Something rather intriguing has been happening in the global e-commerce industry lately.

In response to Meta's AI agent product Muse, Amazon chose to restrict access — while Shopify took the completely opposite stance, actively embracing it.

Both are e-commerce giants. So why have the two companies adopted such diametrically opposed attitudes toward AI agents?

Shopify co-founder and CEO Tobi Lütke believes this actually reflects a fundamental difference between two business models.

Amazon is more like a digital department store, relying on a vast product inventory, traffic, and transaction infrastructure to build its competitive advantage. Shopify, by contrast, serves millions of independent merchants, enabling them to sell through any channel.

For the former, AI agents may mean losing control over the shopping entry point. But for the latter, agents may instead open up new transaction opportunities.

Tobi Lütke recently appeared on the "Sources" podcast hosted by Alex Heath for an 84-minute interview centered on AI shopping, the agent business model, and the future of the entire e-commerce industry.

Among the many points raised, several stand out as particularly worth paying attention to.

First, the point at which e-commerce intervenes in users' shopping decisions has moved upstream. AI is transforming shopping from product consumption into intent consumption.

Take a home gym, for example. In the past, users might first break down the specific products they needed, then search for individual items like treadmills and purchase them one by one. Now, they simply ask AI to buy everything needed to set up a home gym.

Second, Tobi is skeptical of the widely assumed business model in which personal agents take a commission. Existing e-commerce already has a complete and mature profit-distribution system, so the margin that can truly be freely allocated may not be as large as people think.

Third, in the agent era, the traditional advertising model may face a fundamental trust problem. If an AI assistant takes money from merchants and then recommends products to users, does it represent the buyer or the seller?

Today, Silicon Base Lord will use this interview to explore how AI is changing the underlying logic of the e-commerce industry.

/ 01 / From product consumption to intent consumption: shopping decisions have moved upstream

Over the past two decades, the e-commerce industry has been competing around one thing: how to make it easier for consumers to find the products they want to buy.

From the earliest product categories and keyword searches to later personalized recommendations and livestream selling, all of this has essentially been about optimizing the matching efficiency between products and consumers.

But the emergence of AI agents may fundamentally change this logic. In the interview, Tobi Lütke mentioned that Shopify has already observed a very interesting phenomenon.

At present, shopping transactions completed through AI agents are still at a relatively low base, so it is not particularly meaningful to simply discuss growth rates. What is truly worth paying attention to is that consumer purchasing behavior is changing.

A typical case is the home gym.

In the past, if someone wanted to build a gym at home, they might first search for a treadmill, then research equipment such as dumbbells, training benches, and floor mats, and finally purchase them one by one according to budget and space.

Throughout the process, e-commerce platforms were mainly responsible for providing product information and completing transactions. As for what to buy, how to match it, and how to allocate the budget, consumers still decided for themselves.

But now, some consumers have begun making completely different requests to agents:

Help me buy everything needed to set up a home gym.

This may look like merely a change in the prompt, but the business logic behind it is actually completely different.

Because what the agent faces is no longer a purchase demand for a single product, but a project that requires a complete solution. It needs to understand the user's budget, space, and fitness goals, determine which equipment is worth buying, and then complete product screening, matching, and ordering.

Tobi even imagines that in the future, consumers may only need to upload a photo of a room or a floor plan, and AI can participate in the entire procurement planning process.

There are many similar scenarios. For example, wanting to complete a 3D printing project with children on the weekend. In the past, parents might need to first research the project plan, then figure out which electronic components, magnets, and printing consumables are needed, and finally place orders separately with different merchants.

Now, they only need to tell the agent what project they want to complete, and it has the opportunity to handle the material list, product procurement, and delivery arrangements all at once.

This is an important trend observed by Tobi: more and more complex, high-ticket purchasing behaviors are beginning to be completed through agents.

I think the truly important change here is that AI is moving the starting point of shopping decisions from product demand further upstream to life demand.

In the past, from intent to specific products, there was still a huge decision-making process in between, and this work was often completed by consumers themselves.

Now, AI has the opportunity to turn these decision-making tasks into a service that can be provided at scale.

This means that the value of agents may not only lie in improving existing shopping efficiency, but may also create consumption that did not happen in the past because decision-making costs were too high.

/ 02 / Why does Amazon block agents, while Shopify actively opens the door?

When consumers begin to hand shopping decisions over to agents, a new question follows:

What should e-commerce platforms that originally controlled the consumer entry point do?

The different choices of Amazon and Shopify happen to provide two very representative cases.

In Tobi's view, this divergence is actually not surprising at all.

Over the past two decades, the two companies have represented two different paths of e-commerce development.

Amazon's business model is closer to a digital department store. Consumers enter Amazon, search for products, compare prices, check reviews, and finally complete purchases. The platform attracts consumers through a huge product supply, logistics fulfillment capabilities, and transaction system, and then uses traffic to help merchants complete sales.

Therefore, where consumers search, which products they see, and what they ultimately choose to buy are very important to the platform's commercial value.

Especially the advertising business. When merchants need to buy on-site ads to gain exposure, consumers' search and browsing behavior become important monetization resources for the platform.

But agents may change all of this. Suppose a consumer asks AI to help buy home fitness equipment. AI directly completes demand analysis, product comparison, and purchasing decisions, and finally only needs to call the e-commerce platform to complete the transaction.

Then many core parts of traditional e-commerce platforms may be bypassed. Consumers may not necessarily need to open the e-commerce app, nor browse products one by one, and certainly do not need to make decisions according to the platform's recommendation order.

For platforms that rely on traffic distribution and advertising monetization, AI agents may not just be a new shopping channel, but potentially a competitor for the entry point.

Of course, Tobi did not assert that Amazon restricted Muse specifically to protect advertising revenue. He also acknowledged that aggregator platforms do have a stronger incentive to maintain control, and that the current technical implementation of agents also has its problems.

But from the perspective of business models, the potential conflict between the two is already quite clear.

By contrast, Shopify's situation is completely different. Shopify does not operate a large e-commerce marketplace in the traditional sense, but provides independent merchants with website building, payments, order management, and sales tools.

Consumers can place orders through merchants' independent websites, or complete purchases through channels such as social media and offline stores.

Tobi has a very simple judgment about this: the more places merchants can sell products, the greater Shopify's value.

Therefore, whether the final order is placed by the consumer personally or by an AI agent representing the consumer, it is acceptable to Shopify. What it truly cares about is whether the transaction can be completed smoothly and whether merchants can get more orders.

This is also why Shopify is building more APIs and transaction systems so that agents can more conveniently call product and checkout capabilities.

The difference between the two business models is thus reflected.

Amazon needs to consider how to maintain the relationship between consumers and the platform, while Shopify focuses more on how to help merchants connect to new consumer entry points.

Of course, this does not mean that Shopify will definitely become the winner in the agent era. But at least from the perspective of business models, the conflict of interest between Shopify and agents may be much smaller than that of traditional traffic-based platforms.

I think this is also the key to understanding the next round of e-commerce competition.

When AI begins to control shopping decisions, the focus of competition among e-commerce platforms may shift from how to keep users on their own platforms to how to make their products and services preferentially used by agents.

/ 03 / Agents want to take 2% from every transaction — but who pays?

If AI changing the way people shop has already begun to show signs, then another more realistic question is:

How exactly should agents make money?

In this interview, host Alex Heath mentioned a common idea among current Silicon Valley agent startups.

Many entrepreneurs believe that in the future, personal agents can establish a business model through transaction commissions. For example, if AI helps a consumer complete a $100 order and takes 2% from it, it can earn $2.

At first glance, this model seems quite reasonable.

After all, the agent helps users complete product research, comparison, and procurement, creating real value, so taking a certain percentage of transaction revenue also makes sense.

But Tobi directly raised a question: who exactly pays this 2% commission?

There is a reality here that is easy to overlook. E-commerce transactions themselves already have a complete and mature profit distribution system: payment institutions, credit card networks, e-commerce service providers, and other participants all need to obtain corresponding revenue from transactions.

According to Tobi, of Shopify's roughly 3% transaction fee, a large part actually flows to credit card companies. In other words, there may appear to be several percentage points of transaction revenue, but the margin that can truly be freely allocated may not be that large.

Now, agents suddenly join the transaction chain and also want to take 2% from it, and the problem becomes complicated.

If merchants bear it, that is equivalent to further increasing customer acquisition costs. If the platform bears it, it means existing participants need to give up their own profits. If consumers bear it, then the agent must prove that the value it provides is enough for users to pay extra.

Therefore, Tobi is not in a hurry to judge that commissions will definitely become the final business model of agents.

In his view, the more important question is how much value agents can actually create for users.

For example, an agent can help consumers compare the fee schemes of different banks, discover unreasonable long-term payment fees, and even automatically complete service switching. If it can help users save hundreds of dollars each year, then consumers have every reason to pay a certain fee for it.

Under this model, agents do not necessarily need to participate in product transaction commissions, but can charge users directly. In the future, there may also be pay-per-use, subscription fees, or hybrid models related to transactions.

But more worth discussing than the business model is another question raised by Tobi: when agents truly make decisions on behalf of consumers, the traditional advertising model may encounter a trust crisis.

Suppose you have a personal shopping assistant that knows you very well, knows your budget, lifestyle habits, and purchasing preferences, and can help you compare different brands and choose the most cost-effective products.

But one day, you discover that it recommends a certain product not because this product is most suitable for you, but because the merchant paid it an advertising fee.

Then, will you still trust it?

In Tobi's view, this constitutes damage to user trust.

The traditional advertising model mainly stands on the seller's side. Merchants pay for exposure, hoping to influence consumers' purchasing decisions. But the value of personal agents is precisely built on the premise that they represent the buyer's interests.

If an agent must both represent consumers in choosing the most suitable products and decide the recommendation order based on fees paid by merchants, a conflict of interest may arise between the two.

This is also why the advertising business model in the agent era may need to be redesigned.

However, Tobi does not believe that advertising will disappear because of this.

Because no matter how powerful AI becomes, new brands still need to find consumers, and new products still need to gain exposure.

In the past, the advertising systems of Facebook and Google helped a large number of small merchants find their first batch of users. In the future, the agent world will likewise need some mechanism for product discovery and supply-demand matching.

The only difference is that this mechanism must answer a new question: how to help merchants gain traffic without damaging consumers' trust in AI assistants?

This could be one of the most significant directions in the commercialization of Agents. Whoever can strike a balance between sellers' customer acquisition needs and buyers' decision-making trust may succeed in building a new generation of commercial distribution platforms.

Text / Aqi