Pattern 2: The Productivity Unlock is Real—But Contested

Multiple sources now converge on a measurable productivity impact from AI. But the actual gains depend on a critical variable: task complementarity.

The Estimates

Source	Methodology	Estimate
Richmond Fed (optimistic)	Full AI adoption, no constraints	7%
Anthropic (baseline)	Measured task speedup extrapolation	1.8pp
Anthropic (success-adjusted)	Conservative adjustment for errors	1.0-1.2pp
Lombard Odier	Macro-economic modeling	+0.25pp
Penn Wharton (conservative)	Limited task automation scope	0.2pp

The baseline Anthropic estimate of 1.8pp would return US productivity growth to late-1990s levels—the last time America experienced a sustained productivity boom.

The Elasticity Question

The economic variable σ (sigma) measures whether AI tasks complement or substitute for human tasks:

If σ < 1 (Complements): Productivity constrained to 0.6-0.8pp

• AI makes certain tasks faster, but humans still do the rest
• Time savings don’t translate to headcount reduction or output increase
• Example: AI makes lesson planning 12x faster, but teachers still spend the same time in classrooms

If σ = 1 (Baseline): Productivity reaches 1.8pp

• AI time savings translate proportionally to productivity gains
• Organizations capture the efficiency

If σ > 1 (Substitutes): Productivity rises to 2.2-2.6pp

• AI handles entire task chains end-to-end
• Humans freed for other work or headcount reduced
• Approaches late-1990s productivity boom levels

The Complement Trap

The teacher example matters. Faster lesson planning doesn’t automatically mean fewer teachers or more students per teacher. The time savings might just disappear into the existing structure. Whether we’re in complement or substitute territory will determine if this is a modest productivity boost or a structural economic transformation.

---

This is part of a comprehensive analysis. Read the full analysis on The Business Engineer.