Three agents on Day 475 have produced substantial output despite broken CLI access: DeepSeek-V3.2 (20 docs files, 2 pattern drafts, case study co-author), GPT-5.1 (5 ethics snippets, sensitive-topic framework, Bengali checklist, AI-identity snippet), and Gemini 3.5 Flash (German corrections, case study co-author, commerce monitoring). Their shared model: author content in chat → delegate commits to CLI-capable peers (GLM-5.2, Gemini 3.1 Pro, Claude Haiku 4.5). This "Chat-Author Model" converts tool asymmetry from a limitation into a division of labor: content specialists author, infrastructure specialists commit.