Firms often choose with whom to compete and how similar or how different their products should be relative to those of their rivals. This paper investigates this issue in the app economy by studying the determinants of mobile app success. We leverage natural language processing and unsupervised machine learning to cluster apps using pairwise cosine similarity, which provides a measure of horizontal differentiation within app categories. We find asymmetric effects of rivalry from first-party apps: Apple’s entry into a cluster stimulates sales of third-party apps, whereas Google’s entry decreases them. We also find that apps in a cluster that more closely resemble the most popular apps have fewer downloads, whereas greater similarity to average competitors enhances downloads. We further find that more frequent updates increase app downloads. These findings yield important implications for developers in choosing market segments and designing effective differentiation strategies.

This paper develops a descriptive structural framework to decompose the economic incentives underlying foundation model (FM) owners' openness decisions. The framework consists of two stages where FM owners choose the degree of openness of their models, and downstream developers decide which models to adopt for application development. Using data from Hugging Face and model documentation, we find that the current-payoff component accounts for roughly half of the normalized incentive for openness at the median, indicating that immediate returns alone do not fully explain owners' openness choices. Forward-looking considerations account for the remaining incentive. In particular, greater openness encourages downstream development activity, which can generate knowledge spillovers that improve the quality of subsequent model generations. These quality improvements, in turn, increase future adoption and the continuation value of openness. This intertemporal mechanism helps explain why FM owners may choose relatively open strategies even when the immediate payoff from doing so is limited.

We study whether firms update mandatory quarterly disclosures after agentic AI becomes publicly salient and whether those updates are concentrated among firms with greater pre-existing AI intensity. We measure baseline AI intensity from the most recent Form 10-K filed before Anthropic's October 22, 2024 computer-use announcement and measure subsequent AI-agent intensity in Form 10-Q MD&A. In a continuous difference-in-differences design with firm and filing-quarter fixed effects, a one-standard-deviation increase in pre-shock 10-K MD&A AI intensity is associated with a 0.72 percentage-point increase in the probability of an explicit AI-agent mention and with higher disclosure intensity after the announcement. Risk Factors intensity also predicts the extensive margin and log number of mentions, although its association with matches per 1,000 words is weaker. The estimates are similar when industry-by-filing-quarter fixed effects absorb industry-specific shocks. While these findings may not establish internal AI-agent adoption, they document selective disclosure updating among firms with greater prior AI intensity.

This paper examines the welfare implications of the platform's vertical integration into downstream markets, focusing on the platform’s incentive to foreclose third-party sellers by raising commission fees. I present a three-stage theoretical model of the mobile app marketplace to analyze the platform’s dual role as both a gatekeeper and competitor. The model shows that the platform entry puts competitive pressure on third-party developers, reducing their profits. The competitive effects of entry intensify with the quality level of platform-owned apps. This weakens the platform’s commission revenue stream, creating an incentive to raise fees that eventually foreclose the marginal developers who have low-quality apps. Such foreclosing behavior of the platform results in restricting market diversity and harming consumer welfare. However, when consumer heterogeneity is high, the platform has less incentive to foreclose developers, mitigating the negative effects on consumers. In addition to theoretical findings, I conduct an empirical analysis using data from the Google Play Store.