Proof-of-Concept for Accelerating Dutch Policy-Making through Agentic
AI
About the Project
We developed the Policy Pipeline Simulator (PPS) as a proof-of-concept (POC)
that demonstrates how AI agents can simulate and improve the complex Dutch policymaking process. The simulation focuses on the Woondeal framework, the Netherlands’ collaborative mechanism for tackling the national housing crisis.
The Dutch Government: BZK & VRO
The project was developed by Team Epoch in partnership with two Dutch government ministries; The Ministry of the Interior and Kingdom Relations (BZK) and the Ministry for Housing and Spatial Planning (VRO). Both ministries are centrally involved in addressing the national housing crisis through the collaborative Woondeal process, which deals with policy complexity, housing affordability, and sustainable spatial development of cities and regions. Their shared goal for the simulator is to explore how AI agents can accelerate and improve the complex policy-making procedures in the spatial and housing sectors by visualizing bottlenecks and trade-offs.
Relevance
The Dutch government faces a dual challenge of limited administrative capacity and increasing policy complexity, particularly in spatial and housing sectors. Procedures are often labor-intensive and responsibilities fragmented. This simulation is relevant because it tests the feasibility of using AI to accelerate the creation of housing deals like the Woondeal Utrecht 2025–2030, aims to reduce the national housing shortage to 2% by 2031, and provides a tool for “what-if” scenarios, allowing policymakers to visualize bottlenecks and trade-offs.
“The Policy Pipeline Simulator establishes the foundational claim that an AI-driven government is not only possible but necessary—one that operates in a democratically legitimate, ethically responsible, and administratively effective way within the policy process.”
Technical Details
The simulator is a multi-agent system built primarily in Python and hosted on Google Cloud. It follows a four-phase workflow. The Research Phase uses Retrieval-Augmented Generation (RAG) to ground AI responses in factual data from official government documents and web sources. The Discussion Phase follows, where specialized agents represent different stakeholders (e.g., Finance, Legal, Municipalities) and negotiate policy proposals. The Reporting Phase synthesizes findings into official policy drafts, such as a simulated Parliamentary Letter (Kamerbrief). Finally, the Reflection Phase is a stage where users can trace decisions and adapt proposals based on feedback. The system utilized advanced Large Language Models, including Gemini and GPT-4.
Final Presentation
We presented the Policy Pipeline Simulator to over 200 ministry officials at the Ministry of the Interior and Kingdom Relations, including the Secretary General of BZK — making this one of the highestprofile student project demonstrations in Epoch’s history.
