This paper investigates whether fully agentic AI can nowcast stock returns using a state-of-the-art Large Language Model (LLM) that autonomously searches the web, filters sources, and synthesizes information into quantitative predictions. The study focuses on Russell 1000 stocks daily starting from April 2025, when AI web interfaces enabled real-time search. The key innovation is a completely out-of-sample and look-ahead-bias-free framework: predictions are collected at the current edge of time, ensuring the AI has no knowledge of future outcomes. This temporal design is irreproducible because once the information environment passes, it cannot be recreated. The framework is 100% agentic, meaning no curated news or disclosures are fed to the model; it operates entirely on its own web searches.

The results show that AI possesses genuine stock selection ability, but only for identifying top winners. A strategy longing the 20 highest-ranked stocks generates a daily Fama-French five-factor plus momentum alpha of 18.4 basis points and an annualized Sharpe ratio of 2.43. These returns derive from an implementable strategy trading highly liquid Russell 1000 constituents, with transaction costs representing less than 10% of gross alpha. However, this predictability is highly concentrated: expanding beyond the top tier rapidly dilutes alpha, and bottom-ranked stocks exhibit returns statistically indistinguishable from the market. The authors hypothesize that this asymmetry reflects online information structure: genuinely positive news generates coherent signals, while negative news is contaminated by strategic corporate obfuscation and social media noise. The study provides strong evidence that agentic AI can generate economically significant and implementable alpha, but its predictive power is asymmetric and concentrated in top performers.