Agentic Commerce World: An Auditable and Verifiable Environment for Vibe Commerce
2026-08-03 • Artificial Intelligence
Artificial Intelligence
AI summaryⓘ
The authors created ACWorld, a virtual setup where buyer and seller AI agents can negotiate and complete transactions while keeping their own goals and rules separate. They designed a protocol to check and record every step agents take, making it easier to test and review their behavior. Using this setup, they tested various models on tasks ranging from simple trades to searching through large product lists, finding that looking at the whole process gives better insight than just the final result. Their work helps understand how AI agents can work together in buying and selling scenarios.
vibe codingAI agentsbuyer agentmerchant agenttransaction stateagent evaluationbenchmarkprotocollarge catalog searchprocess-level evidence
Authors
Shicheng Fan, Mingdai Yang, Duohao Wang, Canyu Chen, Yongfeng Zhang, Hua Wei, Manling Li, Julian McAuley, Kun Zhang, Philip S. Yu, Kejing Yu, Zhiwei Liu
Abstract
In vibe coding, people describe software in natural language and delegate implementation to AI agents. By analogy, vibe commerce allows people to express buying or selling goals in natural language and delegate the corresponding tasks to agents. Commerce, however, requires independently controlled Buyer and Merchant agents to interact in a shared market while preserving their private objectives and distinct authority. We introduce Agentic Commerce World (ACWorld), an environment for evaluating such agents across ongoing transactions. Through its Vibe Commerce Protocol (VCP), ACWorld validates agent actions before updating shared transaction state and records the resulting interactions, making agent behavior auditable and evaluation reproducible. The ACWorld Benchmark contains a 200-task capability-coverage track and a 60-task large-catalog track that searches 785,022 transactable listings. Across ten models, mean scores range from 65.9% to 85.6% and from 56.1% to 91.4%, respectively. Our analysis shows that process-level evidence is necessary: final state alone can miss evaluated errors, incomplete trajectories still retain useful process signals, and large-catalog tasks expose bottlenecks across stages.