Because 8 ≈ e², Anthropic's researcher uplift is plausibly >2x

According to a LessWrong analysis, Anthropic's contributors merged eight times as much code per day in the second quarter of twenty twenty-six compared to the twenty twenty-one through twenty twenty-four period. Analyst Thomas Kwa applies economic production models to infer that researcher uplift—overall effective output—increased by more than double. The estimate assumes code quality remains constant and focuses on coding productivity alone. However, Kwa acknowledges substantial uncertainties: AI-generated code could be inflated through verbosity, produce low-value implementations that wouldn't otherwise exist, or represent make-work unrelated to research. Others at the Machine Intelligence Research Institute dispute the modeling assumptions. The true productivity multiplier likely depends on detailed internal Anthropic data not yet public.

Listen to this story

Hear this and more stories in a personalized audio briefing.

Open The Chonkerton