Copilot for Business from GitHub is now generally accessible

After a brief beta period that began in December of last year, GitHub today announced that Copilot for Business, the $19/month enterprise version of its AI-powered code completion tool, is now generally available. License management, organization-wide policy management, and more privacy features are all added by Copilot for Business. Before, you had to contact GitHub’s sales team to sign up for the business version, but now you can do it yourself.

CEO of GitHub Thomas Dohmke said, “[This] effectively wraps up our v1 Copilot story. “We had the general availability last summer after announcing the preview in June 2021, which seems like a very long time ago. We are now prepared to introduce it to businesses, teams, enterprises, and really everyone. In fact, Copilot for Business launched with more than 400 organizations already signed up, and we are seeing a ton of interest.

GitHub also revealed today that Copilot now supports proxy connections, including those with self-signed certificates, and that the AI model powering its Copilot code completion tool has been improved (using OpenAI). The group is continuously improving the models and including fresh functionality as it becomes accessible in Azure’s OpenAI Service. He compared this procedure to “spec bumps” in the hardware industry, where the model improves slightly before being brought to Copilot by the team.

As the model improves, the team is also adding new features, such as “fill-in-the-middle,” where the model can begin adding words in the middle of a line in addition to finishing it because it is aware of what comes before and after the current cursor position, for instance. According to Dohmke, the model does this by examining related files you work on and using that information to craft model queries. It uses adjacent files and nearby information in addition to what you type in your own open file to craft the prompt that is sent to the model for inference.

With these most recent updates, the team also made it possible for Copilot to identify typical security flaws in code that is returned from the model with the aid of another model. If it discovers those, it will immediately move on to a more secure suggestion.

However, the goal should be to improve while also speeding up. Dohmke stated that the group is always trying to reduce latency. Developers become impatient very quickly when Copilot takes too long to generate its code, according to data from GitHub.

Copilot will soon be able to generate 80% of a developer’s code, according to Dohmke. Currently, the average across all programming languages is around 46%, with Java coming in at 61%. With just a brief prompt, ChatGPT and Bing can now generate entire applications (or at least portions of them). The obvious question, which I posed to Dohmke, is when these features will also be added to Copilot since those models also make them possible.

We have nothing to announce today, but we think the Bing team’s newly launched feature is really exciting. Wink, wink, nudge, nudge. He did add that, for a product like Copilot, the team always aims to make it easier for developers to complete their tasks, and the current models might merely result in incorrect code.