An introduction
What are AI Shopping Agents?
AI shopping agents are autonomous systems that carry out shopping tasks on a person's behalf: interpreting intent, searching across merchants, comparing options, and, increasingly, completing the transaction. Ask for a black wool coat under 800 dollars that ships this week, and the agent finds it, evaluates it, and can buy it. Agents differ from chatbots in one decisive way. Agents act. The output is not an answer. It is an order.
From answers to orders
The capability arrived in stages. Answer engines like ChatGPT, Perplexity, and Google's AI Mode first absorbed product research. Then came transaction: ChatGPT's Instant Checkout, built on the OpenAI and Stripe Agentic Commerce Protocol, Perplexity's shopping and buying features, and Google's AP2 payment authorization standard. The stack for agent led buying, discovery, decision, and payment, now exists end to end.
The projections that matter, stated with their sources. Bain estimates 15 to 25 percent of United States ecommerce, between 300 and 500 billion dollars, could be agent mediated by 2030. Morgan Stanley's range is 190 to 385 billion, a 10 to 20 percent share. McKinsey models up to one trillion dollars in agent orchestrated United States retail revenue, and three to five trillion globally, by 2030. Gartner projects 20 percent of digital commerce transactions executing through AI platforms by the same year.
For calibration: ecommerce itself took roughly twelve years to triple its share of United States retail. The agent curve is projected to go from roughly zero to 15 to 25 percent in five.
How agents choose
An agent's selection is a retrieval and reasoning act, not an auction. It weighs whether it can parse the product accurately, which depends on structured data. It weighs whether the brand appears in its underlying answer corpus, which is AI visibility. It weighs whether policies are legible: returns, shipping, sizing. And it weighs whether it can transact, which depends on protocol integration.
There is no paid placement yet. Which means the current window rewards exactly one thing: being the most machine legible, most cited brand in the category. That window is the opportunity.
Common questions
Are people actually letting agents buy for them?
Adoption is early and concentrated in repurchases, gifting, and constraint heavy searches. But every major platform shipped transaction capability within the last year, which is why the 2030 projections are steep.
Can a brand opt out?
Of agent checkout, yes. Of agent judgment, no. Agents will describe and compare the brand from whatever data exists, accurate or not.
What should a premium brand do first?
Audit how agents currently see it: what they say, what they recommend instead, and whether product data parses. Measurement precedes strategy.
Further reading