GitHub Copilot Chat aims to replace Googling for devs

GitHub made Copilot Chat available in public beta this week, adding a natural language interface into developers' working environments in a bid to reduce time spent researching ideas elsewhere.

GitHub Copilot Chat is part of a package of technical previews released in March under the moniker GitHub Copilot X, each of which integrates OpenAI's GPT-4 generative AI model with developer features such as chat and voice interfaces, pull requests and documentation. OpenAI's Codex model underpins the original GitHub Copilot for writing code. With this release, GitHub Copilot Chat serves as the centerpiece of the GitHub Copilot X product line and touts integration into Visual Studio and VS Code IDEs.

"This is designed to keep developers in the flow, which has become increasingly challenging," according to a GitHub press release. "They … are spending more and more time spelunking through documentation and search results to figure out how to connect to an API, use a new database or understand what a colleague was trying to accomplish."

As Ev Kontsevoy, CEO at Teleport, put it in an interview this week: "I've heard numerous times where engineers would basically joke that 'All I need to do my work is a text editor where coding happens, and Google,'" he said. "Having those two motions, producing value with your brain on one side and looking things up and copy pasting something that's routine [on the other] tightly integrated in one tool is a big step forward."

Kontsevoy, whose company markets a secure systems access platform, has a vested interest in the success of GPT-based chat interfaces; Teleport markets one with its Assist feature for infrastructure-as-code. For another GitHub Enterprise user, chat still represents a disruption to developer workflows.

"I like how the current iteration of GitHub Copilot is integrated directly into the console while I'm typing – I can add comments about what I'm doing directly in the coding screen and Copilot reads it and recommends code," said Kyler Middleton, senior principal software engineer at healthcare tech company Veradigm. "I've provided the same feedback to our GitHub Enterprise team -- I require any tool we're going to integrate here to take our developers out of their 'focus zone' as little as possible, and most of their upcoming tools require too much 'stop programming and go interact with widget x' today."

Basically, you have to get out of the zone and go to Google and start searching things, and then you're expose to all kinds of distractions. You go on Reddit to look something up and then you end up looking at cats.

But this kind of interruption was happening anyway, even more disruptively with Googling, Kontsevoy said.

"Basically, you have to get out of the zone and go to Google and start searching things, and then you're exposed to all kinds of distractions," he said. "You go on Reddit to look something up, and then you end up looking at cats."

Broader generative AI growing pains continue

Generative AI has been a craze in enterprise IT since OpenAI released ChatGPT to the public in November 2022. Most IT vendors are adding support for generative AI to their tools, including observability, DevSecOps and infrastructure as code tools, in addition to coding assistants. However, the initial trepidation about whether the benefits outweigh the downsides for this still-new tech lingers for some engineers.

On one hand, it's exciting to contemplate the prospect of AI-generated unit tests among the features coming under GitHub Copilot X, said Rob Zazueta, a freelance technical consultant in Concord, Calif. On the other, Zazueta sounded a note of caution about the quality of such tests, especially early on.

"It doesn't escape my attention that every interaction starts with the caveat, 'I'm powered by AI, so surprises and mistakes are possible,'" Zazueta said. "Reading code is a separate skill from writing code -- you can develop it while learning to write code, but the best way to really learn is to review the code of others on a regular basis … [a] junior developer may not have the sharp code reading skills to ensure the [AI-generated] code actually fixes the bug without introducing any others."

Teleport's Kontsevoy said he sees generative AI as the same as any other developer tool, which can be used proficiently or not. But he shares another widespread concern about the still-unknown legal and licensing ramifications of using AI-generated code for proprietary applications.

"The question is what will happen on the legal side of things over time, because it's one of the examples where technology is running ahead of legislation," Kontsevoy said. "We need to make sure that in the future, we will not retroactively get ourselves in trouble."

Beth Pariseau, senior news writer at TechTarget, is an award-winning veteran of IT journalism. She can be reached at [email protected] or on Twitter @PariseauTT.