Microsoft’s GitHub announces Copilot assistant that can learn about companies’ private code

Microsoft-owned GitHub on Wednesday introduced a more expensive Copilot assistant for developers inside companies that can explain and provide recommendations about internal source code.

The launch might help Microsoft boost revenue in its cloud business segment by taking fuller advantage of partner OpenAI’s technology. It could better position Microsoft to fend off challenges from its foremost cloud rival, Amazon. In October Amazon Web Services said it was starting to test the ability to customize its CodeWhisperer programming assistant with private code.

Microsoft was the first large technology company to release software that could help developers complete their lines of code, with the launch of the Copilot in 2021, drawing on GitHub’s publicly accessible code collection. An experiment suggested that developers who used the tool worked 55% faster. In December 2022 GitHub began selling a business-focused tier for $19 per person per month.

Last month Microsoft CEO Satya Nadella told analysts on a conference call that GitHub Copilot had accumulated 1 million paid users.

The new GitHub Copilot Enterprise will cost more than double the business offering, at $39 per person per month. Interested parties can join a waiting list ahead of a full release in February 2024.

“We heard from many, many customers that they want the ability to tune or customize Copilot on their codebase,” Thomas Dohmke, GitHub’s CEO, told CNBC in an interview earlier this week. Some clients even have their own special programming languages, he said.

On GitHub’s website, people with Copilot Enterprise licenses can choose repositories to fine-tune Copilot on proprietary code. From there, they can ask GitHub’s Copilot chatbot about elements of existing code and suggest lines of code in development environments. Over time, the Copilot will also be able to summarize code changes.

The price bump might appear considerable. Stephen O’Grady, co-founder and principal analyst at industry research group Redmonk, called it “an enormous bargain.”

Coding assistants such as Copilot, Amazon’s CodeWhisperer, Duet AI from Google and others from startups such as Replit, Sourcegraph and Tabnine promise to provide insight on multiple languages and frameworks at any hour of the day. Copilot is the most widely adopted AI coding system available, Dohmke said.

These systems are still nascent, though, with less than 10% enterprise adoption, according to Gartner, a technology industry research firm.

In research reports, Gartner has advised clients to do their own evaluations of productivity improvements from coding assistants, rather than exclusively relying on software companies’ claims. The assistants have made mistakes and raised concerns among security executives, Gartner said. GitHub, for its part, suggests that developers test, review and check code that Copilot recommends.

Yet the additional knowledge of corporate code in Copilot Enterprise means advanced developers might be able to spend less of their time on procedural components, and more junior ones can get up to speed more rapidly, O’Grady said.

“If these systems save you one hour of time for a developer, you do the math,” he said.

GitHub talked about Copilot Enterprise and other updates, such as a Copilot partner program to add capabilities through integrations, at its annual Universe conference in San Francisco on Wednesday.

“Copilot provides incredible value for enterprises, incredible productivity gains, and the enterprise gets way more value out of Copilot than the price reflects,” Dohmke said. The higher price factors in the additional costs to deliver the product to clients, he added.

Analysts at Piper Sandler estimated that annualized GitHub Copilot revenue could reach $3 billion by 2026 in a Monday note, assuming that 16% of GitHub’s 100 million users were using it. The model from the analysts, who have the equivalent of a buy rating on Microsoft shares, did not include the impact from GitHub’s Copilot Enterprise.

WATCH: GitHub CEO: A.I. model is not sentient — human developers are still in charge

Leave a Reply

Your email address will not be published. Required fields are marked *