Informatica's VP of Engineering Nancy Chen explains how the team built Copilot, an AI-powered tool that lets users generate data integration pipelines using natural language. Since launching in May 2025, customers have generated ~10,000 pipelines and submitted 25,000+ expression-generation requests, with ~60% accepted without modification. The team initially fine-tuned their own models but switched to OpenAI as foundation models matured. Key engineering challenges included prompt engineering, metadata grounding, output validation, and adapting to probabilistic AI behavior versus deterministic software. The team found that accuracy improvements came more from better context and guardrails than from larger models. Future plans include agentic AI workflows and code-first experiences.
