# GitHub’s COO Explains Why AI Hasn’t Replaced Developers

Podcast: AI & I
Published: Jun 17, 2026
Reading time: 14 min
Canonical: https://podbrew.app/briefs/ai-and-i-github-s-coo-explains-why-ai-hasn-t-replaced-developers

Guest host Mike Taylor welcomes Kyle Daigle, the COO of GitHub and Microsoft's chief marketing officer for developer products. Daigle, a 13-year veteran at GitHub, offers insights into the platform's evolution and the transformative effect of artificial intelligence on software development.

They delve into the significant surge in agent-generated code, which now accounts for a massive increase in GitHub commits and pull requests. The discussion highlights how AI tools like GitHub Copilot are empowering non-developers, such as legal and marketing professionals, to build applications, effectively dissolving traditional boundaries. GitHub's preparations for an agent-native future, including agentic code review, model routing, and a commitment to developer choice, are also key topics.

This rapid expansion of AI-driven code challenges existing development practices and business models, necessitating new infrastructure and strategic shifts from GitHub. This transition marks a pivotal moment, making software creation more accessible and redefining what it means to be a developer in an increasingly agent-powered landscape.

## Key takeaways

- AI tools like GitHub Copilot are empowering non-traditional users, including legal and finance professionals and knowledge workers, to build applications and assets.

- GitHub is implementing agentic code review and merge functionalities to automate and simplify parts of the pull request workflow.

- Agent-created pull requests are a primary driver of GitHub's current growth, with 17 million agent-generated PRs recorded in March alone.

- GitHub projects a 14x increase in commits this year compared to last year's 1 billion, largely due to the amplified output from AI agents.

- The rise of AI agents signifies a shift where individual developers can leverage multiple AI tools to extend their capabilities and scale code creation.

- The continuous operation of AI agents challenges GitHub's existing human-centric freemium and per-seat business models.

- GitHub plans to support extensive AI agent usage, which suggests a transition towards usage-based pricing for these automated services.

- GitHub's strategy is to foster an open ecosystem by partnering with diverse AI model providers rather than creating a "walled garden."

- GitHub leadership actively uses diverse operating systems and development tools, including competitor products, to intimately understand various developer experiences.

- Understanding why developers choose competitor tools helps prevent 'blind spots' and informs strategic decisions about product focus and developer choice.

- The long-term AI strategy anticipates improved models, more critical token economics, and the ability to run capable small language models locally on devices.

- Deep personalization, context, and fine-tuning are identified as crucial for the future of AI, enabling agents to intuitively complete user thoughts for a superior experience without explicit user instructions.

- Iterative 'hill climbing' leverages continuous user feedback, including acceptance data and sentiment, to improve AI models and development tools.

- Effective improvement requires balancing objective 'hard measures' (e.g., performance) with subjective 'soft measures' (e.g., user sentiment), as they don't always align.

- The 'hill climbing' philosophy focuses on consistent, small, data-driven improvements rather than one-time, large-scale 'moonshots.'

- Model routers automatically select the most cost-effective AI model for a specific task, preventing the overuse of expensive, high-capacity models.

- Companies like GitHub and Microsoft Foundry are developing these routers to help developers optimize AI agent costs by saving on token consumption.

- Personal AI agents can deliver unbiased, specific feedback on communication nuances, such as clarity, repetitive phrasing, and metaphor usage.

- People tend to be more open to receiving critical feedback from AI systems than from other individuals, perceiving it as less threatening.

- A continuous AI feedback loop, which reviews past communications and tracks the application of previous suggestions, fosters recursive self-improvement in personal skills.

## 01:43 - 03:50 GitHub expands its definition of a developer to include non-coders using AI tools

GitHub has always held a broad view of what constitutes a developer, recognizing that many individuals, like Kyle Daigle, start writing code for personal projects before adopting the 'developer' label. Daigle himself began coding to fund art school, illustrating how personal need or curiosity often precedes professional identification.

While GitHub provides serious developer tools for large enterprises, its newer AI-powered offerings, like GitHub Copilot, are broadening its user base. The company observes its own legal and finance teams, alongside general knowledge workers and non-traditional coders, utilizing these tools to create small applications and assets.

This expansion signifies a deliberate effort to make coding more accessible, providing an 'on-ramp' for more people to engage in software creation. The goal is to facilitate anyone's ability to build apps, whether for personal use, family, a startup, or a business, effectively blurring the lines of who can be considered a developer in today's landscape.

> A lot of the folks that the industry would call knowledge workers or just non-by-trade developers are using these tools to build little apps or assets for them.

## 03:51 - 06:01 GitHub Introduces Agentic Tools for Code Review and Open Source Management

GitHub is rolling out new agentic tools to help streamline the pull request (PR) process. This includes agentic code review, which aims to get PRs into a better state for easier human review.

Another significant feature is agentic merge within the app. This allows users to configure GitHub Copilot to handle final PR processing steps, such as initiating the merge and ensuring all CI and policy checks pass.

For open-source projects, GitHub is specifically focused on empowering maintainers. They are developing tools that give maintainers control over accepting PRs, deciding who can contribute, and defining the level of validation needed for contributions.

GitHub's approach is to provide flexible building blocks for maintainers, allowing each community to decide its own best practices. The company prefers not to dictate standards, like the "Vouch system" used by some, but rather to observe emerging community practices and then integrate widely adopted solutions.

> Every community is choosing a slightly different way to approach the problem, and for GitHub, we've- We've always wanted to leave that in their hands, like give them tools and enable them, but if a standard comes out of that or most are using a certain practice, we'll lock that in, but we don't really ever wanna be the first to create a standard or an approach.

## 06:01 - 08:02 Agent-Created Pull Requests Drive GitHub's Exponential Growth

GitHub is experiencing an unprecedented surge in platform activity, largely fueled by the integration of AI agents collaborating with human developers. This shift is leading to a dramatic increase in code generation and development work flowing through the platform.

The scale of this growth is evident in key metrics: last year, GitHub saw a billion commits. This year, if current trends continue, the platform is on track for 14 billion commits. A significant contributor to this explosion is agent-created content; in March alone, AI agents generated 17 million pull requests.

This influx of agent-generated code is not considered low quality or 'slop.' Instead, it represents a fundamental transformation where developers can work alongside one or more AI agents, leveraging their skills, resources, and context to amplify their output. This new development paradigm is pushing the industry beyond early adoption phases, with significant growth still ahead.

GitHub is actively preparing for this next wave of expansion, anticipating continuous growth rather than a plateau. The platform's central role means that regardless of where code is built or what tools are used, much of it ultimately ends up on GitHub for sharing and collaboration, necessitating robust support for this agent-native development economy.

> in March, there were seventeen million pull requests that were created by agents.

## 08:02 - 10:02 Adapting GitHub's Business Model to Agent Usage

GitHub faces a challenge in its traditional freemium and per-seat licensing models with the emergence of 24/7 AI agents. The current model, designed for human developers, doesn't account for agents working around the clock, which fundamentally alters usage patterns.

The goal is to enable users to run many agents simultaneously, potentially 150, which points towards a shift from per-seat licensing to a usage-based model for AI activity. While the specifics are still being determined, this change is necessary to accommodate the new way software is being built.

GitHub has a history of evolving its offerings, similar to how it eventually provided free private repositories after initially charging for them. The company remains focused on ensuring individual developers have a great core experience and what they need to succeed, while also supporting enterprises at scale.

> GitHub's always evolving as the industry and community does, but we're always sort of focused on, I need to make sure you, the dev, have what you need to be successful.

## 10:02 - 12:02 GitHub's developer-first philosophy influences Microsoft's broader tooling and events

GitHub has long maintained a unique philosophy: they build tools explicitly for developers, not for the enterprise buyers. While corporate purchases are welcome, the primary focus remains on creating authentic and useful experiences for individual developers, a principle that has guided the company for over a decade.

This developer-centric approach is now being extended across Microsoft through Kyle Daigle's dual role as COO of GitHub and Chief Marketing Officer for developers at Microsoft. His goal is to ensure that all of Microsoft's developer tools and technologies offer holistic, authentic experiences for their users.

An example of this influence is the recent Microsoft Build event in San Francisco, which adopted a significantly different format. The event prioritized hands-on usability over pitches, focusing on whether attendees could go to a session and immediately use the discussed technology. It also notably featured community speakers, including in the keynote, emphasizing that software development is a team sport.

This shift aims to bring GitHub's deep understanding and focus on developers to a wider Microsoft audience, fostering a more practical and community-driven approach to tool development and engagement across the entire ecosystem.

> we're not building for the buyers, we're building for the developers in a hundred percent.

## 12:02 - 14:30 GitHub prioritizes developer choice and an open ecosystem in the competitive AI market.

GitHub believes no single company can answer every developer question, mirroring the collaborative nature of open-source software development. They aim to avoid "walled gardens" where developers are trapped by a single vendor, instead focusing on enabling users to integrate various tools seamlessly. This approach contrasts with systems that might create an "unintentional mouse trap" requiring new accounts or learning curves for external tools.

The company's core philosophy revolves around developer choice, building for builders, and empowering them with flexibility. GitHub intends to support developers using their platform to access and integrate other external tools, actively partnering with various providers to simplify this process.

In the AI space, GitHub is investing in its own technology, like new Microsoft AI models, while also collaborating with leading AI companies. They are partnering with Anthropic, OpenAI, and Google, as well as any other provider bringing models or coding agents to market, to offer a broad range of options to developers.

> we always want to enable developers that are building with GitHub to go use these other tools, and we'll partner with everyone to make that as simple as is possible.

## 15:20 - 17:20 GitHub leadership avoids blind spots by actively using diverse tools and understanding competitors.

GitHub's Kyle Daigle actively uses a variety of operating systems and development tools to maintain a comprehensive understanding of the developer experience. He regularly codes on his MacBook, Windows PC, and an Omaree Linux box on weekends, specifically to switch between environments. This practice ensures he understands how different platforms impact development workflows.

Daigle intentionally uses the GitHub Copilot app exclusively on Windows to guarantee that developers on that platform receive a high-quality experience, not just Mac users. This dedication to diverse testing extends across GitHub's teams, scrutinizing aspects like coding agents, desktop apps, and memory management on various systems.

He notes that a significant risk for successful companies is developing a 'blind spot' by exclusively focusing on their own tools, a pitfall GitHub has encountered previously. While the company prioritizes its own products, leadership makes a conscious effort to understand why developers choose competitor tools. This perspective informs their strategy, even if it means not adopting those tools directly, to maintain developer choice and relevance.

> It's such a blind spot... when you're doing something and you're doing it well, you really laser focus... and I think that's myopic.

## 17:30 - 20:04 GitHub's Strategy Balances Short-Term AI Agent Gains with Long-Term Personalization

GitHub is adopting a two-pronged strategy for AI product development, focusing on immediate capabilities while building for future advancements. In the short term, the emphasis is on enabling robust multi-agent sessions, recognizing the current industry trend and the clear utility these sessions provide to users.

Looking ahead, the long-term vision accounts for several key shifts. Models are expected to continue improving, and the economics of tokens will become an even greater factor in determining which models are widely adopted. Additionally, there's a strong belief that small language models (SLMs) will soon be capable enough to run serious workloads directly on local devices.

A consistent truth across the evolution of AI, from ChatGPT to GitHub Copilot, is the importance of personalization, context, and fine-tuning. While experiments exist, the industry still lacks a comprehensive long-term vision for deep personalization. Achieving this means agents can intuitively complete user thoughts, rather than requiring users to constantly codify preferences, leading to a significantly better user experience.

> Using an agent that you feel like is completing a thought for you, will give you that great experience, especially if you didn't have to personally codify that thought.

## 20:04 - 22:59 GitHub drives continuous model improvement through an iterative 'hill climbing' process using user data.

GitHub employs an iterative 'hill climbing' strategy to continuously enhance its AI models and development tools. This process heavily relies on user feedback, such as thumbs up/down data, acceptance rates of suggestions, and general engagement with the tools.

A key aspect of hill climbing involves balancing both 'hard measures' like performance metrics and 'soft measures' like user sentiment. It's common for technical improvements, based on hard evaluations, to sometimes lead to a decline in user satisfaction, even with consistent latency and performance.

The ultimate goal is to empower everyone with these 'hill climbing machines,' especially within enterprise environments using products like M365. By leveraging extensive internal data—from documents to chats—solutions like 'Frontier Tuning' with base models such as MAI Thinking One can deliver significant results without requiring extensive manual effort.

The 'hill climbing' approach emphasizes continuous, incremental improvements rather than seeking large, infrequent breakthroughs. It's about a consistent cycle of 'Climb, climb, improve, new eval, improve, new data, improve,' constantly refining models and tooling based on real-world usage.

> it is just Climb, climb, improve, new eval, improve, new data, improve, and just keep going.

## 23:00 - 25:29 Implementing Model Routers to Optimize AI Agent Costs

To combat the escalating costs associated with AI agent subscriptions, companies like GitHub are developing model routers. These systems are designed to automatically select the most appropriate and cost-effective AI model for specific tasks, rather than relying on developers to manually choose a default model that might be overly powerful and expensive for the job at hand.

The core idea is to prevent developers from consistently using high-cost models, often referred to as 'model of the day,' for every task. Many tasks do not require the most advanced and expensive large language models, leading to unnecessary token consumption and ballooning expenses. Model routers, like those being developed by GitHub and Microsoft Foundry, can dynamically route tasks to less expensive models when appropriate, based on task intent or complexity.

This automated model selection is crucial for optimizing token usage and reducing overall AI agent costs. Even for small, simple operations within a larger workflow, a model router can ensure that an efficient, cheaper model like Haiku is used instead of a more powerful, costly one like GPT-4 or Claude 3 Opus, significantly impacting the bottom line for organizations heavily utilizing AI agents.

> A lot of the reasons why tokens are expensive is because we're all going and choosing our model of the day or week or hour, and those models are incredibly expensive.

## 25:30 - 28:05 Personal AI agents provide critical communication feedback for self-improvement

Kyle Daigle employs a personal AI agent, affectionately named Baxter, to rigorously analyze his written and spoken communication. This agent provides daily "comms reports" that scrutinize everything Kyle produces, from emails to interview responses.

The AI offers precise feedback, highlighting recurring phrases Kyle uses, assessing the clarity of his language, and evaluating the effectiveness of his metaphors. Given Kyle's preference for using metaphors, the agent specifically provides examples of clear and impactful ones.

This method of receiving critical feedback from an AI agent proves exceptionally effective for self-improvement. Kyle notes that individuals are more receptive to constructive criticism from a robot than from a human, as it feels less threatening and more objective.

Baxter's process involves a continuous seven-day loop, reviewing all of Kyle's past communications, including emails and Slack messages. The agent then delivers feedback and monitors whether Kyle applies previous suggestions, creating a powerful, recursive cycle for ongoing personal development.

> Humans are way more willing to take critical feedback from robots than other humans.

---

Get podcast briefs for shows you follow: https://podbrew.app/