Amazon faces a dilemma — fight AI shopping agents, or join them

Amazon CEO Andy Jassy could see how dramatically artificial intelligence was altering e-commerce.

In June, he told employees that AI agents will start to infiltrate aspects of everyday life, “from shopping to travel to daily chores and tasks.”

Four months later, Jassy said on an earnings call that Amazon expects to partner with third-party agents, and has engaged in conversations with some providers, though he didn’t offer names.

Now, Amazon is looking to hire a leader in corporate development to help forge strategic partnerships in areas including “agentic commerce,” according to a recent job posting.

Amazon’s rapid evolution in its view of AI-powered commerce underscores how quickly online retail is changing, and the risks the company faces if it doesn’t act aggressively to maintain control over its future.

The company has watched as OpenAI, Google, Perplexity and Microsoft have released a flurry of e-commerce agents in recent months that aim to change how people shop. Instead of visiting Amazon, Walmart or Nike directly, consumers could rely on AI agents to do the hard work of scanning the web for the best deal or perfect product, then buy the item without exiting a chatbot window.

The first shopping agents from AI leaders were released about a year ago. Consulting firm McKinsey projected that agentic commerce could generate $1 trillion in U.S. retail revenue by 2030.

It’s a trend that poses a threat to Amazon’s margins and relationships with customers. When a consumer uses ChatGPT to initiate a purchase, for example, OpenAI collects “a small fee” from each transaction.

“With an agent on ChatGPT, retailers risk relinquishing transactions on their site to pay a toll on someone else’s highway for the same transaction,” Sucharita Kodali, a retail analyst at Forrester, said in an interview.

Some companies are trying to find a middle ground between working with agent providers and competing against them. Walmart, Shopify and others have adopted a frenemy strategy, announcing partnerships with AI companies while continuing to develop their own tools and setting guardrails around how agents can access their sites.

Shopify CEO Tobi Lutke wrote in a post on X on Tuesday that his company is “building all the layers of infrastructure to power a new cambrian explosion of creativity in shopping.”

“I’m really really excited about Agentic Commerce,” Lutke wrote. “There is so much amazing stuff being built. Everything I test just feels delightful and right.”

Amazon has so far been playing defense.

The company recently updated the code underpinning its website to block external AI agents from crawling it, part of an effort to wall off its valuable training data from rivals. As of Tuesday, Amazon had blocked 47 bots, including those from all the major AI companies, according to its website.

Amazon has even taken the matter to court. In November, Amazon sued Perplexity over an agent in the startup’s Comet browser that allows it to make purchases on a user’s behalf. The company alleged Perplexity took steps to “conceal” its agents so they could continue to scrape Amazon’s website without its approval.

Perplexity called the lawsuit a “bully tactic.”

Meanwhile, Amazon is investing heavily in its own AI products. The company released a shopping chatbot called Rufus last February, and has been testing an agent called Buy For Me, which can purchase products from other sites directly in Amazon’s e-commerce app.

Morgan Stanley expects that by 2030, nearly half of American shoppers will use AI agents and the technology could add up to $115 billion in U.S. e-commerce spending.

“We believe agentic commerce — in effect the ability to have a personal digital interactive shopper — is set to be the best next substantial GenAI-enabled unlock,” Morgan Stanley analysts wrote in a report in November.

They noted that a mid-single digit percentage of consumers currently start their “purchase journey” through AI, but that could increase over time as roughly 40% to 50% of Americans currently use AI for product research.

Traffic from AI chatbots to U.S. retail sites has surged in recent months, especially during the holiday season, but research suggests Google search still performs better in terms of conversion rate and revenue per session.

AI-powered shopping remains a nascent market.

OpenAI’s Instant Checkout tool, launched in ChatGPT in September, is only available for some products sold by Walmart, Shopify, Target and Etsy. Users can only purchase one item at a time, and they can’t connect loyalty memberships like Walmart+.

Agents are also prone to glitches.

Scot Wingo, founder of e-commerce software startup ReFiBuy, recently tested Perplexity’s Instant Buy tool that lets users purchase items directly in its search engine.

Wingo tried to buy a cable knit sweater from Abercrombie & Fitch, but Perplexity’s agent repeatedly spit out error messages, even though both products were in stock on the retailer’s website. He eventually gave up.

Earlier this month, Wingo was searching for a coffee machine on ChatGPT when it suggested a Breville espresso maker. When he clicked on the product, he was surprised to see an image of a garden rake.

“These crawlers go out, they pull in this data and you never know exactly what they’re gonna get,” Wingo said.

As Amazon contemplates its next move with shopping agents, it’s quietly allowing them to access some of its properties.

Subsidiaries like shoe seller Zappos, fashion site Shopbop and deals site Woot don’t appear to have any language blocking agents in their robots.txt files, which dictate how crawlers can access specific webpages.

“A lot of times they’ll use the subsidiaries to experiment,” Wingo said. “Zappos has its own experience and database, so it’s not like they’re letting all the horses out of the barn.”

The company could ultimately take a page from its rivals if it decides to let agents access its primary e-commerce platform. Shopify and Walmart have set boundaries around what external shopping agents can and can’t do on their sites.

Amazon may be willing to let agents access its catalog, but it likely wants to protect more valuable data from its competitors, Wingo said, such as its vast trove of customer reviews and sales rankings, both of which indicate a product’s quality and can help improve an AI chatbot’s answers.

“Those are probably the two most proprietary data points that if I’m Amazon, I want to protect,” Wingo said.

Amazon isn’t giving up on its homegrown tools.

Rufus’ capabilities have improved since Amazon first launched it last year, and the company has been surfacing the chatbot across more areas of its site to drive user adoption.

Amazon recently added a feature where Rufus can auto-buy items on a Prime shopper’s behalf once they hit a certain price. The chatbot now also suggests products from sites across the web, not just on Amazon.

Amazon also began testing a feature in recent weeks that allows Rufus to create custom shopping guides, similar to OpenAI’s “shopping research” tool launched last month.

“Instead of the innovator’s dilemma, I would say Amazon is in what I would call the leader’s dilemma,” said Jordan Berke, founder and CEO of retail consulting firm Tomorrow. “Their market share is so significant that they have the most to lose.”

WATCH: How Amazon came to dominate the U.S. apparel market