Commerce agents are starting to show revenue lift. Now show us the experiment.

Anthropic says one partner using its commerce-agent technology saw cart size rise about 30–35% and customers become about 60% more likely to complete a purchase. The result is promising, but the public evidence does not disclose the partner, sample size, baseline conversion rate, study duration, or statistical methodology.

The signal

Anthropic introduced new blueprints for shopper and merchant agents on September 2. In reporting on the launch, Anthropic head of product Angela Jiang said one partner saw cart sizes rise by about 30–35%, while customers were about 60% more likely to complete a purchase.

Why this belongs in the machine-economy dataset

Not every economically important agent needs to move money itself. An agent can create economic value by changing what customers discover, compare, add to a cart, and ultimately buy. That makes conversion and order value valid machine-economy measurements even when settlement remains human-controlled.

What is actually autonomous?

Anthropic says the shopper blueprint can make preference-based suggestions and add items to shopping carts. The merchant blueprint can make recommendations involving inventory, prices, and marketing. Anthropic also says these agents do not complete purchases on behalf of shoppers or merchants.

Why this is a brief—not a centerpiece

The headline numbers are encouraging, but the public evidence is incomplete. Anthropic has not publicly identified the partner in the material we reviewed, and we do not have the experiment size, baseline conversion rate, study duration, control methodology, gross-margin impact, return rate, or statistical confidence.

What this does not prove

A larger cart is not automatically more profitable. Higher purchase completion can be offset by discounts, returns, lower-margin product mix, customer-support costs, or cannibalization from other channels. The reported figures are also company-supplied rather than independently audited.

What we watch next

The useful measurements are baseline versus agent-assisted conversion, average order value, contribution margin, returns and cancellations, repeat purchase, customer-acquisition cost, human intervention, and deployment cost. If those numbers emerge, this category could move from interesting commerce signal to measurable agent-created value.