GitHub COO Kyle Daigle Says the Real A.I. Coding Battle Is About More Than Code
The way software is built may never be the same. Advances in large language models have enabled what renowned A.I. researcher Andrej Karpathy referred to last year as “vibe coding,” or using natural language to generate code. According to Inferya, more than 25 million apps had been built with vibe coding tools as of 2026.
Today, Microsoft’s Github Copilot, Claude Code, OpenAI Codex and Cursor are among the most widely used A.I. coding tools. According to agent intelligence provider Exceeds, GitHub Copilot is used by 40 percent of large companies, but it is the primary tool for only 17 percent of developers. Claude Code leads primary tool share at 28 percent, followed by Cursor at 24 percent, GitHub Copilot at 17 percent, and OpenAI Codex at 11 percent.
While competitors like Anthropic focus on building increasingly powerful coding assistants, GitHub is positioning itself as an ecosystem. It provides a centralized, cloud-based platform where developers can work directly in their integrated development environments (IDEs) with real-time code suggestions. It also offers access to multiple frontier models, including Claude Code and Codex, alongside lower-cost open-weight models.
This platform-based approach has become a key differentiator, as building applications involves far more than writing code. “It’s not solely about generating code. It’s about creating software, and that requires you to do more than just the code portion of it,” Kyle Daigle, GitHub’s chief operating officer and chief marketing officer (CMO) of developer at Microsoft, told Observer.
A 13-year veteran of GitHub—since before Microsoft’s 2018 acquisition—Daigle helped bring Copilot to market in preview in June 2021. He now leads developer-first strategy across GitHub and the broader Microsoft ecosystem, including product launches, branding and developer relations.
Copilot offers developers and enterprises the “advantage of not just being locked into a single model family or needing to change the interface that you’re using,” Daigle said, pointing to “auto mode” as a differentiator. This feature interprets task intent, estimates complexity and selects a cost-effective model to complete the task.
The cost of A.I. coding
Cost has become a central concern for A.I. coding, with a range of industry horror stories emerging over high token consumption. Most notably, Uber reportedly consumed its A.I. budget in just four months. Likewise, Axios shared the story of an anonymous A.I. consultant who claimed their company spent half a billion dollars after failing to put usage limits on Claude Code licenses for employees.
Across the industry, vendors like GitHub and Anthropic have experimented with billing changes. GitHub, for example, has shifted from request-based billing to usage-based billing. Under the former, each interaction counted as a single premium request unit; under the latter, costs vary based on the model used and the number of tokens consumed.
Usage-based billing has drawn criticism from some developers for driving up costs. “The ongoing criticism of GitHub’s usage-based pricing and reliability underscores a broader infrastructure problem that hasn’t kept pace with the speed of A.I. coding,” Martin Reynolds, field CTO at software delivery platform Harness, whose role includes working with customers to shape the product roadmap, told Observer via email.
From this perspective, the industry as a whole not only needs to offer access to powerful coding tools, but needs to do so at scale in a cost-effective way. Measures like Copilot’s model harness and multi-model selection can help control costs, but there’s still more to be done to support builders in the future.
