# AI Pipeline Builder – Q1 MVC - Beta ## 1. Summary We are delivering an **AI‑powered pipeline starter** that eliminates the “blank page” problem for new‑to‑CI projects in GitLab. In Q1 FY27, this takes the form of a **Duo‑guided pipeline creation user flow** backed by a **Pipeline Authoring foundational agent** (not a full visual editor) aimed to be released as Beta in 18.11. This epic tracks the work needed to: - Help new‑to‑CI users create their **first working pipeline** in minutes (from days/weeks). - Establish the **Pipeline Authoring agent** as the CI/CD expert behind that experience. - Lay the foundation for future iterations (modification, optimization, migration, troubleshooting and richer visualization). --- ## 2. Problem - Today, **~74% of GitLab projects never create a first CI pipeline**, even though those projects have code and could benefit from CI. - New‑to‑CI users hit a **“blank page” barrier**: - They don’t know *where* to start (multiple entry points, confusing options). - They don’t know *what* to write in `.gitlab-ci.yml`. - They don’t fully understand what their pipeline does once created. This results in stalled CI adoption, slow time‑to‑value, and missed expansion opportunities. --- ## 3. Q1 FY27 – MVC Scope for Beta release ### 3.1 Target user Primary: Developers or maintainers responsible for getting a *first working CI pipeline* running in a given project. - Project has no CI pipeline today: - Truly 0 pipelines created - Existing or newly imported project with code in GitLab. - The user wants to get to a **reliable build + test pipeline** quickly, without first learning all CI syntax and advanced patterns. - They are comfortable owning the repo, but are **not CI experts** and don’t want to start from a blank `.gitlab-ci.yml`. ### 3.2 Experience: AI‑powered pipeline starter (Duo‑guided) **High‑level flow (Q1 MVC):** 1. **User discovers CI entry point** - From project overview / Build area / similar starting point, they see a clear **“Create pipeline with GitLab Duo”** call‑to‑action. 2. **Duo‑guided creation** - Duo, backed by the **Pipeline Authoring agent**, analyzes repository context (language/framework, basic structure). - Duo proposes a **build + test pipeline** suitable for the project (no CD in Q1). 3. **Review & understanding** - User is shown an annotated `.gitlab-ci.yml` explaining the major pieces (install, test, build). - The pipeline can be visualized in the existing pipeline visualization tab to help users understand structure before committing. 4. **Commit & run** - User commits the generated `.gitlab-ci.yml`. - Pipeline runs via the existing CI execution flow. - Commit is possible directly from the visualization This MVC **reuses the existing Pipeline Editor + visualization** rather than introducing a net‑new visual builder UI in Q1. ### 3.3 Agent: Pipeline Authoring foundational agent The AI behaviour is powered by a **specialized CI/CD expert agent**: - **Pipeline Authoring agent** (AI Catalog ID `2845`). - Registered as a **foundational agent** in GitLab (behind a feature flag). - Responsibilities in Q1: - Generate a syntactically valid, runnable `.gitlab-ci.yml` for supported languages/frameworks. - Explain CI/CD syntax and configuration decisions when requested. - Validate generated configuration using existing linting capabilities. --- ## 4. Explicitly Out of Scope in Q1 To stay focused on **0 → 1 activation**, we explicitly **exclude** the following from this epic’s Q1 deliverable: - ❌ **Full visual pipeline editor / drag‑and‑drop** pipeline building. - ❌ **CD / deployment configuration** (environments, rollout strategies, etc.). - ❌ **Advanced pipeline optimization** (caching, parallelization strategies, pipeline cost optimization). - ❌ **Composing from CI/CD Catalog components, templates, or CI Functions/Steps** in the first iteration. - ❌ **Complex YAML patterns** (deep includes, extends chains, advanced rules) beyond what’s needed for a simple build + test pipeline. These may be addressed in subsequent phases once the activation problem is meaningfully improved. --- ## 5. Q1 Success Metrics We use the existing CI Section and AI starter metric framework instead of inventing new ones. At a high level: - **Activation (Primary):** - Increase **first‑pipeline creation rate** for the target cohort (projects with code, 0 pipelines). - Baseline: ~26%. - Q1 target: **mid‑30s%** (e.g., ~36%) for that cohort. - **Time‑to‑first‑pipeline:** - **Target:** <15–18 minutes from project creation/import to first successful pipeline run for users who complete the AI‑guided flow. - **Quality of generated config:** - **95% of generated pipelines are syntactically valid YAML** (can be committed and executed). - Failures should be attributable to project code/tests/environment rather than invalid CI syntax. We can refine exact numeric targets in coordination with Analytics and the CI Agents program doc, but the **shape** of the metrics should remain consistent with the above. --- ## 6. Dependencies and Related Work - **CI Section AI Agents Program – Pipeline Builder Agent** Program‑level plan & metrics for the Pipeline Builder / Authoring agent and future Fix‑a‑Pipeline / optimization agents. - **Pipeline visual canvas / next‑gen authoring discussions** This epic focuses Q1 on AI‑powered activation using existing pipeline editor + visualization. Future phases may revisit richer visualization alternative editors once the 0 → 1 problem is addressed. --- ## 7. Future Phases (Post‑Q1, High‑Level) The following are **future** directions, not part of the Q1 delivery: - AI‑assisted **pipeline modification & optimization** (beyond initial creation). - Deeper integration with **CI/CD Catalog components, templates, and CI Functions/Steps**. - Enhanced **visualization** and potential new visual builder experiences where technically feasible. - **Migration flows** (e.g., from GitHub Actions or Jenkins) built on top of the same agent capabilities. --- ## 8. Definition of Done (Q1 MVC) - Pipeline Authoring agent is available as a **foundational agent** and wired into Duo. - A new or improved **Duo‑guided “Create pipeline” flow** is live for eligible projects. - The flow: - Generates a **valid `.gitlab-ci.yml`** for supported languages/frameworks. - Allows users to **review and understand** what was generated. - Commits and runs pipelines successfully using existing CI execution. - Instrumentation is in place for the primary metrics: - First‑pipeline creation rate for the target cohort. - Time‑to‑first‑pipeline. - Share of generated configs that are syntactically valid and runnable.