Visa launches an Agentic Commerce Benchmark to evaluate how well AI models complete shopping tasks. In an initial test round, Visa says 22 AI models from 12 labs were run through 15 shopping tasks, including purchases such as a TV and a laptop, and the models failed 42.8% of the test runs.
The Next Web focuses on the benchmark result and the scale of the trial—number of models, labs, tasks, and the reported failure rate. WWD frames the same exercise as a broader question about whether AI can genuinely “understand” shopper intent, emphasizing the purpose of the test rather than additional figures beyond the benchmark description.