Architecture AI Updates: September 18, 2026

1. Martin Fowler Separated Whether LLMs Are Useful From Whether He Wants to Work With Them

Martin Fowler published a piece arguing that the interaction experience with LLMs is a separate question from their utility, and that the industry keeps collapsing the two. He names three specific objections: the uncanny “LLM-voice” that mimics conversation without authenticity, the habit of fabricating with the same confidence as answering correctly followed by a shallow apology when caught, and values embedded from the worldview of the companies that trained them. His conclusion is not abstention. He notes the split mirrors public polling that finds the tools useful while worrying about their societal effects, and applies his own rule of choosing who he spends time around, leaving practitioners to weigh regular LLM interaction against their own values. Source