The Agentic Flood: Why GitHub COO Kyle Daigle is Betting on Developer Choice

With AI agents expected to generate 14 billion commits this year, the primary challenge in software development is shifting from writing code to managing the chaos of automation. GitHub COO Kyle Daigle argues that the most durable advantage in this agent-native era is not a proprietary model or a locked-in ecosystem, but the preservation of developer choice. By embracing usage-based infrastructure and agentic code review, GitHub is working to solve the PR flood that threatens to overwhelm maintainers. For leaders and developers, the advantage lies in recognizing that the immediate pain of managing agentic noise is a prerequisite for the long-term gain of massive, automated productivity. Those who build systems to handle this scale now will define the next decade of software creation.

The Hidden Cost of the Agentic Surge

The shift toward agentic workflows is a structural transformation of the developer experience. With 17 million pull requests created by agents in March alone, the traditional manual review process is effectively obsolete. Daigle notes that the industry is currently climbing the hill of agent adoption, moving beyond early experimentation into a phase where agents handle the heavy lifting.

The danger here is the PR flood. When agents generate code faster than humans can review it, maintainers risk being drowned in noise. Conventional wisdom suggests that more code equals more progress, but the reality is that without control mechanisms, the signal-to-noise ratio collapses.

"There is so much more code being created. And I think at times everyone goes like, oh, this is all just like, slap. This is all just code that is being pushed up. And no one cares. It is not really true."

-- Kyle Daigle

GitHub is responding by providing the building blocks of control for maintainers, allowing communities to define their own standards for contribution. The insight here is that the system must remain flexible. By refusing to force a single standard, GitHub avoids the mouse trap of a walled garden, betting that developer choice will ultimately drive the most stable long-term ecosystem.

The 18-Month Payoff: Why Hill Climbing Beats Moonshots

Most organizations chase moonshot AI features, but Daigle emphasizes the power of hill climbing, a continuous, iterative process of small improvements based on real-world usage data. This requires patience that most teams lack. By leveraging feedback loops, such as monitoring whether developers accept or reject AI suggestions, teams can tune models to be more personalized and effective over time.

This approach creates a significant competitive moat. While competitors might chase the latest headline-grabbing model, those who focus on the ever-improving loop of evaluations and user sentiment are building a product that adapts to the specific needs of its users.

"It is not these moonshots or like a hill climb. It is just climb, climb, improve, new eval improve, new data improve and just keep going to get to the point where we are able to launch these models."

-- Kyle Daigle

The downstream effect is a more durable product. When a model is frontier-tuned to understand an individual developer’s context, it becomes an extension of the user rather than just a tool. This creates a lasting advantage: the agent becomes smarter the longer you use it, making it harder for users to switch to a competitor that lacks that accumulated, personalized context.

Solving the Token Economics Trap

A major hidden consequence of the agent economy is the potential for costs to spiral, where a $200 subscription balloons into a $2,000 monthly bill. The solution is not just to limit usage, but to implement intelligent model routing.

By automatically shifting tasks between high-performance, expensive models and smaller, efficient models, developers can maintain productivity without the financial overhead. This is a classic systems-thinking trade-off: the immediate discomfort of managing token costs is solved by building an automated layer that handles the complexity for the user. Over time, this shifts the incentive from choosing the best model to letting the system choose the best model for the task, which reduces the cognitive load on the developer.

Key Action Items

• Implement Automated Routing: Over the next quarter, audit your agentic workflows to identify tasks that can be offloaded to smaller, cheaper models. Do not rely on a single model of the day.
• Establish Maintainer Controls: If you manage open-source or internal projects, prioritize building vouching or filtering systems for incoming PRs now. This prevents the flood from becoming an operational crisis in 6 to 12 months.
• Adopt the Hill Climbing Loop: Stop chasing perfect features. Start measuring user acceptance data, such as thumbs up or down, weekly to refine your internal tools. This creates a compounding advantage that pays off in 12 to 18 months.
• Build Personalization Pipelines: Invest in fine-tuning or frontier-tuning models on your team’s specific codebase and documentation. This creates a memory that makes agents significantly more effective than off-the-shelf solutions.
• Audit Your Mouse Traps: Review your current tech stack for vendor lock-in. Prioritize tools that allow for interoperability, as the long-term winner will be the platform that respects developer choice rather than restricting it.