Building Reliable Agentic Systems: Lessons from Front Lines of Data Engineering

Talk Title:
Building Reliable Agentic Systems: Lessons from the Front Lines of Data Engineering, by Julian Wiffen, Chief of AI and Data Science – Matillion
Abstract:
Moving beyond the initial excitement of generative AI requires a fundamental shift from isolated experiments to integrated, agentic systems. This talk explores the technical and organizational lessons learned from building a production-grade digital data engineer - a system designed not as a simple chatbot, but as a holistic model for user interaction within complex data environments.
A central pillar of this approach is the power of constraining a large language model’s output. By forcing probabilistic models into binary or categorical decisions - the "Yes or No" constraint - we can reliably transform messy, unstructured data into structured insights and programmatically pull levers within software applications. This methodology moves AI from providing general advice to executing precise, self-correcting workflows that bridge the gap between natural language intent and deterministic execution.
The presentation will detail how the adoption of agentic tools fundamentally changes the data team. When agents handle the messy prep work and repetitive orchestration, the human role evolves from manual script-writing to knowledge curation and strategic oversight. We will discuss the necessity of establishing a rigorous Evaluation Gym to move past vibes-based testing toward the statistical verification required for non-deterministic systems. Attendees will gain a practical framework for building reliable agentic loops and a roadmap for managing machine-scale data operations where human expertise is focused on context more than code.
Bio:
Julian leads AI research and Data Science at Matillion, where he is building Maia - widely recognised as the world’s first AI Data Engineer.
He focuses on delivering real-world GenAI at scale, architecting agentic systems that operate inside live enterprise data workflows with governance, guardrails and measurable productivity embedded from day one. His approach is pragmatic and production-led - building AI capabilities that drive operational impact.
Under his leadership, Matillion has accelerated its AI roadmap by years and delivered significant productivity gains across global enterprises, positioning the company at the forefront of AI-driven data engineering.
Julian holds a Master’s degree in Artificial Intelligence from the University of Oxford, where his research explored genetic algorithms capable of deriving complex relationships from complex datasets - foundations that continue to shape his approach to intelligent systems today.