AI Architecture Updates: June 18, 2026

1. PRINCE: Patterns for Building Reliable Agentic AI Systems

Sarang Sanjay Kulkarni (Thoughtworks), on martinfowler.com. A new case study describes PRINCE, a Bayer preclinical research platform that lets scientists query decades of study reports, and distills its reliability into two engineering concepts. Context engineering routes the right context to the right capability at the right time rather than loading everything into one prompt, feeding planning, retrieval, evidence, and synthesis stages separately. Harness engineering supplies the control layer around the models via LangGraph orchestration, with state persistence in PostgreSQL, cross-provider LLM fallbacks, multi-level retries, human intervention points, and observability through Langfuse. The agentic RAG workflow splits work across Clarify Intent, Think and Plan, Researcher, Reflection, and Writer agents, arguing that reliability comes from engineering both what the model sees and the harness within which it acts. Source

2. Fragments on LLM Programming, Domain-Driven Design, and Context Discipline

Martin Fowler, on martinfowler.com. Fowler’s June 16 Fragments collects practitioner observations on building with LLMs that bear on system design. It highlights Chelsea Troy’s framing of four conversation registers with an LLM, exploring, brainstorming, deciding, and implementing, and her advice to start fresh conversations when switching registers to keep context windows healthy. Fowler also argues that Domain-Driven Design becomes more important amid AI-driven change, citing Eric Evans’ DDD Europe 2026 experiments, and surfaces Charity Majors’ point that AI enthusiasts and skeptics are both right, making engineering discipline essential as teams ship faster. The throughline is that AI amplifies existing practices and rewards deliberate workflow and context management rather than ad hoc prompting. Source