GitLab 19.4 Expands AI Agents Across DevOps With New Models and Controls
The release comes as companies move from using AI assistants for individual coding tasks to deploying agents across engineering teams.
GitLab has released GitLab 19.4, expanding its agentic automation capabilities across the software development lifecycle while giving organisations more control over how AI agents access tools, consume resources and execute work.
The release comes as companies move from using AI assistants for individual coding tasks to deploying agents across engineering teams. GitLab is addressing this shift with features that allow developers to delegate larger objectives, trigger automation from merge requests and connect external AI agents to GitLab through the Model Context Protocol (MCP).
The Left Shift Bureau
One of the key additions is the /goal command in GitLab Duo CLI. Developers can give an agent a broader objective instead of breaking it into individual tasks. The agent works toward completing the goal and a separate model verifies the result against the original objective at each step.
The developer can stop the process, modify the goal and restart it, while the workflow continues to operate locally under the organisation's existing permissions and rules.
"This release takes agentic automation from something individual developers use to something an organization can scale at speed and under the controls already in place. The platform running the automation is what governs which tools an agent can touch and attributes what it consumes, so extending it to the next team is a measured decision rather than an open-ended risk," said Manav Khurana, GitLab Chief Product and Marketing Officer.
GitLab is also adding three GitLab-hosted open-weight models—Kimi K3, MiniMax M3 and GLM 5.3—to its Duo Agent Platform. The company said the models give teams more flexibility to balance quality, latency and cost, with the hosted models offering up to four times more calls per GitLab Credit than many comparable frontier models.
Group owners can set default models for individual features and control which models are available across teams and projects.
The latest release also expands GitLab's MCP server with tools that allow external agents to trigger pipelines, investigate failed jobs, manage merge requests, update work items and triage vulnerabilities.
Administrators can govern these tools through existing GitLab Duo Agent Platform settings. Read-only tools default to “Always Allow”, while write and delete operations default to “Always Ask”, providing a checkpoint before an agent makes changes.
GitLab is also introducing more detailed usage visibility, including per-user caps and usage exports down to individual billable events.
The release also includes an experimental GitLab Duo Agent Platform integration for Slack, allowing users to interact with GitLab through conversations and turn discussions into tracked engineering work.
Last year, the company launched version 18.7 of its DevSecOps platform, bringing new AI-driven automation, governance controls, and developer enhancements.
The Left Shift Bureau
The Left Shift Bureau
