GitLab Enhances AI Offerings with Introduction of Duo Chat

Earlier this year, GitLab launched Duo, a new suite of AI features designed to enhance developer productivity. Duo helps users by summarizing issues, generating descriptions for epics and tasks, and providing code suggestions, along with explanations for vulnerabilities. Today, GitLab expanded this offering with Duo Chat, a ChatGPT-like experience that enables developers to interact more dynamically with existing Duo features. Duo Chat is currently in its beta phase.

In a recent interview at KubeCon, David DeSanto, GitLab’s Chief Product Officer, shared that the goal is to integrate more capabilities from Duo into the chat feature by the time it transitions from beta to general availability.

DeSanto, emerging as a prominent spokesperson for GitLab, also revealed that the company has migrated its chat backend to Anthropic's Claude. This shift might surprise some, considering GitLab's partnership with Google for other AI functionalities like code completion. “We hadn’t selected a specific model for chat,” DeSanto explained. “Initially, we utilized our own open-source model. However, we realized that, to effectively manage the chat context within GitLab, a 100k context window significantly enhances our information retrieval."

Furthermore, DeSanto mentioned GitLab's collaboration with Oracle for cloud-based GPUs. “We’re cloud-agnostic, seeking the best solutions. All three providers—Google, Oracle, and Anthropic—integrate seamlessly with our systems, fostering a productive partnership.”

Currently, Duo Chat offers a variety of code-related features, including ‘explain this code’ and code refactoring. Additional capabilities such as test case generation and vulnerability explanations will be introduced shortly.

DeSanto emphasized the critical role of context during testing, noting that it directly impacts the quality of the chat experience. Unlike other platforms, GitLab can access all of a developer's code, even if it isn't opened in their IDE, and the larger context windows help maintain continuity in conversations.

During a closed preview phase, the team observed interesting user behavior: seasoned developers often felt overwhelmed by code suggestions, preferring to navigate AI tools via Duo Chat instead.

“I initially thought chat would cater more to junior developers, but it turns out that they prefer inline code completion,” DeSanto noted. “Senior developers tend to gravitate towards code generation in chat. They sometimes find suggestions noisy and unnecessary but still seek guidance or assistance with tasks like refactoring.”

GitLab’s latest security feature utilizes AI to clarify vulnerabilities for developers, making it easier for them to understand and address potential issues.