Why Walmart Isn’t Sleeping on Agents
Retailers have mixed feelings about web-browsing agents that can control a person’s browser to complete tasks for them.
Some firms like DoorDash are concerned that an explosion of agents could overrun their sites with bots, lowering the value of their ads and creating distance between themselves and customers in the process. (That was the subject of our feature story on Wednesday.)
Others like Walmart say it’s too early to tell.
Hari Vasudev, chief technology officer of the retailer’s U.S. business, believes agents will become a mainstay for consumers but doesn’t think they’ll be as useful for shopping as AI firms that develop them think.
For toothpaste and paper towels? Sure, automate away. But shopping for special events like birthday parties often involves specific preferences and details that may be hard for AI to work out, he said.
“It’ll be a while before agents take over,” he said.
Still, Vasudev expects web-browsing agents from OpenAI, Anthropic and others to drastically change the way companies like Walmart do search engine optimization and sell advertisements on their sites.
Advertising could become more focused on brand-building or targeting a consumer’s social circle, otherwise known as top of funnel, rather than visual ads or links intended at prompting a purchase, Vasudev said.
Retailers like Walmart could also build agents of their own to interact with Operator and other web-browsing AI, to provide recommendations or additional product information, he said.
Walmart would “have an agent that understands, interacts with the customer agent and decides how to personalize the experience,” he said. “What’s really going to happen is we have to establish a shared understanding of customers’ needs” for both of these agents.
On the flip side, Operator and other agents representing consumers could provide retailers’ agents with information about those consumers’ preferences, he said, like how cost-sensitive they are or specific brands they like.
A world of agents might also give rise to a new form of advertising, he said. A customer may want their agent to discover new products, and new types of ads would help those agents do so and “ensure that we’re informing the agent about everything that’s available” at Walmart, he said.
It might seem like retailers are getting ahead of themselves. Agents are far from mature, as we pointed out in our OpenAI-DoorDash story and our feature about Salesforce’s early struggles with Agentforce.
But retailers are keenly aware of the consistent improvement of large language and vision models over the past two years, and how fast-growing products that use them—namely, ChatGPT—are already influencing shopping and product recommendations.
That’s why Walmart isn’t sleeping on agents.
In related news…
AI Works… Longer
Walmart has a good reason not to sleep on agents, if the results of recent study pan out.
Model Evaluation and Threat Research, a nonprofit group whose work we wrote about in AI Agenda recently, just published a report showing that core technologies underpinning agents are progressing—fast.
In short, the length of a task AI can take on doubles every seven months, the research found. If the trend over the last six years continues for a few more years, the report said, “generalist autonomous agents will be capable of performing a wide range of week-long tasks.”This assumption rests on continued improvements in model performance, of course, so we are far from getting there. As the paper noted, “the best current models—such as Claude 3.7 Sonnet—are capable of some tasks that take even expert humans hours, but can only reliably complete tasks of up to a few minutes long.”—Amir Efrati
New From Our Reporters
A Podcaster Develops AI to Search His Own Episodes
