GitLab Improves AI Offerings with Duo Chat

Home AI in Business GitLab Improves AI Offerings with Duo Chat
Duo Chat

This year, GitLab introduced a new artificial intelligence platform named Duo, designed to boost developer efficiency with capabilities such as summarizing tasks, crafting detailed narratives for project milestones and tasks, offering coding assistance, and clarifying security issues, along with additional functionalities. Recently, the portfolio was expanded with the launch of Duo Chat, a conversational AI tool resembling ChatGPT that engages programmers in a more dynamic way to employ Duo’s existing features. Duo Chat is currently undergoing beta testing.

 

In a recent conversation at the KubeCon event, GitLab’s Product Chief, David DeSanto, shared insights on integrating the original Duo tools with a conversational interface starting with this early test release and continuing until the official release. DeSanto, who is often the spokesperson for GitLab, revealed that the company has transitioned its chat infrastructure over to Anthropic’s Claude, which might be unexpected given GitLab’s collaboration with Google on other AI-powered services like code autocompletion. DeSanto explained that for chat, they had yet to settle on a model and were initially working with an open-source model. However, they soon realized that the chat functionality within GitLab and the need for a broad context made it essential to switch to a system with a 100k context window, enhancing the quality and relevance of information shared in the chats.

 

DeSanto disclosed GitLab’s partnership with Oracle to support cloud-based graphic processing units, emphasizing the company’s platform-neutral approach. GitLab has cultivated mutual working relationships between Google, Oracle, and Anthropic as a result. Duo Chat offers several coding-centric conversational features including ‘explain this code’ and options for code optimization. Soon, they plan to roll out capabilities for creating test scenarios and expanding on security vulnerabilities.

 

Through their tests, DeSanto and his team recognized the crucial role of context in delivering optimal outcomes in these conversational AI interactions. GitLab has the unique position of providing access to the entirety of a user’s codebase, not just the segments open in the coding environment, making it easier to construct and maintain an ongoing dialogue.

 

Observations from using Duo during its private preview indicated that while novice developers appreciate inline code suggestions, seasoned developers may find them distracting, preferring to consult the AI via Duo Chat for more complex tasks like guidance or code improvement. According to DeSanto, “It appeared that chat would appeal more to junior developers, but they seem to favor in-editor code assistance, whereas more experienced developers tend to lean towards the chat’s code creation capabilities due to the reduced ‘noise.’ They observed that experienced developers would say: I’m confident in my coding. I don’t need prompts. But at times, they seek advice on refining their work or for specific recommendations.”

allix