GitHub Copilot has been expanded again. Workspace, currently still in technical preview, promises AI assistance at every step of the development process. What exactly does it mean?
GitHub Copilot was released in October 2021. The AI development platform is now equipped with GPT-4 and has already contained many AI features via “Copilot X” for a year. Everything from generating code and refining pull requests to answering questions about technical documents is already in the Copilot suite. Now GitHub is going one step further: Copilot Workspace offers AI assistance from the very beginning of a software project. It sounds like software engineers should be kept awake at night, but what exactly does Workspace do?
Working
Workspace can take a look at the current state of a project on GitHub and suggest changes. It explains those changes in fairly clear language. The “Copilot-native” development environment seems mainly designed for refining a current codebase. However, the GitHub team also suggests that developers can come up with an entirely new project with the tool’s help. Workspace can materialize a general idea by throwing up the first piece of programming code. Other GitHub Copilot features are then ready to tackle every other aspect of software development.
A suggested Workspace use case is adding AI to Pong. Workspace starts by answering a question that appears on a GitHub page: Is there AI in Pong yet? After a brief analysis, the Workspace tool generates an answer. A user can then add additional information point by point. Workspace then generates a plan that can be expanded again. With an additional push of a button, Workspace changes the files to implement the change.
The user therefore remains firmly at the controls. The thought process of Workspace is also clear to follow. If something is unexpectedly wrong (which is sometimes the case with every GenAI tool), it can be corrected. The only opaque thing is the underlying AI model itself, which was developed in collaboration with OpenAI and GitHub parent Microsoft.
The Bard among programming assistants
Workspace builds on previous developments within GitHub Copilot. For example, GitHub Copilot Chat arrived a few months ago and essentially offered a repackaged version of the standalone features that made up Copilot X. Things like code review, pull requests, writing documentation; AI assistance has been provided in these areas within the GitHub Copilot suite for some time.
The approach of Workspace is to act as “rubber duck” or colleague to function, says a GitHub developer on YCombinator. “It is very useful to make an idea more tangible/concrete.” It is said that a Workspace session starts a discussion within a team and simplifies the step towards implementation. All this with just a few clicks (and even on a smartphone).
It’s very reminiscent of the explanation Google CEO Sundar Pichai gave when Google Bard (now Gemini) debuted last year. Programming help was also promised there, albeit to a limited extent. The promise was primarily to serve as a “thought partner,”to bounce ideas off“. GitHub Copilot Workspace limits itself to coding tasks to be more accurate, but its promises are as nuanced as Bard’s.
Is AI programming the future?
Despite these nuances, many news organizations are quite bombastic on the topic of AI programming assistance. For example, GitHub Copilot is said to be the beginning of the ‘automation of the coding industry’. Problems abound, however: allegations of plagiarism tarnished the AI tool’s reputation shortly after launch.
Since then, the limitations of AI code have become increasingly apparent. For example, debugging generated code is seen as extra tedious, as AI often makes very different mistakes than a human. Large codebases are also unsuitable for analysis, as (current) AI models have a limited context window (actual short-term memory). The assistance is limited to specific tasks, minor refinements and solving already defined problems. The fact that Copilot Workspace focuses precisely on this task is therefore not surprising. Starting up a project that is as interpretable as it can be also offers AI the opportunity to be compensated freewheelen with ideas as a “thought partner”.
No Devin, no worries
What Copilot Workspace is not is a full-fledged AI software engineer. Above all, the implementation of GenAI within GitHub is modular, even if Workspace and Copilot Chat unite different AI components. Workspace is therefore not intended to complete an entire codebase with a few clicks.
The promises are more economical than what we saw with Devin from start-up Cognition AI. That debuted in March as “the first AI software engineer,” although it is still in beta. Once that tool sees the light of day, we can look past the staggering benchmarks and know whether software engineers really need to look for another job. In any case, that suggestion does not work for Workspace. Ultimately, people still have to come up with good ideas themselves, with AI helping out.
Also read: Devin is the first AI software engineer: should developers be concerned?