AI Architecture Updates: May 6, 2026

1. Martin Fowler Highlights Lattice Framework for Disciplined AI Coding

Architecture. In his May 5 fragments post, Martin Fowler points readers to Rahul Garg’s Lattice — an open-source framework that bakes engineering disciplines (tests, structure, review checkpoints) directly into AI coding-assistant workflows rather than leaving them as prompt afterthoughts. Fowler ties it to ongoing work on Structured Prompt-Driven Development and warns that without such guardrails, agentic coding risks recreating the “tar pit” of poor internal quality that historically plagued large systems. Source

2. Fowler Flags the Local-vs-Cloud Model Tradeoff for Architects

Architecture. The same fragment post contrasts Apple’s roughly 10%-of-revenue capex on AI infrastructure against competitors spending 50–75%, framing it as a deliberate bet on local, on-device models versus cloud-hosted frontier ones — a split with direct implications for how architects design coding-agent pipelines, latency budgets, and data-residency boundaries. Fowler positions the choice as architectural, not just procurement. Source