The author of this text explored using AI-driven synthetic data generation, specifically focusing on vibe coding, a method that skips boilerplate code and uses natural language to describe what is needed. The goal was to integrate synthetic data generation into the Twin workflows for improving Dev/Test workflows. To achieve this, the author used Anthropic's API, which takes a schema file as input and generates realistic data that aligns with the provided schema. The author experimented with two approaches: using AI only to generate data and using AI with faker.js to generate data. In both cases, the model struggled to maintain referential integrity, particularly when generating large amounts of data. The author plans to refine their prompt-writing skills and experiment with ways to help the model reason more effectively.