandboxing.

Foundry IQ

A powerful context engine that connects to existing stores and provides intelligence across work data, similar to Work IQ.

More from Latent Space

View all 221 summaries

Ask anything from this episode.

Save it, chat with it, and connect it to Claude or ChatGPT. Get cited answers from the actual content — and build your own knowledge base of every podcast and video you care about.

Get Started Free