This study evaluates the effect of generative artificial intelligence (AI) on software developer productivity via randomized controlled trials at Microsoft, Accenture, and an anonymous Fortune 100 company. These field experiments, run by the companies as part of their ordinary course of business, provided a random subset of developers with access to an AI-based coding assistant suggesting intelligent code completions. Although each experiment is noisy and results vary across experiments, when data are combined across three experiments and 4,867 developers, our analysis reveals a 26.08% increase (standard error: 10.3%) in completed tasks among developers using the AI tool. Notably, less experienced developers had higher adoption rates and greater productivity gains. This paper was accepted by Jean-Pierre Dube, marketing. Funding: M. Demirer and T. Salz thank the MIT GenAI Initiative for funding. Supplemental Material: The online appendices and data files are available at https://doi.org/10.1287/mnsc.2025.00535 .
As the economy digitizes, menu costs fall, and firms can more easily monitor prices. These trends have led to the rise of automated pricing (and repricing) tools. We employ a novel e-commerce data set to examine the effect of algorithmic pricing in the wild. Evidence from an event study suggests that firms that start employing repricing tools drop their prices by 16.93%, with market prices falling by 9.67%. However, algorithmic pricing companies have developed “resetting” strategies (which regularly raise prices in the hope that competitors will follow) in order to avoid stark Bertrand-Nash competition. We find that these strategies are effective at coaxing competitors to raise their prices; when a resetting strategy is adopted on a market with less than six serious competitors, both competitor prices and market prices eventually increase by 11.4%. Although the resulting patterns of cycling prices are reminiscent of Maskin-Tirole’s Edgeworth cycles, a model of equilibrium in delegated strategies fits the data better. This model suggests that the average price over the cycle will be the monopoly price. Moreover, if the available repricing technologies remain fixed, cycling and prices could rise significantly. However, cycling is still relatively rare in the data, even when studying a convenience sample of products with at least one merchant using a repricing tool. This paper was accepted by Omar Besbes, revenue management and market analytics. Supplemental Material: The online appendix and data files are available at https://doi.org/10.1287/mnsc.2022.02462 .
We study a vote buying setup where a committee votes on a proposal important to the vote buyer. We characterize the cheapest combination of bribes that guarantees the proposal's passing in different voting environments. We find that for both simultaneous and sequential votes, the vote buyer publicly offers small bribes to a large super-majority of members. Each member accepts because he anticipates that the proposal will pass regardless of his vote. We discuss the committee design that maximizes cost of capture: demanding majority requirements combined with diversity among members make the committee more expensive. In small committees, sequential voting increases cost while the opposite is true for large committees. On the other hand, additional members and transparent voting rules lower the cost.
How do the productivity effects of AI evolve across successive generations of tools, and to what extent do task-level gains ultimately translate into final output? We study these questions in the context of software development, using data on more than 100,000 GitHub developers combined with their AI usage telemetry. In a matched event study design, we find that autocomplete, interactive coding agents, and autonomous coding agents each significantly increase coding activity ("commits"), with respective cumulative effects of 40%, 140%, and 180%. These gains, however, attenuate sharply across the production hierarchy: the 180% cumulative effect falls to 50% for the number of projects, and to 30% for actual releases. This pattern is consistent with the weak-link hypothesis: the strong productivity gains from AI are attenuated by human bottlenecks in the production chain, with an estimated elasticity of substitution of 0.25 between AI and human effort, which indicates strong complementarities. We further confirm these results across four major app marketplaces, finding a moderate increase in the number of new apps but no increase in total usage. Large task-level AI productivity gains have therefore translated only partially into shipped and used software thus far.