---
id: "quote-llm-vs-lam"
type: "quote"
source_timestamps: ["§ The Everything Engine Needs Your Data"]
tags: ["ai-evolution", "task-execution"]
related: ["concept-large-action-models"]
speaker: "Amy Webb"
speakers: ["Amy Webb"]
quote: "If LLMs predict what to say next, LAMs predict what should be done next, breaking down complex tasks into smaller pieces. Unlike LLMs that primarily generate content, LAMs are optimized for task execution..."
sources: ["futures"]
sourceVaultSlug: "hbr-seg-futures"
originDay: 2
articleStem: "hbr-foci-73-living-intelligence"
sourceUrl: "https://hbr.org/2025/01/why-living-intelligence-is-the-next-big-thing"
sourceTitle: "Why “Living Intelligence” Is the Next Big Thing"
---
# LLMs vs LAMs

> "If LLMs predict what to say next, LAMs predict what should be done next, breaking down complex tasks into smaller pieces. Unlike LLMs that primarily generate content, LAMs are optimized for task execution..."
> — [[entity-amy-webb|Amy Webb]]

**Context:** A succinct differentiation between the current generation of *generative* AI (LLMs) and the upcoming generation of *agentic* AI ([[concept-large-action-models|LAMs]]). The "say next" vs. "do next" contrast is the single most quotable definition of the LAM concept.

> *Enrichment caveat:* The distinction is coherent, but the broader AI field would typically frame this shift as *agents*, *tool use*, and *computer-use models* rather than as a new model class named "LAM."