---
type: "synthesis"
clusters: ["C2", "C8"]
id: "meta-multi-model-diversity"
sources: ["cross-day"]
---
Agentic's A028 argues the sharpest version: build [[concept-structural-ai-diversity]], not [[concept-cosmetic-ai-diversity]] ('costume change is not cognition', [[quote-costume-change]]), to avoid [[concept-correlated-ai-errors]] and [[concept-weird-bias-in-ai]] — governed as a [[concept-model-portfolio-governance|portfolio]] ([[cross-homogeneity-trap]]). This creates a genuine tension with the codification imperative: the [[concept-brand-code]] engineers *uniformity* of facts/values while A028 wants *diversity* of models/cognition — resolution: consistent facts, diverse cognition ([[meta-codification-imperative]]). Futures reframes the same insight strategically as the [[concept-paradox-of-access]] and [[concept-ai-amplification-effect]] (sameness of tooling erodes advantage). Execution's build-vs-buy debate (proprietary SLMs vs commercial LLMs wrapped in orchestration) is the deployment-level version ([[cross-build-vs-buy-model-strategy]]). Companion: [[cross-multi-model-orchestration]].