---
id: "claim-acemoglu-underestimate"
type: "claim"
source_timestamps: ["¶4", "§ Information Systems"]
tags: ["productivity", "economic-forecast"]
related: ["entity-daron-acemoglu", "concept-electricity-factory-analogy"]
confidence: "high"
testable: true
speakers: ["Harang Ju"]
sources: ["agentic"]
sourceVaultSlug: "hbr-seg-agentic"
originDay: 6
articleStem: "hbr-ext-17-workplace-set-up-for-agents"
sourceUrl: "https://hbr.org/2026/01/is-your-workplace-set-up-for-ai-agents"
sourceTitle: "Is Your Workplace Set Up for AI Agents?"
---
# Acemoglu's 0.5% productivity estimate is a floor, not a ceiling.

Nobel laureate [[entity-daron-acemoglu|Daron Acemoglu]] estimated that AI will add only ~0.5% to productivity over the next decade. Ju claims this is a significant underestimate because Acemoglu's model assumes AI merely automates tasks within existing, human-centric structures. When complex analysis is rewired for agents (months → minutes), improvements can be thousand-fold, making 0.5% 'the floor for doing nothing differently, not the ceiling for what's possible' (see [[quote-acemoglu-floor]]). This rests on [[concept-electricity-factory-analogy|the electricity factory analogy]] and is stated more sharply as the contrarian insight that [[contrarian-acemoglu-estimate|Acemoglu's estimate is fundamentally flawed]].

**Confidence:** high (as asserted) · **Testable:** yes.

**Enrichment / validation:** Acemoglu and co-authors actually estimate ~0.4–0.6 *percentage points* of annual labor-productivity growth over the decade — the article's ~0.5% is directionally accurate. But 'thousand-fold' goes well beyond current evidence: human–AI field experiments raise output per worker ~50%; agentic research frameworks (Agent Laboratory) report ~84% cost reductions — large, but closer to one order of magnitude. Historical general-purpose technologies (electricity, computing, the internet) produced transformative gains over decades, not immediate thousand-fold jumps. A domain expert reads 'thousand-fold' as a metaphor for extreme *local* task speedup (months → minutes), not a validated economy-wide multiplier.