### Background Following the successful completion of the [POC phase](https://gitlab.com/gitlab-org/ai-powered/ai-framework/team-hq/-/issues/45 "POC phase"), this epic encompasses the work streams and items required to develop and ship the first version of Duo Workflows. This epic will guide the project through its Dogfooding/Alpha, Closed Experimental, and Open Experiment phases. ### Key Components 1. Workflow Service: A Python-based engine orchestrating workflow logic and state management. 2. Workflow Executor: A runtime entity for safe action execution, either locally or in a CI environment. 3. Human-in-the-Loop Features: Enabling user interactions within workflows for enhanced control and oversight. ### Goals 1. [Dogfooding/Alpha (Internal)](https://gitlab.com/groups/gitlab-org/-/epics/15201 "Duo Workflow - Dogfooding/Alpha Release") - Develop a version of Duo Workflow for internal testing and refinement. - Generate interest and buy-in for GitLab's AI vision among team members. 2. [Closed Experiment](https://gitlab.com/groups/gitlab-org/-/epics/15046 "Duo Workflow - Closed Experiment Release") - Create a version of Duo Workflows for limited customer testing. - Gather initial feedback from select customers in controlled environments. 3. [Open Experiment](https://gitlab.com/groups/gitlab-org/-/epics/15200 "Duo Workflow - Open Experiment Release") - Launch a publicly accessible version of Duo Workflows, labeled as an "Experiment". - Collect broader user feedback and validate the solution's effectiveness. ### Focus Use Cases Our primary focus for testing and user articulation will be: 1. Coding tasks - Enable Duo Workflow to read issues / MRs and write code based on user goal- refactor, bootstrap a project, add functionality or tests 2. Enhancing code quality - Improve test coverage, manage dependency upgrades, assist with refactoring, and aid in bug fixing. Note: While optimizing for these use cases, Duo Workflow will remain flexible for other applications. ### Definition of Done - Goal: Ship an experiment enabling observation of user reactions and engagement in a production setup. - Primary Use Case: Coding tasks - Focus: Introduce a first use case as an experiment, maintaining handcrafted agent usage. - Rationale: Learning from actual usage is critical to understand user interactions, trust levels, and important features. - Criteria: Successfully deploy a workflow that fixes a broken pipeline, operabl from the vscode IDE extensions, with sandboxed local execution, user confirmation for commands, documentation, and performance tracking. ### Release Iterations 1. Dogfooding/Alpha (17.4): - Ship core Workflow Service and Executor - Implement basic UI and API integrations - Enable Docker support and GitLab checkpointing 2. Closed Experiment (17.5): - Enhance Dev Container support - Improve GitLab checkpointing UI and functionality - Implement initial human-in-the-loop features 3. Open Experiment (17.6): - Expand IDE integration and support - Integrate with CI Pipeline and GitLab Workspaces - Refine and optimize based on early user feedback Upon completion of these iterations, we will officially enter the Experiment state and begin preparations for the Beta phase. ### Key Metrics and Success Criteria - User engagement rates - SWE-Bench - User satisfaction scores - Performance and reliability metrics ### Links - [Duo Workflow Board](https://gitlab.com/groups/gitlab-org/-/boards/7545476?label_name%5B%5D=Category%3ADuo%20Workflow) - [GitLab Experiment Support Policy](https://docs.gitlab.com/ee/policy/experiment-beta-support.html)