GitLab Duo

Situation

GitLab customers needed to learn how to use Duo’s AI capabilities in their work. At the same time, the product was evolving quickly and some features were still experimental. The learning path was part of a broader company initiative around AI.

The challenge was to deliver useful instruction without creating a course that would need a complete rewrite whenever the product changed. Learners also needed to practice in a real workflow, rather than only watch isolated feature demonstrations.

Defining the learning problem

I focused the curriculum on what people needed to accomplish: understand code, work through project issues, diagnose a failing pipeline, and respond to a security finding. Those tasks provided a more durable organizing structure than a tour of individual controls.

That distinction shaped both the learning objectives and the maintenance plan. Product details could change within a module while the task and its place in the larger workflow remained useful.

The experience I designed

I built a modular learning path combining explanation, demonstration, knowledge checks, and hands-on practice. The DevOps Swag Store lab used a prepopulated project so learners started with the same issues, pipeline failure, and security scenario.

This was the first program in which our team used prepopulated projects in the lab environment. It made the starting conditions more consistent and directed practice toward the intended workflow instead of project setup.

For example, the pipeline exercise asks learners to examine a failed job, review Duo’s explanation, edit the YAML, and verify that the next run succeeds. The vulnerability demonstration follows the same practical pattern: understand the finding, examine a proposed fix, and review the generated changes.

How I organized delivery

I authored the written content directly in GitLab and used merge requests to collaborate with product managers and technical experts. This brought the review process into the same environment as the source content.

I created independently maintainable modules and focused demos so changes could be addressed where they occurred. Course content and lab activities stayed connected through a shared set of tasks, rather than growing into separate collections of feature explanations.

Impact

The program established a reusable approach to AI enablement: task-based modules, preconfigured practice environments, and a review workflow that included product expertise.

The delivered courses, demos, and lab exercises show how that model worked in practice. Each focuses on a concrete task, and the modular structure allows the relevant instruction to be updated as the product evolves.