Could an AI-infused IDE represent the next chapter in software development? GitHub seems to think so.
In the lead-up to the GitHub Universe conference set for early autumn in San Francisco, GitHub unveiled the Copilot Workspace, a development environment enhanced by “Copilot-powered agents” designed to assist developers in conceptualizing, planning, coding, testing, and executing code through conversational language.
Jonathan Carter, who leads GitHub Next, the platform’s experimental software division, introduces Workspace as an advancement of the AI-driven coding aide Copilot. This new iteration extends its utility beyond coding assistance to a more versatile tool, leveraging the recently added Copilot Chat feature that enables developers to interact with code queries conversationally.
Carter explains, “Our investigations revealed that initiating tasks often poses the greatest hurdle for developers, especially in determining the approach to a coding challenge, identifying the relevant files, and weighing different solutions. Our goal was to create an AI ally that engages with developers right from the spark of an idea, lowering the barriers to start and collaborating throughout the codebase modifications.”
With over 1.8 million individual subscribers and 50,000 corporate clients, Copilot’s user base is substantial. Yet, Carter anticipates attracting an even broader audience with the introduction of appealing new features like Workspace.
Carter believes that since developers dedicate much of their time to resolving coding problems, a ‘cognitive collaboration’ with AI could significantly empower them daily. Copilot Workspace is envisioned as a supportive development environment that enhances existing tools and processes, aiming to simplify a range of development tasks. The potential for delivering substantial benefits in an AI-centric development setting, unbound by traditional workflows, is immense.
The imperative to turn Copilot into a profitable venture is strong.
A Wall Street Journal article suggests that Copilot incurs an average loss of $20 monthly per user, with some users costing GitHub up to $80 monthly. Meanwhile, the competition is intensifying, with Amazon’s CodeWhisperer now free for individual developers and startups like Magic, Tabnine, Codegen, and Laredo entering the fray.
With the support of OpenAI’s GPT-4 Turbo model, Workspace can devise strategies to address bugs or add new features within a GitHub repository, leveraging insights from the repository’s comments, issue discussions, and broader codebase. It offers developers code suggestions for fixes or new functionalities, complete with a checklist for validation and testing, as well as options to modify, save, refactor, or reverse the changes.
The code recommendations provided can be executed within Workspace itself and disseminated to colleagues through a shareable link. Once team members access Workspace, they have the liberty to adjust and enhance the code according to their requirements.
The most straightforward method to initiate Workspace is by utilizing the “Open in Workspace” feature, situated next to the issues and pull requests on GitHub repositories. By selecting this option, a user can articulate the software engineering task in plain language—for instance, “Create documentation for the modifications in this pull request.” This task is then cataloged in a series of “sessions” within Workspace’s specialized interface.
Workspace methodically processes requests, first formulating a specification, then devising a strategy, and ultimately executing it. Developers have the option to delve into each stage for a detailed examination of the proposed code and modifications, with the flexibility to eliminate, redo, or rearrange these stages as needed.
Carter notes, “Developers frequently encounter obstacles at the outset of a new project, uncertain of the initial steps.” He adds, “Copilot Workspace removes this obstacle, providing developers with a structured plan as a starting point for further development.”
Workspace is set to launch its technical preview on Monday, tailored for various devices, mobile included.
It’s crucial to note that during its preview phase, Workspace does not fall under GitHub’s IP indemnification policy. This policy typically offers legal fee support to customers who are challenged with third-party claims of IP infringement due to their use of AI-generated code. This is particularly relevant as generative AI models, like GPT-4 Turbo, have a tendency to reproduce elements from their training data, which includes copyrighted material.
GitHub has yet to finalize its plans for commercializing Workspace, opting instead to utilize the preview period to gauge its utility and observe developer interaction with the tool.
The pressing issue, however, is whether Workspace will address the fundamental problems associated with Copilot and similar AI-driven coding aids.
According to GitClear, a company specializing in code analysis, an examination of over 150 million lines of code submitted to project repositories in recent years indicates that Copilot has led to an increase in erroneous code being integrated into codebases. This trend has resulted in more code being rewritten rather than reused or optimized, posing challenges for those tasked with maintaining the code.
Security experts have cautioned that tools like Copilot could exacerbate pre-existing flaws and vulnerabilities in software. Moreover, a study by Stanford indicated that developers who rely on AI coding assistants might end up with less secure code. GitHub has countered this by highlighting its use of an AI-driven system to prevent vulnerabilities and a feature to identify and filter out duplicated code from public repositories.
Despite these concerns, developers are not turning their backs on AI.
A recent StackOverflow survey from June 2023 revealed that 44% of developers are already incorporating AI tools in their workflow, with another 26% planning to do so shortly. Gartner forecasts that by 2028, AI coding assistants will be used by 75% of developers in enterprise settings.
With a focus on human oversight, Workspace aims to mitigate the issues that can arise from AI-generated code. The effectiveness of this approach will become clear as Workspace is adopted by more developers.
Carter emphasizes, “The main objective for Copilot Workspace is to harness AI to simplify the development process, thereby enabling developers to unleash their creativity and navigate more freely. We are convinced that the synergy of human intelligence and AI will outperform either working in isolation, and that’s the foundation of our commitment to Copilot Workspace.”