(Editor’s Note: Agents of Transformation is an independent GeekWire series and 2026 event, underwritten by Accenture, exploring the people, companies, and ideas behind the rise of AI agents.)
Imagine telling your AI assistant that you need a new winter jacket. It already knows your style preferences and budget from previous purchases. The AI searches across dozens of retailers, analyzes reviews, checks for sales, and comes back with a list of ranked options.
You pick one you like. The AI asks if you want to wait until the price drops. A week later, there’s a sale. The AI completes the purchase, applies loyalty points, selects the fastest free shipping option, and sends you a confirmation. Your jacket shows up within days.
This is the promise of “agentic commerce” — AI systems that research, compare, and even buy on your behalf. Tech giants, startups, and retailers are all racing to build it. McKinsey projects the market could reach $1 trillion in the U.S. alone by 2030.
For the latest installment in our Agents of Transformation series, we interviewed startup founders, consumer brand marketing leaders, industry analysts, and others to better understand how agentic commerce could change the way we shop — now, in the future, or maybe not much at all.
Some key takeaways from our reporting:
- Agentic commerce could happen within a retailer’s “owned environments,” such as a website or app. Or it could be in a third-party platform, such as ChatGPT or Gemini.
- There is a lot of hype around agentic commerce, but today’s tools look more like fancy search than truly autonomous shopping.
- New behind-the-scenes technology infrastructure is emerging to let AI agents talk to retail sites, payments services, and login systems.
- Amazon sits at the center of the shift, simultaneously defending its ad-driven marketplace from outside agents and testing its own AI features.
- Brands are rethinking everything from how their sites show up in search to how their homepages are laid out.
- There are new security concerns as agents roam the open web and can be tricked by bad actors. Nearly 80% of financial institution leaders surveyed by Accenture expect that fraud will increase due to agentic commerce.
Major players are making moves.
- OpenAI just released a shopping research experience and announced a partnership with Walmart to let customers complete purchases within ChatGPT.
- Google rolled out agentic checkout options last month.
- Perplexity partnered with PayPal just before Black Friday.
Adobe reported that AI-driven traffic to U.S. retail sites jumped 670% year-over-year on Cyber Monday.
But it’s still early days. For ChatGPT, referrals to e-commerce apps represented only 0.82% of all sessions over Thanksgiving weekend. In a recent OpenAI study of about 1.1 million ChatGPT conversations, 2.1% of activity was classified as “Purchasable Products.”
There’s still a big gap between the pitch and what these tools can actually do today. Practical use cases remain limited.
“I am shocked at the promises versus reality,” said Emily Pfeiffer, a principal analyst and digital business expert with Forrester.
Still, the builders we spoke with see the current moment as the beginning of a fundamental shift.
“I think this is much bigger than even the invention of the online store,” said Jonathan Arena, co-founder of e-commerce AI startup New Generation.
Bots meet the buy button
McKinsey outlines three main ways agentic commerce could work:
- Agent-to-site (an AI assistant interacting directly with a retailer’s site)
- Agent-to-agent (a shopper’s agent working with a seller’s agent to complete a purchase)
- Brokered agent-to-site (an intermediary platform routing requests between agents and retailer sites)
Today’s reality is closer to a fancy search than full autonomy. AI chatbots can suggest products, but completing a purchase still typically requires clicking through to a retailer’s site. A handful of retailers have experimented with checkout-in-chat, but Pfeiffer said some polished demos don’t actually work in the real world.
“The experiences that are out there today, in my opinion, are extremely premature,” she said.
There’s also a broader debate about whether AI shopping assistants are solving a problem that doesn’t exist for specific purchase categories. For fashion, gifts, home decor — things where discovery is part of the value — many consumers may not want an agent to shortcut that process.
Agentic commerce could work best for low-consideration, commodity purchases — like household staples and replenishment items.
The concept becomes more complex outside of a brand’s own site or app, in AI search tools where an agent might eventually handle the entire shopping process without a user ever opening a retailer’s website. Pfeiffer believes this is where truly autonomous commerce is most likely to show up, though probably in specific situations rather than as a full replacement for browsing.
But she said any substantial shifts will take time. “If we get there, it’s not soon,” Pfeiffer said.
Teaching the internet to talk to AI
Agentic commerce isn’t possible without the right infrastructure. E-commerce websites were designed for humans typing keywords into a browser — not AI agents that need to read pages and place orders on their own.
New tools are starting to fill that gap.
- Anthropic has released the Model Context Protocol (MCP), which standardizes how AI agents share context across tools and platforms.
- Google launched the Agent Payments Protocol (AP2) in September, providing a framework for agents to make verifiable purchases.
- OpenAI, working with Stripe, has developed the Agentic Commerce Protocol (ACP) for completing transactions within ChatGPT.
For retailers, this patchwork can be confusing and expensive, especially as there’s no guarantee which protocol will become dominant.
“Each protocol is a burden for the merchant,” said Kumar Senthil, founder of Firmly, a Seattle-area startup building software that hides some of this complexity. His company, which recently partnered with Perplexity, lets merchants connect to multiple protocols through a single interface.
Firmly is trying to solve a basic problem: merchants can’t afford to integrate with every AI platform, but they also don’t want to miss out on any of them.
Senthil, who previously built Samsung’s e-commerce platform, said online retailers need to have “microstores” everywhere. Their traditional websites, he predicts, will go dark.
“The stores are going to be distributed across the internet,” he said.
But AI assistants need to draw on data from somewhere — which means a brand’s homepage could still serve an important purpose, even if the act of purchasing gets dispersed.
Brands like Brooks Running are refocusing their sites to make them easy for AI systems to read and understand. “We’re continuing to emphasize crawling, indexing, and ranking technical SEO opportunities through the lens of AI,” said Ryan Ngo, vice president of North America marketing and e-commerce at the Seattle-based company.
Beyond making a website “AI-ready,” Arena said brands should let shoppers ask questions about their products in plain language, using built-in AI chat on their own sites. “People are going to be frustrated that your website can’t answer them,” he said.
In Pfeiffer’s view, the bigger strategic risk lies in places brands don’t control — AI-powered search tools like ChatGPT or Gemini that could become powerful new gateways for finding and buying products. In that world, brands face the same decisions they once confronted with Amazon: what to share in each place people might shop, what to keep exclusive, and how to protect pricing and sensitive data.
What happens to Amazon?
Amazon helped shape modern online shopping when the Seattle-based giant started selling books on the internet more than three decades ago. The company is now a giant in online retail, and it’s staring at another potential shift with the rise of agentic commerce.
Amazon is in a tricky spot. The company captures roughly 40% of U.S. e-commerce spending and has a fast-growing advertising business that brings in around $70 billion a year — revenue that depends on humans browsing and clicking.
In November, Amazon sued Perplexity to stop the startup from using its AI browser agent to make purchases on its marketplace, citing computer fraud laws and security risks, along with a “significantly degraded shopping and customer service experience it provides.” Amazon has maintained what Bloomberg described as “a walled garden” that doesn’t allow autonomous shopping on its site.
Perplexity CEO Aravind Srinivas called the lawsuit “a bully tactic” and argued consumers should be free to use whatever AI assistant they prefer.
“Amazon should love this. Easier shopping means more transactions and happier customers,” Srinivas wrote. “But Amazon doesn’t care. They’re more interested in serving you ads, sponsored results, and influencing your purchasing decisions with upsells and confusing offers.”
Amazon CEO Andy Jassy acknowledged on a recent earnings call that agentic commerce “has a chance to be really good for e-commerce” and said that he expects the company to partner with third-party agents over time. But he also said agents “aren’t very good” at personalization and often display incorrect pricing and delivery estimates.
“So we’ve got to find a way to make the customer experience better and have the right exchange value,” Jassy said.
Amazon’s AI shopping assistant, Rufus, now has more than 250 million active customers. Amazon says that customers using the assistant during a shopping trip are 60% more likely to complete a purchase.
The company has also been testing a “Buy For Me” feature that lets customers purchase products from other brands’ sites, from inside Amazon’s mobile shopping app.
Senthil, the Firmly CEO, sees Amazon as potentially vulnerable. He questioned whether Amazon’s delivery speed advantage — long considered a competitive moat — will matter as much in a world where consumers place less emphasis on faster shipping times.
The rise of third-party AI agents, such as Perplexity’s Comet browser, could also weaken Amazon’s grip on customers. E-commerce journalist Jason Del Rey noted that if agents own the relationship and steer shoppers across sites, Amazon risks looking more like fulfillment infrastructure. That raises a long-term question, he said — if agents sit between shoppers and stores, who ends up capturing most of the value?
But others don’t expect AI tools to displace Amazon for now.
“It is highly unlikely that ChatGPT will be a dominant shopping cart mainly because e-commerce isn’t a problem that needs fixed,” said Sucharita Kodali, a retail industry analyst with Forrester. “It’s perfectly easy to buy on Amazon as hundreds of millions of people around the world already do every year.”
Kodali added: “It’s unclear what value ChatGPT is bringing to retailers, other than dis-intermediating Google.”
Last month Google unveiled a suite of AI shopping features powered by Gemini, including “agentic checkout,” which lets users set rules such as maximum spend or product specifications. It’s also building the infrastructure layer with AP2.
Microsoft, meanwhile, is positioning itself to help retailers and brands adapt to agentic commerce, whether building assistants into their websites or surfacing their offerings in third-party chatbots.
“We prioritize robust frameworks, open standards, and trust infrastructure so intelligent agents can operate reliably and securely throughout the commerce ecosystem,” said Kathleen Mitford, corporate vice president of global industry at Microsoft, responding to questions via email.
When AI knows you’re going on vacation
Finding the perfect winter coat based on your personal preferences may be just the start when it comes to AI assistants knowing what to purchase for you.
“Imagine an agent recognizing that the bathing suit you’re buying isn’t just another item, but part of preparing for an upcoming vacation and tailoring recommendations accordingly,” Mitford said.
That example would require consumers to offer up more personal data such as calendars and budget information. But it could enable a better experience, according to Arena.
“We’re talking about a brand being able to personalize experiences to all of their customers across the internet — not only on a first-party website that they own,” he said.
John Larson, who helped launch business messaging company Zipwhip (acquired by Twilio), said conversational commerce is evolving toward two-way interactions, enabling retailers to have more effective interactions with customers.
“We do believe that real conversational commerce leveraging agentic AI is absolutely the future,” said Larson, now an investor in Seattle startup Ambassador. “You’re getting your needs met, and you’re having a conversation.”
Lorrin Pascoe, CMO at Vancouver, B.C.-based footwear retailer Vessi, said he believes AI agents will become an important way to reach customers. “For us, it’s really realizing that this isn’t a gimmick,” he said. “It is something that is foundational in changing behaviors.”
Vessi began in 2018 as an online-only footwear company. This month, it’s opening its first U.S. store in Bellevue, Wash. — reversing the course that brick-and-mortar retailers took when e-commerce pushed them online. It’s a reminder that retail rarely follows a predictable path, and in the same way, there’s no telling where agentic commerce will ultimately land.
