AI coding tools are moving into the everyday work of software development, according to GitHub CEO Thomas Dohmke. He sees tools such as Copilot becoming part of the standard toolkit, while GitHub’s findings show that developers use suggestions selectively and that less experienced developers may benefit especially.
Why GitHub expects AI use to spread
Speaking at the Collision conference in Toronto, Canada, Dohmke said AI and software development are “inextricably linked for the rest of our lives.” In an interview after his talk, he pointed to companies watching competitors to judge whether they have adopted Copilot. Even when adoption is limited to some developers, he said, a company may see it as an advantage.
Dohmke also described Copilot as a natural tool for developers to use. He expects it to become part of the standard toolset, while acknowledging that some developers will continue working without it. He compared that possibility to Cobalt developers still existing.
That view reaches beyond code completion. Dohmke said AI tools will become integrated across the development lifecycle, suggesting their role could extend through more stages of building software.
Developers accept suggestions selectively
GitHub’s analysis of a sample of almost one million users found that developers accept just under 30% of Copilot’s code suggestions. The reported acceptance rate rises with experience: after six months of use, developers accept closer to 35% of suggestions.
Dohmke said he does not expect those figures to change all that drastically in the near future. He added that a 50% acceptance rate would “make us happy.” The figures indicate that Copilot’s influence does not depend on developers accepting every suggestion; the tool can contribute code while developers decide what fits their work.
GitHub also highlighted Copilot’s usefulness for less experienced developers. The company defined experience by the average number of repository actions on GitHub before a developer began using Copilot. That definition ties the finding to activity on the platform, rather than to a broader measure of professional experience.
A bigger prediction for AI-written code
In a report co-written by Dohmke, Marco Iansiti and Greg Richards from Keystone.AI, GitHub said it anticipates that 80% or more of code will be written with AI as developers become more fluent in prompting and interacting with AI. The report connects that possibility to new models that let natural language power parts of the development lifecycle.
GitHub presents this as a way to democratize software development for more people. In practical terms, natural-language interaction could help people work with development tools without relying only on conventional code suggestions. The report’s forecast depends on developers building fluency with AI and on the arrival of models that support this kind of workflow.
GitHub’s estimate of the economic impact
GitHub argues that generative AI developer tools could help address a shortage of skilled developers and increase productivity. The company estimates that these tools could add $1.5 trillion to global GDP by 2030, and says each missing skilled developer accounts for $100,000 in GDP loss.
GitHub believes AI developer tools could make up for about 15 million additional developers, which underpins its estimate of the total economic impact. The company describes its estimate as conservative. These are GitHub’s projections, not reported outcomes, and they rest on the company’s view of how much AI tools can expand developer capacity.
Together, the adoption figures and forecasts point to a shift in how software may be made. Copilot suggestions are accepted selectively today, while GitHub expects AI to play a much larger role as developers learn to work with it and tools spread across the development lifecycle.