We’re joined by Pete Kowalczyk, AI Conversation Designer at the Government Digital Service (GDS), to talk about designing AI experiences across the vast and complex landscape of UK government services.
Pete explains how GDS is exploring knowledge graphs, persistent state and goal detection to move beyond basic RAG systems. The aim is to connect information with action and guide people through non-linear journeys, such as moving house or registering a birth. A key goal is to reduce the burden created by fragmented processes and legacy systems.
We discuss why conversational systems need to track user goals, how small interaction loops can be built into rich experiences over time and why transitions between stages of a conversation carry so much weight for trust and usability.
We also get into progressive disclosure, deciding when an AI should stay concise and when it should offer more, and how to judge whether a link belongs in a response at all. Pete points to a bank surfacing patterns in someone’s spending as an example of how organisations can build proactive features from data they already hold, rather than only responding when asked.
Pete shares how the designer role is evolving. Designers now need sufficient technical understanding to prototype and work directly with LLMs through practices like vibe coding, while maintaining the creative ability to understand experiences beyond incremental chatbot improvements. He closes with a warning against the industry’s current resolution rut, where success gets defined too narrowly as fast ticket resolution rather than genuinely useful, proactive AI.
Show notes
Lead with Tempo 2026
London Design Festival
Video of Pete’s talk at the Conversational AI Meetup in London
Take our AI Maturity Assessment
Subscribe to VUX World
Subscribe to The AI Ultimatum Substack
