Update from data repository

Author: github-actions

_publications/broken-ladders.md

@@ -0,0 +1,22 @@
+---
+cite: |-
+  Bárány, Zsófia L., and Miklós Koren. 2025. "Broken Ladders: AI, Teamwork, and the Dynamics of Skill Formation in the Workplace." Working Paper.
+links:
+  - text: "PDF"
+    url: "https://cdn.thnk.ng/pdf/broken-ladders/paper.pdf"
+statement: ""
+team:
+  - "barany"
+  - "koren"
+grants:
+  - erc-advanced-2022
+  - elvonal
+title: "Broken Ladders: AI, Teamwork, and the Dynamics of Skill Formation in the Workplace"
+date: "2025-05-01"
+tags:
+  - working
+  - macromanagers
+description: "We study the efficiency of AI adoption in knowledge work using a model of team production and learning. Seniors pick higher-value problems and solve the hardest tasks; juniors learn by working with them. AI can solve problems at a lower cost than juniors, but it lacks the ability to recognize the value of the problems it is solving. Its output has to be reviewed by a human; this supervision time is an important bottleneck for the productivity gains from adoption. AI raises output in the short run, but it can limit learning and decrease output in the long run. Inefficient AI adoption can be the result of seniors not internalizing the value of the mentoring they provide, or of AI or AI-using juniors crowding out (other) juniors from learning from seniors. Whether AI adoption is dynamically inefficient depends on the static gains from AI, and the magnitude of dynamic gains from learning."
+---
+
+We study the efficiency of AI adoption in knowledge work using a model of team production and learning. Seniors pick higher-value problems and solve the hardest tasks; juniors learn by working with them. AI can solve problems at a lower cost than juniors, but it lacks the ability to recognize the value of the problems it is solving. Its output has to be reviewed by a human; this supervision time is an important bottleneck for the productivity gains from adoption. AI raises output in the short run, but it can limit learning and decrease output in the long run. Inefficient AI adoption can be the result of seniors not internalizing the value of the mentoring they provide, or of AI or AI-using juniors crowding out (other) juniors from learning from seniors. Whether AI adoption is dynamically inefficient depends on the static gains from AI, and the magnitude of dynamic gains from learning.