TL;DR:
- GitLab introduces Duo Chat, expanding its AI offerings.
- Duo Chat offers an interactive ChatGPT-like experience for developers.
- David DeSanto, GitLab’s Chief Product Officer, discusses the transition of Duo capabilities into the chat domain.
- GitLab shifts its chat backend to Anthropic’s Claude.
- Partnership with Oracle for cloud-based GPUs highlights GitLab’s cloud-agnostic approach.
- Duo Chat’s current features include code explanation and refactoring.
- Senior developers prefer code generation in chat, citing noise concerns.
Main AI News:
In a bid to empower developers and enhance productivity, GitLab made waves earlier this year with the introduction of Duo, an array of AI features designed to streamline issue summarization, epic and issue description generation, code suggestions, vulnerability explanations, and more. Today, GitLab takes a significant stride forward by introducing Duo Chat to its AI repertoire, ushering in a ChatGPT-like experience for developers to engage with. This interactive chat platform grants access to the existing Duo features, all while delivering a more engaging experience. Duo Chat is currently in beta.
During a recent interview at KubeCon, David DeSanto, Chief Product Officer at GitLab, shared insights into the company’s vision. He expressed a clear intent to migrate a substantial portion of the existing Duo capabilities into the chat domain, bridging the gap between the initial beta launch and the eagerly anticipated General Availability (GA) release.
A notable twist in this development journey is GitLab’s choice of chat backend. While the company had previously partnered with Google for various AI features, including its code completion service, GitLab has now switched gears, moving its chat backend to Anthropic’s Claude. DeSanto provided context for this decision, explaining, “We had not chosen [a model] for a chat. We were using our own model—an open-source model—and we determined that based on the way chat works within GitLab and the large context that’s needed, a 100k context window makes it a lot easier to get better information in and back.”
GitLab’s quest for versatility continues as it partners with Oracle for cloud-based GPUs, demonstrating its cloud-agnostic approach. DeSanto affirmed, “We’re going to find the best, and all three of them—Google, Oracle, and Anthropic—actually work with each other through us, so it’s been a good relationship.”
At present, Duo Chat boasts a range of code-related chat capabilities, encompassing features such as ‘explain this code’ and code refactoring. The roadmap includes plans to introduce test case generation, vulnerability explanations, and other enhancements in the near future.
Throughout the testing phase, GitLab’s team has gained valuable insights into the significance of context within this chat-based experience. GitLab enjoys a competitive advantage in its ability to access all of a developer’s code, even if it’s not currently open in the Integrated Development Environment (IDE). Moreover, the generous context windows prove instrumental in facilitating coherent and meaningful conversations, ensuring that past interactions are easily accessible.
An intriguing revelation surfaced during the closed preview period, as developers engaged with the tool. Seasoned developers, it appears, do not always embrace code suggestions, considering them to be potentially disruptive. Instead, they gravitate towards accessing these AI tools through Duo Chat. DeSanto elaborated on this phenomenon, stating, “I thought that chat would be a lot more for associated developers, but it seems like they want the inline code completion—and then senior developers gravitate more to code generation in chat. The reason was noise. They said: I know what I’m typing. I don’t need you to tell me. But they do maybe want some guidance or some refactoring, for example.”
Conclusion:
GitLab’s introduction of Duo Chat signifies a strategic move to cater to the diverse needs of developers, offering an interactive AI-powered platform. The transition of Duo capabilities into chat and the choice of Anthropic’s Claude for the chat backend reflect GitLab’s commitment to providing a context-rich and efficient developer experience. This development positions GitLab as a significant player in the AI-enhanced developer tools market, appealing to both novice and seasoned developers seeking productivity and guidance.
