GitLab introduced Duo earlier this year, a suite of AI functionalities designed to enhance developer efficiency. These functions include summarizing issues, generating descriptions of errors and solutions, providing script recommendations, and explaining risks. Recently, the company expanded its product lineup with Duo Chat, a ChatGPT-like interface that allows developers to engage with bots to access current Duo features in a more interactive manner. Duo Talk is currently in its alpha stage.
The plan is to transition many of the existing Duo capabilities into Duo Talk before the official release, as stated by GitLab’s Chief Product Officer David deSanto at KubeCon. DeSanto, who plays a prominent role as the company’s spokesperson, mentioned that GitLab has switched its chat backend to Claude from Anthropic. Despite GitLab’s collaboration with Google on various AI initiatives, including code execution services, this change may come as a surprise. DeSanto revealed that they have not yet chosen a specific model for the chat platform, opting instead for an in-house open-source solution that has proven effective in facilitating seamless interactions within GitLab’s ecosystem.
Furthermore, DeSanto disclosed that GitLab is collaborating with Oracle to develop cloud-based GPUs, emphasizing their cloud-agnostic approach. The partnerships with Google, Oracle, and Anthropic are seen as mutually beneficial, with each entity complementing the others in advancing AI capabilities.
Duo Skype presently offers features such as code explanations, script refactoring, and other standard code-related chat functionalities. Additional features like test case technology and risk explanations are set to be integrated soon.
During the testing phase, the importance of the development environment in optimizing the chat experience became evident. GitLab’s ability to access all code, even if not currently open in the IDE, provides a competitive edge in the industry. Moreover, the larger environment windows facilitate better tracking of past conversations.
Feedback from developers indicated a preference for using Duo Chat over traditional code suggestions, as experienced developers found the latter to be distracting. Top developers, in particular, favored code generation within the chat interface. DeSanto noted that while it was initially expected that associated developers would benefit more from the chat feature, inline code completion emerged as a preferred choice, as developers expressed a clear understanding of their code and a preference for guidance or debugging assistance when needed.
