Walmart just showed OpenAI how to actually sell groceries, and the lesson is brutal: consumers don't want AI agents handling checkout, they want shopping advice that doesn't suck.

The Summary

  • OpenAI's Instant Checkout feature for Walmart flopped, so Walmart pivoted to embedding its own Sparky chatbot into ChatGPT and Google Gemini instead
  • The shift signals that agentic commerce needs trust and context before autonomy, consumers aren't ready to hand over their wallets to third-party AI
  • Walmart is betting on advisory agents over autonomous checkout, a move that could redefine how retailers approach AI integration

The Signal

The original pitch was clean: OpenAI builds checkout functionality into ChatGPT, you tell it what you need, it completes the purchase through Walmart. Frictionless commerce powered by the world's most popular AI interface. Except nobody used it. Walmart is now reversing course, pulling back from OpenAI's checkout dreams and instead putting its own Sparky assistant inside ChatGPT and Gemini as an advisory layer.

The failure wasn't technical. It was psychological. Consumers trust Walmart's AI to help them find pasta sauce. They do not trust a third-party LLM to execute a $200 grocery order without oversight. The gap between "help me decide" and "do it for me" is wider than the AI optimists thought. Instant Checkout assumed people wanted to skip steps. What they actually want is better information at each step, not fewer steps controlled by someone else's algorithm.

Sparky's new role is narrower and smarter. It lives inside the chat interfaces people already use, answers product questions, suggests alternatives, helps with meal planning. But the actual transaction still happens in Walmart's app or site, where the customer maintains control. This isn't a retreat from AI commerce, it's a recognition that agent adoption follows a trust curve. You start with advice, earn credibility, then maybe, eventually, get permission to act.

The bigger tell: Walmart chose to build across both ChatGPT and Gemini. That's a hedge against platform risk, but it's also a statement about where the agent economy is headed. No single LLM will own consumer commerce. The winners will be the retailers who can make their own agents portable, plugging into whatever interface the customer prefers. Walmart isn't betting on OpenAI or Google. It's betting on itself.

The Implication

If you're building in the agent space, watch what Walmart just did. They tested autonomous transactions, saw the friction, and downshifted to advisory mode. That's the actual adoption curve for most agent use cases. People will let AI advise before they let it act. Build trust at the recommendation layer before you ask for wallet access.

For OpenAI, this is a warning shot. Being the best LLM doesn't guarantee you win commerce. Retailers with existing customer relationships and supply chains will build their own agents and use your model as infrastructure. The platform play only works if customers trust you more than they trust the brand. Right now, they don't.

Source: Wired AI