Skip to content

Prompt Engineering Upleveling for GitLab Engineering Teams

Everyone can contribute. Help move this issue forward while earning points, leveling up and collecting rewards.

Overview / Context

Prompt engineering has rapidly emerged as a critical skill for software developers who are building AI-powered features. Although some teams at GitLab have begun to adopt these practices, there’s a need to broaden prompt engineering knowledge beyond the AI-focused groups so that more teams can autonomously design, prototype, and iterate on AI features. This is part of our larger goal to support faster, better, and more effective AI development across the entire engineering organization.

This initiative is intended to:

  1. Uplevel engineers’ ability to create and refine prompts.
  2. Reduce the skill bar for effective prompt engineering through documentation, tooling, and abstraction.
  3. Enable consistent, high-quality AI features—regardless of a team’s prior AI expertise.

References & Related Work

Problem Statement

  • GitLab engineers outside the AI-focused groups often lack hands-on experience with prompt engineering best practices.
  • Current resources are scattered or too general; there is no single, cohesive, “one-stop shop” for engineers to learn how to craft, experiment with, and iterate on prompts.
  • The barrier to entry for building and improving AI features is still high; we want to lower that barrier so any GitLab engineer can confidently contribute to AI-powered capabilities.

Proposed Objectives

  1. Consolidate and Curate Learning Resources
    • Centralize best practices, examples, and references in a single, easily discoverable place.
    • Provide real-world examples of how prompt engineering is applied in GitLab features.
  2. Develop (or Extend) Tooling / Abstraction Layers
    • Streamline how engineers create, test, and iterate on prompts.
    • Provide a consistent approach—templates, libraries, or frameworks—that reduce guesswork.
  3. Hands-On Workshops or Pilot Sessions
    • Organize short but practical sessions where teams can learn prompt engineering by doing.
    • Encourage cross-team collaboration and knowledge sharing.
  4. Ongoing Support / Community of Practice
    • Maintain a Slack channel or recurring office hours for prompt engineering Q&A.
    • Collect feedback from feature teams to continually improve resources and tooling.

Success Criteria

  • Increased Velocity: Faster delivery of AI features across multiple teams, measured by time-to-merge for initial prompt-based features or POCs.
  • Adoption: The number of teams / MRs that adopt the recommended prompt-engineering guidelines.
  • Quality & Consistency: Fewer prompt-related bugs, improved prompt success rates, or higher user satisfaction with AI outputs.
  • Self-Sufficiency: Teams outside AI-Powered subgroups can effectively design and iterate on prompts with minimal external help.

Scope & Deliverables

  • Phase 1: Gather & review existing material (Issue Gather resources and make a pathway to uplevel ... (gitlab-org/ai-powered/discussions#18 - closed) • Gosia Ksionek, Stephan Rayner • 17.4, “How to Prompt” slides, AI Feature Dev Playbook).
  • Phase 2: Propose updates / expansions to the existing documentation and best-practice guidelines:
    • A simplified “Prompt Engineering 101” module.
    • Common pitfalls / frequently asked questions.
  • Phase 3: Identify or build any minimal tooling or example scaffolding (e.g., code snippets, templates, prompts library).
  • Phase 4: Conduct workshops or knowledge-sharing sessions with volunteer engineering groups (potential pilot).
  • Phase 5: Evaluate impact and iterate on the plan.

Key Stakeholders

  • Engineering: All teams that aim to build or enhance AI features.
  • AI-Powered Stage: For in-depth AI domain expertise and potential platform abstractions.
  • Product Managers: Ensuring alignment with overall product vision, especially around GitLab Duo.
  • Technical Writing / Developer Relations: Potentially to help with documentation and training resources.

Open Questions

  1. What is the fastest way to pilot this content?
    • Identify which team(s) want to adopt or test out the initial training material.
  2. How do we measure success beyond velocity?
    • Consider setting up usage metrics or a short feedback survey after engineers experiment with the recommended approaches.
  3. Do we have capacity or existing frameworks for live workshops?
    • Check with Developer Relations or any existing programs that can help.
Edited by 🤖 GitLab Bot 🤖