Shopping by AI agent is splitting into rival camps, so stay flexible

Online stores are being urged to get in early on shopping by AI agent, but two payments experts say betting on one platform is risky. OpenAI, Google and Anthropic have each built their own checkout inside their chat tools. The market is already splitting, much as it did with mobile phones, apps and digital wallets.

Why early bets can backfire

In past waves of technology, merchants raced to pick the platforms that would win. No single platform ended up owning any of those areas. Many early adopters later had to change course. Adam Hiatt of Spreedly and Don Apgar of Javelin Strategy & Research discussed this on a PaymentsJournal podcast. They expect agentic commerce, where an AI agent shops for you, to follow a similar split path.

Other firms, such as Muse and RockBot, offer personal AI agents that take over shopping tasks. The leading card networks have also set their own standards for how payments connect to these tools.

The hidden cost of every connection

Apgar warned that each extra connection a store builds creates costs across the business, from finance to fulfillment to customer service, chargebacks and returns. His answer is an orchestrator, a single connection that links to many systems without passing all that work downstream.

If a store backs one format and it loses, it could face big costs and disruption while rebuilding. Hiatt also pointed to card tokenization, which swaps a real card number for a secure stand-in code. The card networks are building agent guardrails on it, so agents do not run wild with a credit card.

Where the new sales may come from

Many stores may first build chat helpers on their own sites. Hiatt said these rarely create new revenue. Many agent purchases are not new sales. They come from existing customers who handed their buying to an agent.

He suggested focusing on new channels beyond the store's own website, where new customers could be reached.

What to do

• Build payment and tech setups that work with any AI agent, not just one.
• Make products easy for agents to find by distributing your catalog.
• Add more written detail to listings, since agents may not read images at all.

A person can pick up details from a few photos and a short description. An agent may not be able to, so thin listings could leave a product overlooked.