Vibe researching: can AI agents with skills replace or augment social scientists?

Agentic artificial intelligence (AI) marks a qualitative shift from prior automation technologies in social science research. Unlike chatbots that respond to isolated queries, AI agents can read files, execute code, query databases, search the web, and invoke domain-specific skills to carry out entire research pipelines. This paper focuses on the concept of vibe researching and uses Open Scholar Skills, a plugin for coding agents that spans idea formulation to journal submission, as an illustrative case. I develop a cognitive task framework that classifies research activities along analytically independent dimensions, including output verifiability and process articulability, to identify a delegation boundary that is cognitive rather than sequential. AI agents excel at speed, coverage, and methodological scaffolding but struggle with theoretical originality and tacit field knowledge. A generation–verification asymmetry structures the resulting human–AI division of labor: AI compresses production time by orders of magnitude, while the expert judgment needed to validate outputs that resist mechanical verification remains largely fixed. The paper concludes by tracing implications for the profession and proposes principles for responsible vibe researching.