Agentic commerce

Agentic commerce is buying and selling where an AI agent takes the actions. A person describes what they want, and the agent searches catalogs, compares options, weighs price and availability, asks questions when the request is unclear, and places an order or declines.

What an agent has to get right

• The request. Requirements, preferences and limits such as a budget or a delivery date, including the ones stated only once.
• The options. Prices, stock and product details as they actually are, not as the agent assumes they might be.
• The decision. Choosing the best available option, or declining when nothing fits, and following any procedure the person asked for.

Each of these is a point where an agent can fail while still sounding confident. Because the result is a concrete purchase, it can be graded objectively, which is why commerce makes a rigorous agent benchmark.

At ORO

ORO builds reinforcement learning environments and benchmarks for agentic commerce. ORO Bench grades shopping agents on what they bought, and AssistantEval tests assistant apps and frontier models on everyday delegated tasks.