It used to be that people shopped 'til they dropped. But tireless bots never have to stop -- for better or worse.

Artificial intelligence has been working behind the scenes in retail for years, far beyond ubiquitous product recommendations, as companies use powerful models to predict inventory needs, optimize supply chains, and automate warehouses. But AI has exploded in popularity with consumers following the recent introduction of Muse and Dots, the autonomous AI agents from Facebook parent Meta Platforms and ChatGPT developer OpenAI, respectively.

Despite their apparent ability to help people save money and discover new products, at this point there are more questions than answers about these tools -- for both buyers and sellers.

AI agents are undoubtedly useful, but they don't have a legal obligation to find the cheapest items -- and they tend to gravitate toward big brands. Preferences are also subjective, so what works for one consumer may not for another. And users will want to check the bots' work.

AI agents might be the next step up in e-commerce, but they come with an asterisk.

The numbers, like it or not, are too big to ignore. Muse surpassed five million downloads in just over three weeks after its introduction in September, claiming the top spot in the Apple App Store and Alphabet's Google Play. Bain & Company estimates that agentic AI agents will account for as much as $500 billion in as little as four years, claiming as much as a quarter of all U.S. e-commerce by 2030.

Although the landscape will change, there are already winners and losers. As usual, the largest retailers, like Walmart and Target, will probably reap the biggest benefits as more dollars flow through bot purchases. (So much for AI leveling the playing field for the little guys.)

Consumers, meanwhile, remain at the mercy of the tech companies deploying these agents, with no way to know if they are giving them the "best" option, or even how to define that.

"What does it mean to evaluate whether a shopping agent works or not?" asks Omar Besbes, Vikram S. Pandit Professor of Business at Columbia Business School. "If we ask whether [AI] works in coding or mathematics, we know because there is a right answer. But what does it mean for it to pick the best product?"

His research shows different agents return very different recommendations for the same query, even for simple products like toothpaste. If a bot picks an item that comes in under budget and fits a person's criteria, there's still no way to know whether there were other options that might have been better.

Users aren't privy to the bots' selection rational. Also, although major payment companies including Visa and Mastercard permit bots to shop for their customers, there's no uniform legal standard regarding who would be responsible for, say, fraud.

Still, it's clear why the introduction of Muse and Dots made such a splash. "This could be an iPod moment for AI, when a breakthrough product makes a developing technology available to the mass audience," says David Russell, TradeStation's global head of market strategy.

So far, the AI story has been mostly an enterprise one -- massive companies spending billions on its buildout and others adopting it. However, given that Americans' spending accounts for some two-thirds of gross domestic product, "the consumer is one of the holy grail opportunities," says 22V Research data infrastructure/commodities analyst Dauvin Peterson.

Yet, like the quest for the holy grail, the use of AI shopping agents is a fraught one for both consumers and retailers.

The bots do have positive uses: quickly comparing prices or whipping up travel itineraries. Muse has teamed up with Stripe and Shopify to provide things like free coverage for damaged items and no-fee returns, and pledges to safeguard financial data. Dots has its own big partners, including Target, and touts safety features like encryption.

But here's the thing. Users can't know how or why a bot chooses one product over another, the "best" option can be subjective. Most worryingly, there are minimal guardrails to protect consumers or to ensure that they -- not retailers or tech companies -- are the bots' priority.

"These agents don't work for you," says Lindsay Owens, president and CEO of nonprofit watchdog Groundwork Collaborative. "We could be unleashing the most sophisticated snake oil salesman the world has ever seen."

Yup, in some ways, it's the Wild West in Silicon Valley.

AI agents have an incentive to do right by users -- but only to the extent that shoppers keep using them. If they pick the lowest-priced or highest-rated item most of the time, people will probably stick with them, but that doesn't have to be near 100%. Consumers who see savings early on may skip verifying the agent's future recommendations since that negates the timesaving benefit.

What do the bot parents say? Not much so far. A Meta spokesperson pointed to publicly available statements about Muse's safeguards and partnerships to secure transactions. OpenAI didn't respond.

But the success or failure of bot shoppers is going to come down to this: "Trust is going to be the biggest factor in how quickly consumers adopt this new technology," says Badgley Phelps research analyst Katie Wham.

And that is questionable right now.

Trust comes not just from consistency but transparency, which is particularly relevant given how quickly algorithms and inputs can shift. It also means that first-mover advantage may not determine which companies dominate.

Amazon.com banned Muse, as it has its own AI ambitions. Other retailers want a piece of the pie, too, and Walmart has already rolled out multiple tools to help with agentic commerce. Given the skepticism many Americans feel toward Big Tech, there's a chance for more trusted names in retail and consumer finance to make inroads if they can overcome that hurdle.

A September UBS survey found that only 17% of respondents said they thought agentic AI was a good idea, down a percentage point from the prior quarter, while 43% said it wasn't a good idea -- up a percentage point. Just over 10% said they would use agentic AI to book a restaurant, down from about 12% in June.

"Perhaps six or 12 months ago, we would have expected steady growth in the percentage of people becoming aware of and using AI, the September data indicates that is not exactly playing out, at least for now," writes UBS analyst Jay Sole.

Still, there are winners.

Companies like Walmart, Target, and Costco Wholesale typically take the top spots when it comes to agent recommendations, along with well-known category experts such as Home Depot or Best Buy, as bots read them as reliable and well stocked.

"The incumbents are the firms we're used to working with....[W]e know what we're going to get" and AI agents therefore view these as the safest route, says Matt Stith, Bartlett Wealth Management portfolio manager and director of equity research.

Bots are motivated to ensure companies can swiftly fulfill orders with products that won't lead to difficult returns, which can undermine the ease of use that hooks users. These are the same companies with the resources to "advertise" their availability to bots -- and develop their own AI agents.

UBS analyst Michael Lasser's research shows that AI agents clearly favor traditional retailers over digitally native ones. Still, they can't rest on their laurels. They now have to spend money not only to compete for shelf space, app downloads, and search rankings, but also for bot recommendations: "Retailers that successfully provide product catalogs, inventory feeds, merchant data, store information, reviews, and other structured content to AI agents may gain a significant advantage," Lasser says.

Likewise, third-party marketplaces -- from Amazon to Etsy and Walmart -- may have to rework the lucrative process of charging merchants for placement. Columbia's Besbes found that bots tend to avoid products labeled as "sponsored" in favor of those deemed the "overall pick" -- even in simulations where those labels didn't correspond to real-world rankings, and thus shouldn't have factored into AI agents' thinking. Additionally, while some bots preferred products that were placed in the top-left corner (as many humans do), others chose those on the bottom right, again muddying the value of paying for positioning.

Looking ahead, Gavekal Research's Louis-Vincent Gave wonders if bots will lower demand for things like luxury items or specific car brands, where emotions play a big role in purchases, or if they will sabotage marketing in general. "In a world where AI agents select the best deals...the rationale for online advertising begins to disappear," he says.

Adds Guggenheim analyst Simeon Siegel, "Anyone writing a list of winners now better be using a pencil."