GitHub Copilot aims to help programmers do their jobs faster. It recently made Copilot Chat generally available, a “core part” of its own developer platform, according to GitHub.
With Copilot Chat, a programmer can interact with an AI assistant to translate code to another programming language, write documentation and track errors, among other things. According to GitHub, it is up to the user to decide what to do with it.
As part of the Copilot X initiative, Chat has been a fairly known quantity for some time. A number of Copilot X features were further explored in 2023 to complement the existing GitHub Copilot, which has been positioned as an “AI pair programmer” since 2021. It costs $10 per month or $100 if one opts for the annual plan.
That appears to be a successful business model, as GitHub Copilot is said to be the first generative AI product to generate $100 million ARR.
Different expectations
Again, GitHub Copilot’s goal is to be a “pair programmer,” an AI assistant that removes or speeds up a meaningful portion of work. There’s plenty to find online about the actual user experience. Those responses show that there are widely varying expectations of the tool. This isn’t too surprising, since GitHub itself does not have a narrow definition of what Copilot excels at. It supports all kinds of programming languages. The tool can also facilitate documentation writing, something many organizations fall short on.
Regardless, Copilot’s results are variable. Where one user only spends more time fixing bugs in AI-generated code, another finds huge time savings. What’s striking in this regard is that the positive responses mostly seem to have somewhat unambitious goals in mind.
For example, one programmer on Reddit states that they’ve become twice as productive thanks to Copilot, which essentially acts as “an advanced autocomplete,” “not something that the code invents for me.” This is exactly a point that critics make as well.
Drew DeVault, CEO of GitHub alternative SourceHut, touts that, indeed, Copilot doesn’t invent anything new. The tool can even replicate the exact code it was trained on, which tends to come from open-source projects that don’t all just allow such straightforward reproduction in their licensing. In doing so, according to DeVault, GitHub completely ignores the basis open-source licenses were built upon. Thus, it appears that not only the efficacy, but also the way Copilot was developed, causes annoyance.
Chat is more of the same, but safer and more accurate
In all likelihood, Copilot Chat will change few opinions. For one, the chat feature was already available in beta form to a limited number of users, so it’s not entirely new. However, the general launch does allow organizations to open up the tool for use among their development teams with a single switch inside a dashboard. The big advantage here is that programmers can stay within their own IDE and still refine their code with a chatbot.
Previously, ChatGPT was considered a possible alternative for this, but it has some snags. For example, it is not advisable (and often illegal) to pass on company information to that chatbot, as OpenAI would then potentially have access to it. In addition, ChatGPT does not have access to the documentation of the organization in question and provides generic answers. GitHub Copilot Chat should instead be context-aware, so that it generates code that is constructed in the same way as the existing code base.
In short, there are many advantages to GitHub Copilot Chat, which is now accessible to many more users. Those who would have otherwise turned to ChatGPT now have access to a more sensible and secure alternative. However, those who have already encountered problems with the existing GitHub Copilot tools will ultimately be hard to convince. After all, the underlying technology remains the same as before, with all its pros and cons. However, GitHub’s free interpretation of its uses does show how nimble the tool can be. The only question is whether the end user has expectations that Copilot Chat can live up to.
Also read: GitHub soon to make 2FA mandatory; quick activation desired