# From templates to AI agents [View the visual case study](https://joshdunne.design/work/aws-pipeline-ai) Generate a starting point. Inspect every step. Keep a path back to code. AWS SageMaker ยท May to October 2025 Role: Senior Product Designer Team: 12 developers and 1 product manager Duration: 5 months Scope: Strategy, working prototype, and production design ![SageMaker Pipeline Designer ready to build a machine learning pipeline with AI](https://joshdunne.design/_next/static/media/generated-pipeline-canvas.13d7dced.webp) The AI-first Pipeline Designer starts with intent instead of low-level pipeline configuration. ## Outcomes 17 Pipeline step types. Each required detailed configuration and connections 4 Agent tools prototyped. Generation, editing, configuration, and validation 3 mo Prototype to production design. Including user and leadership validation ## From fixed templates to flexible workflows. Problem: Static templates could not cover each team's workflow. Decision: I proposed an agent that composes pipelines and tested it in working code. Tradeoff: Generation needed an escape hatch: inspect, edit and continue in a notebook. Project outcome: Leadership approved the direction and production designs were delivered. The 50% speed improvement remained a post-launch estimate. ## The starting point The original template gallery packaged fixed workflows. ![The original template gallery packaged fixed workflows.](https://joshdunne.design/_next/static/media/template-gallery.57033141.webp) The original template gallery packaged fixed workflows. ## Ask before acting Production design. Clarify the data, scale and goal before generating a pipeline. ![Production design. Clarify the data, scale and goal before generating a pipeline.](https://joshdunne.design/_next/static/media/agent-clarifying-questions.0f66897f.webp) Production design. Clarify the data, scale and goal before generating a pipeline. ## Make every step visible Production design. The generated pipeline stays connected to the conversation. ![Production design. The generated pipeline stays connected to the conversation.](https://joshdunne.design/_next/static/media/generated-pipeline-canvas.13d7dced.webp) Production design. The generated pipeline stays connected to the conversation. ## Keep the details editable Production design. Inspect and adjust a selected step without leaving the workflow. ![Production design. Inspect and adjust a selected step without leaving the workflow.](https://joshdunne.design/_next/static/media/step-configuration.4aa357b7.webp) Production design. Inspect and adjust a selected step without leaving the workflow. ## From canvas to code ### Continue in a notebook Export the pipeline when the work needs finer control. ![Export the pipeline when the work needs finer control.](https://joshdunne.design/_next/static/media/notebook-handoff.c104c490.webp) Export the pipeline when the work needs finer control. ### Review the changes Review proposed code changes before accepting them. ![Review proposed code changes before accepting them.](https://joshdunne.design/_next/static/media/notebook-review.d5d1920f.webp) Review proposed code changes before accepting them.