GitHub recently released Copilot Workspace, an environment powered by Copilot that will assist developers throughout the software development lifecycle. It is currently available in technical preview for a waitlist of developers, but it will be integrated into the larger GitHub Copilot platform after it exits preview.
4 New Insights about GitHub Copilot Workspace
Let’s take a look at the experiences shared by a few developers who had the opportunity to test GitHub Copilot Workspace across various tasks. The initial reviews highlight the tool’s impressive capabilities.
1) High-Quality Code with Less Mistakes
Dan Shipper, CEO of Every, got access to Copilot Workspace and he wrote about his experience after spending 24 hours testing it.
He tried it for building large, complex features and small, better-defined tasks. His goal was to see what he could ask of it, what kinds of tasks it could handle, and when he might choose to use this instead of ChatGPT. He said one interesting thing about it: “The next popular programming language will probably be English“.
Unlike the constant switching between the browser and programming environment required with ChatGPT, Copilot Workspace’s experience is fully integrated into the codebase on GitHub. This seamless integration allows you to sit back and watch it make the required changes, without the need for copy-pasting code from GPT to the codebase.
Dan needed to replace an ugly logo in their internal tool Spiral with a new design created by one of their talented designers. While not a difficult coding task, it required ensuring the new logo looked right in context and didn’t break any surrounding styles – a simple yet tedious task that often gets procrastinated.
If satisfied with the changes made by it, you can create a pull request and instantly merge the new feature into your codebase. He found this aspect particularly impressive, as Copilot Workspace successfully swapped out the old logo for the new one, without requiring any manual programming or copy-pasting.
He also compared it with ChatGPT and Claude. He found that ChatGPT’s versatility allows you to use it for a wide range of tasks, but it lacks structure and guidance. On the other hand, Copilot Workspace is designed specifically for programming, following a process to help you complete tasks.
While Copilot Workspace is less flexible than ChatGPT, it might be more effective in certain situations.
In a side-by-side comparison with ChatGPT and Claude, he noticed that Copilot Workspace was slower in generating code, taking 2-3 minutes compared to 10-20 seconds for the others. However, the code generated by Copilot Workspace was of higher quality, with fewer subtle mistakes that would need manual cleanup.
For minor tasks like fixing small issues or implementing minor feature requests in huge codebases, Workspace could be quite helpful, potentially getting an engineer 90% of the way there. Its integration with GitHub makes it easy to manage and merge the changes.
When it came to a more complex feature request, such as implementing a team-sharing flow, Dan said that it yielded mixed results. He found it challenging to map Copilot Workspace’s design onto their mental model of the feature, as Copilot Workspace had made specific design decisions based on his vague initial prompt.
Dan sees Copilot Workspace as a step towards the future, where English becomes a programming language, and most software will start as sentences written into interfaces.
2) Can Do Simple Tasks But…
Ashley Peacock, an AI enthusiast also got access to GitHub’s new AI tool and put it to work on one of his side projects. He initially had concerns about it potentially replacing software engineers. However, after using it on a side project involving a Discord bot for League of Legends, his perspective changed.
He maintains a Discord bot that scrapes data to provide information about League of Legends eSports games, including upcoming fixtures, schedules, and live game announcements. Over time, the codebase became messy due to code duplication whenever a new league needed to be added.
To address this, he tried Copilot Workspace and asked it to add support for the LCK (Korea) league, which the bot didn’t previously cover. Copilot Workspace planned and implemented the necessary changes to add LCK support correctly on the first attempt.
Copilot Workspace successfully implemented a plan to add support for a new league (LCK) to the bot, making the correct code changes with impressive accuracy. When tasked with reducing code duplication, it prepared a reasonable plan and refactored the codebase, reducing it by 250 lines.
Although it made a mistake that caused some URLs to return 404 errors, Ashley was able to quickly fix the issue.
Overall, Copilot Workspace dramatically reduced the time required for these tasks, which would have taken hours to complete manually. He acknowledged that while it is prone to mistakes, engineers can efficiently fix them.
Ashley imagines a future where AI tools like Copilot Workspace significantly boost productivity, with engineers focusing on complex work while AI handles simpler, routine tasks after being reviewed and corrected by humans.
3) A Good AI Assistant for Programmers
A YouTuber Dev Leonardo also tested Copilot Workspace on one of his personal projects. He decided to use it to address an issue related to button functionality on his survey generator website.
He opened the Issues page on GitHub and added a comment providing context. Copilot Workspace analyzed the issue, including comments, to understand the current state and proposed changes.
After successfully identifying the requirements, Copilot Workspace generated an implementation plan. However, Dev noticed that the plan failed to specify the button’s intended action.
But Copilot Workspace allows you to edit the plan directly ensuring that the developer always has complete control. He added all the necessary details to the implementation plan.
With the revised plan, Copilot Workspace started the implementation process, making changes across relevant files. The user could preview the code changes line-by-line and compare the updated code side-by-side with the original version. The user decided to test the feature to verify the new implementation.
During testing, a minor styling issue with the button was identified, which the user promptly resolved. Copilot Workspace then automatically generated a descriptive pull request, including the issue number for easy reference during the review process.
The user acknowledged that while coding with an AI assistant is remarkable, Copilot Workspace felt like a peer programmer. The ability to edit and revise every step made a significant difference, fostering a collaborative and iterative process.
4) Provides 80% Code for Developers’ Team
Eddie Aftandilian, a researcher at GitHub Next, shared how his team utilized Copilot Workspace to generate their initial Terraform setup for managing Copilot Workspace infrastructure, despite the team lacking expertise in Terraform.
Impressively, Copilot Workspace accurately coded around 80% of the Terraform setup, providing a valuable starting point for the team when working with an unfamiliar technology.
Conclusion
These key insights shared by developers for GitHub’s Copilot Workspace reveal its impressive capabilities to simplify coding and enhance productivity. Even with its limitations, it represents a step towards a future where AI assists programmers in efficiently tackling basic tasks, thus allowing teams to focus on more complex challenges.
