Back in October 2021, developers on the GitHub software development platform were given access to an exotic AI tool called Copilot. Created in association with OpenAI, the company behind ChatGPT, Copilot uses the same generative AI technology to produce computer code on request, rather than text.
And it is pretty good at it. Github reckons that Copilot successfully autocompletes coding suggestions about 50 percent of the time, which should significantly increase the productivity of the millions of developers who now use it.
And that raises an interesting question: just how influential is Copilot set to become?
Trillion Dollar Question
Now we get an answer thanks to the work of Thomas Dohmke at Github, Marco Iansiti at Harvard Business School and Greg Richards at Keystone.AI who have measured how developers are using Copilot. Their conclusion is that the AI-powered coding is producing a sea change in the software industry that is set to turbocharge global GDP by over $1.5 trillion by 2030. “This symbiotic relationship has the potential to shape the construction of the world’s software for future generations,” they say.
GitHub is a software development platform that allows developers to collaborate with ease on global scales. Launched in 2008, bought by Microsoft in 2018 and now with 100 million developers on its books, Github has rapidly become the go-to repository for software development, particularly for open-source projects.
Copilot is powered by the same Generative Pre-Trained Transformer technology behind ChatGPT. But instead of generating text, it produces code. As such, it has the potential to significantly influence software production.
So Dohmke and co decided to find out by how much. They analyzed the way almost a million Github developers use Copilot and the code it produces. “On average,” they say,” users accept nearly 30% of code suggestions and report increased productivity from these acceptances”.
The pattern of use is interesting too. Dohmke and co say that the productivity impact increases over time and that the benefits are greatest for less experienced developers. That is consistent with other work suggesting that less experienced workers benefit most from AI assistants because they have the most to learn.
In addition, Dohmke and co say that most of the innovation with AI-powered coding is on open-source projects and usually led by individuals rather than corporations. They say their findings “suggest that the open-source ecosystem, particularly in the United States, is driving generative AI software innovation. “
So how much is this increased productivity worth? Dohmke and co point out that software development significantly contributes to global GDP already but that its contribution is limited by the global shortage of developers. They say the increased productivity from Copilot could fill some of this gap and that this increase in productivity would add $1.5 trillion to global GDP by 2030.
Dohmke and co are adamant that this is a conservative estimate, pointing out the adoption of AI-powered coding techniques is accelerating and that what we are seeing now is just the beginning of a massive change. “As more developers adopt these tools and become fluent in the skill set of prompting with generative AI, it is clear that this new way of software development has created an inextricable link between humankind and artificial intelligence that could well define how the world’s software is built for generations to come,” they conclude.
Hang on to your hats!
Ref: Sea Change in Software Development: Economic and Productivity Analysis of the AI-Powered Developer Lifecycle : arxiv.org/abs/2306.15033