Cyrus Gazanfar, chief technology officer of 401(k) asset management startup Beagle, told Business Insider that his company has begun experimenting with Copilot in recent months.
He now asks his team of three engineers to use the tool because he has quickly found that it saves him a lot of time by removing routine work, also claiming that Copilot has already proved useful in creating databases.
Many databases have the same properties, such as a time frame that indicates when the data was created. According to Gazanfar, they can now use Copilot to create a database with these properties.
Bill Mears, vice president of engineering at LookDeep Health, says Copilot draws its knowledge from the large corpus of projects on GitHub, but also uses the context of the project the engineer is working on to suggest adapted code.
Although Copilot’s suggestions are accurate, you should not trust it completely. Ghazanfar advises that the developer should be careful, because sometimes they will have to go back and correct bugs when the proposed code doesn’t work.
GitHub’s CEO, Nat Friedman, also shared about the product’s shortcomings. At a recent event in San Francisco, he compared using Copilot to participating in a lottery.
Developers write their own code, and while Copilot’s suggestions are often wrong, it occasionally passively offers an answer that perfectly solves a developer’s need, Friedman said.
“Copilot will give you a very convincing answer that looks great, and that code looks great, and those API calls look great. But then it turns out that the API doesn’t even exist.”
Mers says, referring to the application programming interfaces that help applications communicate with each other.
According to experts, Copilot is prone to “hallucinations,” a term the artificial intelligence industry has adopted to describe chatbots that confidently offer completely made-up answers.
AI tools would speed up productivity as well as benefit developers, but are unlikely to completely replace the need for engineers any time soon.
