10 Agentic Commerce Research Papers Shaping the Future of Enterprise Product Discovery
Meta-analysis of 2025 agentic commerce research, including empirical findings on agent purchasing behavior, position bias, and the modular retrieval-first architectures that enable reliable shopping agents.
Agentic Commerce: Modular Architectures Over Monoliths
This curated collection of 2025 agentic commerce research reveals a critical architectural insight: the most reliable shopping agents are orchestrators, not monoliths. The flagship paper, ‘What Is Your AI Agent Buying?’ (Allouah et al., 2025), deployed agents in mock marketplaces where they shop via APIs. The findings were sobering: different LLMs make wildly different purchasing decisions, concentrate on small product sets, and exhibit strong position bias. Sponsored tags trigger inconsistent reactions, raising questions about competition fairness and ranking integrity.
What changed. Agentic commerce research has shifted from asking “Can agents shop?” to “How do agents fail?”—with empirical evidence that monolithic LLM-based agents are unreliable and biased.
Why it matters. As agentic systems move into high-stakes domains (commerce, finance, healthcare), understanding failure modes is non-negotiable. Position bias and inconsistent decision-making aren’t edge cases; they’re systematic vulnerabilities that undermine trust and fairness.
Builder takeaway. The emerging architecture for reliable agentic systems is modular and retrieval-first: retrieve candidate items, interpret user intent, validate against constraints, then rank using proven systems. This pattern—retrieve → interpret → validate → rank—mirrors successful GenAI architectures in production and should become the default for builders shipping agentic systems.