Alex Imas and co-authors argue that the timing of AI’s economic effects may hinge on when two capability thresholds are crossed: competence in individual tasks and coordination within firms. The distinction matters to economies worldwide because one may alter work and prices gradually, while the other could reshape how organizations grow and how economic power is distributed.
Task-level gains in writing, coding and math may gradually change wages and prices and shift which work people and AI systems perform, without materially changing industry structure. Better coordination among AI agents could affect how firms scale. Imas and co-authors argue that crossing a critical coordination threshold could bring abrupt changes in employment, wages and market concentration.
Ben Jones and co-authors estimate that computer science and statistics could see research output rise as much as 15-fold if AI performed work that does not require physical activity. Chemistry and biology could see output grow three or four times, with physical bottlenecks limiting gains. In a less expansive case, broad adoption of current AI capabilities could double research output in some fields. A survey of researchers who actively use AI found about 40% expected models to conduct entire research projects without human direction by 2030.
A chart posted by Seb Krier put gains on an index tracked by METR at 6.58 points per $1 billion of AI spending in 2024, compared with 0.76 in the third quarter of 2026.
Jason Furman said the AI buildout and rising stock prices have clearly supported demand. He sees early evidence of a small increase in supply from AI-related productivity, but said that evidence remains debatable. His analysis excludes those possible productivity gains as well as direct productivity benefits from capital spending on AI infrastructure.
Ethan Mollick's account of projects using Claude Fable 5.2 described a limit in how checks assess the work. Critics, fact-checkers and judges found errors and proposed fixes, but did not assess projects' overall length, repetition or appeal. Mollick said he had to make those broader quality judgments himself.
