A couple of years ago, if you were online shopping for running shoes, you might have started with a Google search for the best running shoes for a beginner, read some articles, visited a few websites, and eventually made a purchase from a specific brand or a retailer such as Amazon. Along the way, advertisements, sponsored listings, and recommendations competed to influence what you ultimately purchased.
01The traditional journey
One purchase. Many stops.
Chances to
influence you0
Tested & reviewed
The best running shoes for your first miles
Our pick: Brooks Ghost 18 · best overall
Brooks Ghost 18
★★★★★Daily trainer
Add to cartOrder placed
You found your pair.
Brooks Ghost 18Size 10
Running socksAdded at checkout
The journey is beginning to change
Increasingly, consumers are turning to large language models for product discovery and agentic assistants for execution. Instead of navigating between search results, review sites, retailers, and checkout pages, a consumer can describe what they want, receive a personalized recommendation, and rely on an AI agent for the transaction itself.
02The agentic journey
You ask. The agent acts.
Chances to
influence you0
What are the best running shoes for a beginner?
Start with a cushioned, everyday trainer.
Based on reviews and what’s in stock in your size:
Also consider Hoka CliftonSaucony Ride
Buy the Ghost 18 in size 10.
✓Size 10 in stock
✓Retailer selected
✓Checked out with your saved card
Purchase complete
Your shoes are on the way.
In this new journey, discovery is happening inside an LLM chatbox and the execution is handled by an agent. In this case, Meta recently released Muse, a personal AI agent designed to complete multi-step tasks on the internet for you rather than just answering questions like a standard chatbot.1 The journey itself has shortened, and the ad surfaces that were prevalent are beginning to shrink.
There are fewer opportunities to shape a purchase decision, but the remaining opportunities are becoming significantly more valuable. A consumer who is asking an AI assistant what to buy is a consumer that is ready to make a purchase.
The important question to ask is who gets to influence the consumer before the agent gets to execute the purchase. This creates a divide between businesses looking for incremental demand and businesses that already own it. In today's market, Meta and Amazon are trying to control the piece of the flow where they get to influence the consumer.
Agentic Changes
Agentic commerce changes the value chain because an agent can separate consumer intent from the merchant that ultimately fulfills it.
In my previous post, "Intent Share," I wrote that whichever platform becomes the place where consumers go to execute on their intent gains leverage over the businesses trying to fulfill that intent. More users with intent = more bargaining power.
As noted earlier this week, Amazon blocked Muse from facilitating checkouts on Amazon.2 We can also see Amazon in the past trying to keep outside agents from getting inside its castle with lawsuits against Perplexity3 and blocking other shopping agents from Google and OpenAI.4 Interestingly, Amazon is not against agentic shopping itself. Its own Buy for Me feature allows Amazon's agent to purchase products from other retailers when they are not available on Amazon.5 Amazon isn't anti-agent, it is just against someone else's agent benefiting from the demand it has generated.
At the same time, Muse has integrated with a growing number of businesses, with the Muse mascot appearing in a bunch of different costumes. (See Expedia, Instacart, etc.)6
Meet the cast
Same Muse. Different roles.
ExpediaTravel
ShopifyShopping
InstacartGroceries
Opting In
What is interesting is why these different companies are opting in to the agentic experience.
On one hand, many of these integrators are software-based businesses that are looking for incremental demand. On the other hand, Amazon has invested heavily in physical infrastructure and has honed its process such that it has generated significant demand and has become muscle memory for a consumer looking to make a purchase. By that logic, Amazon should welcome incremental orders from agents. The fact that it is willing to refuse them suggests that the shopping session is worth more than the marginal order.
Companies like Instacart are more inclined to accept agent traffic because they expect incremental demand from AI agents executing transactions. If you want to order groceries with an agent like Muse, and it tells you that you can do so by linking your Instacart account, more users would be inclined to download Instacart and plug in their information.
Expedia and Instacart are demand seekers. The demand from consumers for a specific activity is where they can place themselves in the flow and obtain incremental sales. Retailers with interchangeable inventory and costly customer acquisition welcome agent demand, while retailers with strong direct traffic, valuable advertising inventory, loyalty, or exclusive products have more ability to refuse. Amazon is a demand owner.
Owning the Journey
Why is Amazon resisting? Amazon has raised concerns about security, Muse not identifying itself while browsing, and Meta not informing Amazon before Muse began accessing its store.7 But if Amazon is so customer-obsessed, why not allow customers to purchase goods the way they want to?
For one, sales that go through an agent can cannibalize existing sales. The expected revenue generated by a shopper on Amazon's direct website for the same good will likely be greater than that generated through an agentic workflow, due to the loss of Amazon-owned ads, opportunities to increase the shopping cart, sponsored placements, data about browsing behavior, and the customer relationship itself. Amazon does not just want the transaction; it values the shopping session.
Amazon is more than a checkout company. Amazon does not just capture demand; it also influences demand after the shopper arrives. Someone might go to Amazon thinking that they need running shoes. Amazon can influence which shoe, which brand, how much they spend, and what else they purchase. An external agent threatens to make those decisions before Amazon ever sees the customer.
The part of the product journey where consumers are still undecided is where advertisements are most compelling. This is one of the most important moments because the user has intent to purchase something; they just do not know what to get yet.
The number of ad surfaces shrinks, while the value of the remaining surfaces increases. With this lens, I expect advertising to come to these agentic chats and interfaces, and be very valuable due to the high intent and the potential for undecided consumers to make their decisions in these interfaces.
This is especially interesting because Muse is built by Meta, which generated more than $190 billion in advertising revenue last year, while Amazon generated nearly $69 billion from its own advertising services.8 These are two advertising businesses competing for the same high-intent moment.
Substitutes
In Ben Thompson's "Amazon Blocks Muse, Amazon's Moat, Aggregator v Aggregator" piece, he argues that Amazon is already the aggregator of demand.9 Muse wants to become a layer through which consumers express and execute that demand. However, when an agent wants to make a purchase for a good or a service, and the main merchant or vendor does not allow it, the agent will look for substitutes. If the user allows the agent to make a substitute, Amazon is at risk of losing the entire order. If a user does not allow the agent to make a substitute, they have to return to Amazon directly, allowing Amazon to retain both the customer and the shopping session. As Muse or other AI agents begin to take a greater share of executed orders from users, leverage begins to shift toward the agent.
This is where the breadth of merchant integrations matters. Muse now has access to Shopify's catalog and has announced integrations with retailers including Walmart, Best Buy, Gap, Sephora, and Wayfair.10
Walmart's relative position in online shopping is lower than Amazon's.11 Amazon may be trying to defend its position, whereas Walmart is trading the session value for incremental demand because it's losing sessions to Amazon anyway.
A reinforcing loop
The integration flywheel
This flywheel only works if the merchant or product is substitutable. An agent has enormous leverage when the task is simply to purchase paper towels, whereas the agent has less leverage when it is asked to purchase a specific brand of paper towels at a specific retailer. One advantage Amazon has is its physical moat of logistics centers, Prime, returns, and consumer trust. These make Amazon harder to substitute.
Personal agents also become sticky as they accumulate account information across platforms. Switching to another agent could require time and effort, creating higher switching costs as users become more accustomed to agents doing this work for them.
The next battle in commerce may not be over who owns checkout, but over who owns the consumer's intent when the product at checkout is substitutable.
Notes
- Meta introduced Muse on September 8, 2026. Meta’s announcement. ↩
- Amazon began blocking Muse on September 20, 2026. GeekWire; Bloomberg. ↩
- Amazon sued Perplexity in November 2025 over the shopping agent in its Comet browser. Payments Dive. A preliminary injunction Amazon won in March 2026 was lifted on appeal in August, per GeekWire. ↩
- Amazon’s robots.txt disallows AI agents and crawlers from Google, including Project Mariner, and from OpenAI, including ChatGPT’s live-browsing agent. Modern Retail, August 2025; Modern Retail, November 2025. ↩
- Amazon began testing Buy for Me in April 2025. Amazon’s announcement. ↩
- Expedia and Instacart announced Muse integrations on September 22, 2026. Instacart’s announcement; Skift on Expedia. ↩
- Amazon’s stated reasons, as reported by GeekWire. ↩
- Meta reported $196.2 billion in 2025 advertising revenue (Q4 2025 results). Amazon’s advertising services net sales totaled $68.6 billion across its four 2025 quarters (Q4 2025 results). ↩
- Ben Thompson, “Amazon Blocks Muse, Amazon’s Moat, Aggregator v Aggregator”, Stratechery, September 22, 2026. Subscriber update. ↩
- Shopify catalog access, Shop Pay checkout, and the retail partners are listed in TechCrunch’s roundup. ↩
- Amazon held an estimated 40.5% of US ecommerce sales in 2025; Walmart held 9.2%. Statista. ↩
