The Agent Is the New Storefront
AI agents are becoming the layer between shoppers and stores. Here's what that does to ecommerce, and how brands and startups can win.
The Agent Is the New Storefront
Oct 3, 2026 · @David Mandel
AI agents are becoming the layer between shoppers and stores. Here's what that does to ecommerce, and how brands and startups can win.
On September 8, Meta launched Muse, a personal AI agent that can answer your email, book your dinner and buy things for you. Within a week it was the top free app in Apple's U.S. App Store. About two weeks later, Amazon shut it out. Shoppers who pointed Muse at Amazon.com got a message saying that "continued access by an unauthorized AI agent violates Amazon's Conditions of Use."
Amazon's official reasons were security and transparency. Muse didn't identify itself, and Amazon says it could capture customer credentials, which Meta denies. But the number I keep coming back to is $68 billion. That's roughly what Amazon earned from advertising last year, and nearly all of it depends on people scrolling Amazon's pages. An agent that never sees a sponsored listing is not a neutral visitor.
I think that standoff is the clearest sign yet of something I've been watching all year. The storefront is moving out of the browser and into the agent.
The funnel collapses into a shortlist
For 30 years, online shopping has worked the same way. You search, browse, open a dozen tabs, read reviews and click buy. Brands fight for your attention at every step of that funnel. A whole industry, from SEO and paid search to retail media and affiliate marketing, grew up to sell them that attention.
Agentic commerce collapses the funnel. You tell an agent what you want: "running shoes for flat feet, under $150, here by Friday." It does the searching and comparing, then comes back with two or three picks. You choose one, or let it choose, and it handles checkout, payment and tracking. You never visit the site.
That changes who the customer is. A brand's first customer is now software, an agent deciding which handful of products a human gets to see. If you're not on that shortlist, you don't exist for that purchase. Goldman Sachs summed it up in a 62-page report last month: value follows consumer intent. The strategic question shifts from who owns the storefront to who owns discovery, intent and the transaction.
This is already happening
Consumers aren't waiting for the industry to sort this out. Adobe found that AI-referred traffic to U.S. retail sites rose 693% year over year during the 2025 holidays. Those visitors converted 31% better than other traffic and were a third less likely to bounce. Goldman cites Bain data showing about 44% of online buyers already use AI tools to discover products.
Retailers are moving too. Walmart's agent, Sparky, doubled its weekly active customers year over year, and shoppers who use it spend about 40% more per order. OpenAI launched Instant Checkout inside ChatGPT in September 2025, with Walmart, Etsy and Shopify merchants on board. By March 2026 it had pulled back, admitting the first version "did not offer the level of flexibility that we aspire to provide." Retailers now run their own checkout inside the chat.
The courts are clearing the road as well. In August, the Ninth Circuit threw out Amazon's injunction against Perplexity's shopping agent. It treated the agent as a tool the user controls, not a hacker. So future fights will likely be about terms of service, not computer-fraud law. That's exactly the ground Amazon is now fighting Meta on.
One lesson from all this: discovery is moving much faster than checkout. Most agent use today is research. Fully hands-off buying is still early.
What breaks when the shopper stops browsing
Four things that ecommerce has taken for granted start to give way.
Search visibility.
An agent shows three options, not ten blue links. Mondelez found Oreos showed up in only 10% of AI chatbot cookie recommendations. Being on page one used to be enough. Now only the top three count.
The ad model.
Global retail media hit roughly $170 billion in 2025, built on human eyeballs. Ads won't disappear. They'll follow intent into new formats like sponsored recommendations inside AI answers. But the auction changes when the buyer is a machine optimizing for fit and price.
The customer relationship.
When an outside agent is the front door, the retailer loses the browsing data, the cross-sell and the loyalty touchpoints. It may also pay a new toll: Goldman warns that agent platforms could charge referral fees or placement payments.
Trust and liability.
Who pays when the agent buys the wrong thing? Today the issuing bank eats in-store fraud and the merchant eats online fraud. For agentic transactions, Goldman's answer is literally "TBD." Fraud tools break too. A legitimate agent running from cloud servers looks a lot like a card-testing bot.
None of this happens overnight. Ecommerce gained only about half a point of retail share a year between 2011 and 2019. Three decades in, it's 17.1% of U.S. retail sales. Goldman expects agentic commerce to follow a similar multi-year arc. It will start with things people already buy without thinking (groceries, household staples, tickets, electronics) and come last to luxury, apparel and furniture.
But small shifts add up. Goldman puts $2.6 trillion of U.S. spending in the categories most likely to go agentic first. If agents move just 2% of in-store card spending in those categories online, total ecommerce growth speeds up by about a point. McKinsey goes further, projecting up to $1 trillion of U.S. retail revenue orchestrated by agents by 2030, and $3 to $5 trillion globally.
How brands and merchants win
If the agent is the new storefront, the job is to be the product it picks. Here's the playbook I'd hand a brand or retailer today.
Make your catalog readable by machines.
Agents can't recommend what they can't parse. That means clean product data, live inventory and pricing, and clear shipping and return terms, exposed through the new protocols: MCP, Google and Shopify's Universal Commerce Protocol, and OpenAI and Stripe's Agentic Commerce Protocol. This is the new site speed. It's dull, and it decides who gets in the door.
Compete on what agents weigh.
Agents look at price, speed, availability, reviews and fit, not ad budget. Track how often you show up in AI answers for your category, and fix the gaps the way brands once fixed SEO. That discipline now has a name: generative engine optimization, or GEO.
Build your own agent where you own the relationship.
Walmart's Sparky works because it reuses Walmart's identity, payment and loyalty systems. Goldman expects store-branded agents like this to lead in the near term, while horizontal agents like Muse work through their trust problems.
Fight to be the default.
In the shift to mobile, the prize was the app install. In the agent era, the prize is being the brand the agent picks again and again for a given person. Replenishment, subscriptions and loyalty programs an agent can read and honor are how you get there.
Choose open or closed on purpose.
Amazon is closing the door. Shopify and Ticketmaster are opening it. For most small and mid-size brands, open is the right call, because agents judge on fit, which can surface merchants that could never outbid the giants on ads. Incumbents with big ad businesses face a real tradeoff.
Invest in brand where humans still decide.
In luxury, apparel and furniture, the agent builds the shortlist but a person makes the final call. Brand still carries weight there. Commodity and private-label categories are the most exposed, because an agent will happily swap your product for a cheaper equal.
Let the good bots in.
Fraud rules built to block bots will turn away paying customers. Visa's Trusted Agent Protocol and Mastercard's Agent Pay give merchants ways to verify legitimate agents. Plug into them early.
Where the startup opportunities are
Goldman's list of likely winners reads like a who's who of incumbents: Meta, Alphabet, Amazon, Walmart, Shopify, Visa, Mastercard. That makes sense. Models are becoming commodities, and the moats are distribution and trust. So where does a seed or Series A company fit? In the gaps the giants leave between them.
Agent visibility and analytics.
Brands need to measure and grow their share of AI recommendations. GEO tools are already crowded. The winners will tie visibility to actual revenue.
Agent-ready catalog infrastructure.
Most mid-market merchants aren't on Shopify and can't turn messy product data into feeds for every new protocol. That's a classic picks-and-shovels business.
Agent identity and intent verification.
Goldman calls proving what the consumer actually asked for "a major unsolved issue." "Know your agent" will become a category, the way "know your customer" did.
Disputes, returns and liability.
Online return rates are already around 17%, and about 90% of disputes go the consumer's way. Someone will build the claims layer, and someone will underwrite agent mistakes.
Vertical agents for the hard categories.
Horizontal agents will handle toilet paper first. Apparel fit, home projects and B2B procurement need specialists that know the domain.
Agent-native ads and attribution.
When no human clicks, today's measurement stack stops working. Someone has to rebuild it.
One caution for investors: platform risk is real. OpenAI rewrote its checkout model six months after launch, and Amazon is blocking agents it doesn't control. I'd favor companies that stay neutral across agents and protocols over ones that bet everything on a single platform's rules.
My take
Agents won't kill ecommerce. They'll change who the customer is. I expect this to unfold more slowly than the hype suggests, because consumer habits change more slowly than technology does. But the direction looks settled to me.
For brands, the real risk isn't that agents arrive too fast. It's waking up three years from now to find you've quietly dropped off the shortlist, with no dashboard that told you it was happening. For founders and investors, the first ecommerce wave was won by whoever solved trust: Visa, Mastercard and PayPal built some of the most durable franchises in finance. The second wave has the same bottleneck. The companies that make it safe for software to shop on our behalf are where I'd be looking.
Sources
Goldman Sachs Global Investment Research,
Americas Technology: Entering the Agentic AI Era for the Commerce Landscape
, September 24, 2026
Quartz via Yahoo Finance: Amazon blocks Meta's Muse AI agent
U.S. Census Bureau: Quarterly Retail E-Commerce Sales, Q2 2026