Is AI-powered IDE the future of software development? GitHub is proposing the idea.
At its annual GitHub Universe conference in San Francisco on Monday, GitHub announced Copilot Workspace. It's a development environment that lets developers brainstorm, plan, build, test, and run code in natural language using what GitHub calls “Copilot-powered agents.”
Jonathan Carter, head of GitHub Next, GitHub's software research and development team, pitches Workspace as a more general evolution of Copilot, GitHub's AI-powered coding assistant, to help developers understand their code. It builds on recently introduced features like Copilot Chat where you can ask questions. in natural language.
“Through our research, we found that for many tasks, the biggest friction point for developers is getting started, and especially knowing how to approach the task. [coding] The challenge is knowing which files to edit and how to consider multiple solutions and their tradeoffs,” Carter said. “So we wanted to build an AI assistant that could meet developers at the beginning of an idea or task, reduce the activation energy needed to get started, and collaborate with developers as they make necessary edits across the core base. I thought about it.”
At last count, Copilot had more than 1.8 million paying retail customers and 50,000 business customers. But Carter envisions a much larger base drawn in by enhancements with broad appeal like Workspace.
“Developers spend a lot of time developing, so [coding issues]”We believe that through a 'thinking partnership' with AI, we can help developers unleash their power every day,” said Carter. “Copilot Workspace can be thought of as a companion experience and development environment that can complement your existing tools and workflows and simplify a range of development tasks. We believe that an AI-native development environment has a lot of value to offer. We believe that you are not limited by existing workflows.”
To be sure, there is internal pressure to make Copilot profitable.
The Wall Street Journal reports that Copilot loses an average of $20 a month per user, with some customers paying as much as $80 a month for GitHub. And the number of rival services continues to grow. Amazon's CodeWhisperer was made available to individual developers for free late last year. There are also startups like Magic, Tabnine, Codegen, and Laredo.
Given a GitHub repository or a specific bug within a repository, Workspace, powered by OpenAI's GPT-4 Turbo model, understands the comments in the repository and comes up with a plan to (try) to squash the bug or implement a new feature. can be built. Publish a reply and a larger codebase. Developers get code recommendations for bug fixes and new features, a list of things they need to validate and test that code, and controls to edit, save, refactor, or undo.
Image credit: GitHub
Suggested code can be run directly in Workspace and shared among team members via external links. Once in Workspace, these team members can adjust and tinker with the code as needed.
Perhaps the most obvious way to start Workspace is to click New to the left of issues and pull requests in your GitHub repository.[Open in Workspace]This is the method to do it from the button. Clicking this opens a field that describes the software engineering task you complete in natural language, such as “Add documentation for the changes in this pull request.” Once submitted, this will be added to the list of “sessions” within the new file. Dedicated workspace view.
Image credit: GitHub
Workspace systematically executes requests step by step, creating specifications, generating plans, and implementing those plans. Developers can review any of these steps, review the proposed code and changes in detail, and remove, rerun, or reorder steps as needed.
“When we ask developers what they often get stuck on with a new project, they often say they don't know where to start,” Carter said. “Copilot Workspace eases that burden and gives developers a plan to start iterating.”
Image credit: GitHub
Workspace entered technical preview on Monday and is optimized for a variety of devices, including mobile.
Importantly, because Workspace is in preview, it is not covered by GitHub's IP indemnification policy. This policy promises to help with the legal costs of a customer facing a claim from a third party that the AI-generated code she is using infringes her IP. (Generative AI models are notorious for regurgitating training data sets, and GPT-4 Turbo was trained in part using copyrighted code.)
GitHub has not decided how it will commercialize Workspace, but says it will use the preview to “understand the value it provides and how developers will use it.”
I think the more important question is: Can Workspace solve the existing problems surrounding Copilot and other AI-powered coding tools?
GitClear, the developer of the code analysis tool of the same name, analyzed more than 150 million lines of code committed to project repositories over the past few years and found that Copilot pushed more incorrect code into the codebase. It turns out that more code is being pushed into the codebase. It's a headache for code maintainers because it's re-added rather than reused or streamlined.
Elsewhere, security researchers have warned that Copilot and similar tools can amplify existing bugs and security issues in software projects. And researchers at Stanford University found that developers who accepted suggestions from AI-powered coding assistants were more likely to write less secure code. (GitHub highlighted to me that it attempts to use an AI-based vulnerability prevention system to block insecure code, in addition to an optional code duplication filter to detect backflow of public code.)
But developers aren't shying away from AI.
In a June 2023 StackOverflow poll, 44% of developers said they currently use AI tools in their development process, and 26% said they plan to use AI tools in the near future. Gartner predicts that by 2028, 75% of enterprise software engineers will hire an AI code assistant.
By emphasizing human review, Workspace could actually help clean up some of the mess caused by AI-generated code. Once Workspace is in the hands of developers, you'll know right away.
“Our main goal with Copilot Workspace is to leverage AI to reduce complexity and give developers more freedom to express their creativity and explore,” said Carter. “We truly believe that the combination of humans and AI will always be better than either alone. That's what we're betting on with his Copilot Workspace.”