---
id: "claim-agents-cannot-infer-context"
type: "claim"
source_timestamps: ["¶5"]
tags: ["ai-limitations", "context", "culture"]
related: ["concept-codifying-judgment", "claim-bottleneck-is-explicit-judgment"]
confidence: "high"
testable: true
speakers: ["Jen Stave", "Ryan Kurt", "John Winsor"]
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"
---
# AI agents cannot infer context or absorb norms through observation

**Claim (confidence: high, testable):** Unlike human employees who gradually internalize how an organization thinks by watching, listening, and experiencing the culture, AI agents lack the capacity to absorb norms through observation. They operate strictly on what is made explicit to them.

This creates a specific failure mode: agents deployed without codified judgment eventually go off track and misalign with firm goals because they lack the inferred context necessary to make correct decisions in ambiguous environments. It is the direct rationale for [[concept-codifying-judgment]]. See the anchoring quote [[quote-agents-operate-on-explicit]].

**Enrichment assessment — conceptually sound, slightly overstated if taken absolutely.** LLM-based agents operate on training data plus provided context, not human-style cultural observation; agentic frameworks require operational clarity, explicit decision rights, and unified data. The nuance: agents *can* pick up some tacit patterns indirectly when those patterns appear in data they are fine-tuned or aligned on (e.g., RLHF, decision logs, long-term memory) — but this is still mediated by explicit signals, not unstructured cultural osmosis, and requires intentional curation. See [[cp-agents-learn-norms-from-data]].