GitLab Launches Its AI Agent Platform in Public Beta

Virtually every software development platform is currently working on how to integrate AI agents into the software lifecycle. With Duo and Duo Chat, GitLab started its AI journey in 2023, earlier than many of its competitors. It’s no surprise then that the company is also bringing AI agents to its DevSecOps platform. First announced earlier this year, the GitLab Duo Agent Platform is now in public beta.

In a way, GitLab is following a general trend here, where AI-powered developer tools have gone from first providing code completion and chat answers to now doing some work — autonomously or semi-autonomously — in the background.

Working asynchronously is something GitLab also stresses with today’s release. The company notes that its focus is on offering an orchestration layer that allows specialized AI agents to work together and in tandem with human developers. Because GitLab is typically already the system of record for its users’ software development teams, the company has the advantage of being able to give its agents access to a lot of context (though with the advent of the Model Context Protocol, it’s now easier than ever for agents to pull in third-party data as needed, too).

“As we worked with many customers, eventually, we saw that the world, especially in the context of software development, is advancing super quickly,” Emilio Salvador, the VP of Strategy and Developer Relations at GitLab, told me. “We see more and more this idea of agents and developers working together, right? We have a unique value there, because we’ve been solving that problem for years at GitLab. And it makes total sense for us to build on the existing platform that we have and enable that future environment where we’ll see both agents and humans working together in both synchronous and asynchronous ways.”

Salvador noted that one of the main principles for the new platform was to make it extensible. That means users can build their own agents and create their own workflows. Out of the box, some specialized agents like a code reviewer, a security expert, a planner and a software developer will be included in the platform. These agents will be able to set up projects, create implementation plans, execute them, and then generate tests to make sure everything looks good.

Like with similar agentic platforms, developers can assign specific tasks to their agents, and those tasks execute in the background, either to completion or until the agent needs feedback from the developer. But developers can also define triggers so that when something happens on the platform — say a developer checks in some new code — a workflow is activated to perform a series of tasks with the help of its associated agents.

“Our goal here is to augment developers,” Salvador said. “And I feel that there’s an opportunity, as those agents evolve, for developers also to be able to customize those agents. Think of it as: hey, you have someone working for you. And eventually, what you want is to complete a task that you don’t have time for — within guardrails — and you want that done as soon as possible. But you don’t have time for that. That’s kind of how we envision that. Eventually, agents will be able to take on more and more things, but we having that orchestration capability is fundamental for the developer workflow to work seamlessly.”

The GitLab Duo Agent Platform is now in public beta and available to GitLab Premium and Ultimate customers. In its current iteration, it includes the ability to orchestrate multi-agent workflows , as well as agentic chat in the IDE that can take feedback and that allows users to quickly delegate tasks with slash commands like /explain, /tests or /include.

This chat feature now also supports JetBrains IDEs like IntelliJ, PyCharm, GoLand and WebStorm.