---
id: "cp-agents-learn-norms-from-data"
type: "counter-perspective"
source_timestamps: ["Enrichment: Counter-Perspectives §3"]
tags: ["counter-perspective", "ai-capabilities", "fine-tuning", "memory"]
related: ["claim-agents-cannot-infer-context", "quote-agents-operate-on-explicit"]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-new-27-teach-ai-your-decisions"
sourceUrl: "https://hbr.org/2026/06/teach-your-ai-how-you-make-decisions"
sourceTitle: "Teach Your AI How You Make Decisions"
---
# Counter: Agents Can Learn Some Norms from Data and Interaction

**Nuance / counterpoint (technical perspective):** The article's claim that agents cannot absorb norms through observation ([[claim-agents-cannot-infer-context]], [[quote-agents-operate-on-explicit]]) is slightly absolutist. LLM-based agents can be fine-tuned on organizational correspondence, policy logs, and decision histories; with RLHF, long-term memory, and telemetry, they can adapt behavior via feedback loops and learn patterns in how issues are handled.

**Why it qualifies the source:** Norms reflected in data traces can be learned even if not written as explicit policies; over time agents might approximate some "tacit" norms consistently expressed in interaction logs.

**Where it still agrees:** This learning is mediated by **explicit signals** (labels, ratings, prompts, structured traces) and requires intentional data curation, alignment, and feedback — not human-like cultural immersion. So it softens the phrasing but largely upholds the practical point that explicit structuring and governance are needed.