Agent Wars!

Context

Meta's Muse personal agent became the first consumer AI agent to reach mainstream adoption, briefly overtaking ChatGPT for the number one spot on Apple's US free app charts. Days later, Amazon blocked Muse from shopping on its site, while Shopify moved the opposite direction and partnered with Meta to support Muse fully. NLW uses these two moves to explore what happens when personal agents start making purchase decisions on a user's behalf, and why platforms that spent years aggregating customers now face a choice between blocking agents or being reduced to a commoditized backend behind someone else's interface.

The Big Idea

Muse's breakout success turned personal agents from a tech-adopter curiosity into a mainstream force overnight, and Amazon's immediate block of its shopping access reveals the real fight underneath: platforms that built their business on owning the customer relationship and on advertising now have to decide whether to open up to agents or risk becoming an interchangeable, dumb pipe behind them.

Amazon's decision was framed as a security move, but multiple observers pointed out that agents do not look at ads, and Amazon made $76 billion in advertising over the past year. Shopify took the opposite bet, treating agent support as a way to compete with Amazon on convenience rather than defend an ad business.

Key Insights

Muse became the first mainstream agent

Earlier consumer agents like Manus, Open Claw, Hermes, Town and Instinct stayed either work tools people pushed through rough edges to use, or tech toys for early adopters. Muse is the first to show steady, growing adoption outside tech circles: it hit number two on Apple's free app charts at launch, then overtook ChatGPT for the number one spot, a position ChatGPT has rarely left in the four years since its release.

Users handed over trust fast

The tech press was skeptical people would connect their email, calendar, and personal data to a Meta product. Commentator Joe Weisenthal had predicted back in 2024 that "nobody cares about privacy," and that call held up: users gave Muse and Instinct access to email, messages, credit cards, location, screen access and logins within weeks of launch, with little visible resistance.

The agent now makes the decision

One widely shared account described Muse buying socks, ordering Whole Foods groceries, booking a cleaning service, and picking a burger restaurant the user had never heard of, with the user just saying "okay." The shift being described is not that the agent executes a choice the user already made. It is that the agent researches the options and the user defers to its recommendation, which is a different and bigger behavior change than automating checkout.

Amazon blocked Muse after breakout

Effective immediately after Muse's breakout week, Amazon barred Muse agents from shopping on its site, citing a conditions-of-use violation. This is consistent with Amazon's past posture: in November it sued Perplexity for circumventing its agent blockers, arguing that agent traffic imposes real serving costs beyond any terms-of-service question. What makes this instance notable is the timing, blocking access the moment a competitor's agent proved it could actually drive purchases.

Ad margins may explain the block

Retail margins at Amazon run close to zero to 3%, while advertising margins run near 70%. Because personal agents do not view or click on ads, giving them shopping access trades away Amazon's highest-margin revenue line to enable transactions on its lowest-margin one. Under this reading, Amazon's real customer has always been the advertisers and sellers who pay for placement, not the shopper, which is why blocking Muse protects the business model even if it frustrates the person shopping.

Shopify bet opposite of Amazon

Shopify announced a partnership giving Muse direct backend access for better search performance, plus official support in its agentic checkout product, Shoppay, across every Shopify store. Unlike Amazon, Shopify's revenue comes from powering merchants, not from an ads business agents would bypass, so opening up to agents costs it little and could pull demand away from platforms that stay closed.

Aggregators can be aggregated too

For over a decade, platforms like Amazon won by aggregating both users and the sellers competing for those users' attention. A personal agent that already knows a user's calendar, preferences and history can, in principle, compare Amazon, Instacart, DoorDash and Uber Eats and route the order to whichever offers the best deal. That flips the aggregation relationship: the platform that used to own the interface to the customer becomes just one more option the user's agent chooses between, and the agent's owner becomes the new aggregator.

Smaller platforms resist less

A platform with real scale, like Amazon, can afford to block agents because it does not depend on any single channel. A smaller platform, or one whose revenue comes mainly from transaction fees rather than ads, has the opposite incentive: it gains volume by making itself easy for agents to reach, since a user's agent will route demand toward whichever store cooperates. This is the dynamic Shopify is betting on.

Mental Models & Frameworks

Aggregation theory applied to agents

Ben Thompson's aggregation theory describes platforms winning by owning the interface between users and a fragmented set of suppliers. Applying it to agents: users increasingly want one agent that knows them and gets things done across every platform, while the platforms being aggregated want the opposite, to keep owning the direct relationship with the customer rather than become an interchangeable supplier behind the agent's interface. Use this lens whenever a platform's reaction to agent access seems to contradict simple revenue logic (a purchase is a purchase either way): the platform may be defending its position as the aggregator, not the transaction itself.

B2A: businesses now sell to agents

The idea that businesses increasingly sell not just to a human but to that human's agent, shortened to B2A. Under this model a product's real buyer is sometimes the agent doing the comparison and the purchasing, which means product and marketing decisions that assume a human is reading, clicking, or browsing may need a separate path built for agent access (APIs, structured data, agent-friendly checkout) alongside the human-facing one.

Trade-offs & Nuance

Agentic shopping may not scale

Several voices push back on how much agentic shopping actually matters for AI adoption broadly. The case against: everyday purchases like groceries or flights are not that hard to do manually, so there is limited upside to delegating them, and for many shoppers the browsing and discovery itself is part of the experience they enjoy, not friction to remove. One executive argued that saving a shopper 20% on a $50 purchase does not matter if the dopamine of the shopping process is what they were there for. The case for taking it seriously anyway: NLW notes his own confidence in the skeptical view is not high, since the direction of travel (Meta reaching mainstream scale with Muse in weeks) already surprised most predictions once.

Should platforms charge agents for access

One proposed resolution to platform standoffs like Amazon versus Muse is a business-development deal, where Meta pays Amazon for agent access, sharing data or a per-user fee, rather than either fully blocking or fully opening up. This preserves some of the platform's revenue while still letting the agent function. The trade-off: if every major platform starts charging for agent access, only companies with enough scale to pay those fees can build a truly general agent, which would concentrate the agent market around a few large players rather than opening it up.

Practical Application

Map your own product's B2A exposure

Identify which parts of your product currently assume a human is the one browsing, clicking, or reading, such as a pricing page, an onboarding flow, or a search results page, and check whether an agent acting for that user could complete the same task without ever rendering that experience. If it can't, decide whether that's a deliberate moat or an oversight.

Separate ad revenue from core value

Follow the logic behind Amazon's decision and ask, honestly, whether your product's real profit center is the thing your customer thinks they're paying for, or something adjacent like advertising or attention that an agent would bypass. If a large share of margin comes from something an agent-mediated experience would remove, that's the actual constraint shaping how open you can afford to be to agents, not a policy choice you can make freely.

Track which platforms embrace agents

Watch the emerging split between platforms blocking agent access (Amazon) and those actively courting it (Shopify), since this determines which platforms will see agent-driven demand and which will see it routed elsewhere. A product's distribution strategy in an agent-mediated world may depend more on which agents it partners with than on its own app or website experience.

Decide where your agent gates go

If you're building any kind of agentic workflow, decide explicitly where a human needs to approve a step versus where the agent should run unattended, since the episode's broader discussion (echoed elsewhere in the same show) treats this as still unresolved industry-wide. Removing a gate speeds things up but risks wasted or wrong actions; keeping it slows the system but limits exposure.

Questions to Consider

• If a personal agent could compare prices and availability across every retailer that sells what your company sells, would your product win on that comparison, or does part of your current advantage depend on customers not comparing?
• Does your business earn more margin from the transaction itself or from advertising and attention around it, and what would change if an agent bypassed the advertising entirely?
• Would you rather be the platform that owns the customer relationship and risks becoming a walled-off option agents route around, or the platform that opens up early and wins volume from being agent-friendly?
• Where in your own product or workflow would you want a human approval gate if an agent could complete the whole task unattended, and what would you be trading off by removing it?

Bottom Line

Muse's rise to the top of the App Store made personal agents mainstream almost overnight, and Amazon's immediate block of its shopping access shows that the real conflict is not about agents themselves but about who keeps owning the customer relationship and the advertising revenue built on top of it. Platforms with an ads business to protect, like Amazon, have reason to resist; platforms without one, like Shopify, have reason to embrace agents early and take the resulting demand.

Case Studies Mentioned

Amazon blocking Muse from shopping

Amazon barred Muse from making purchases on its site immediately after Muse's breakout week, citing a conditions-of-use violation, while officials framed the move as protecting how service providers choose to participate in agent access. The pattern matches Amazon's earlier lawsuit against Perplexity for circumventing its agent blockers over the added server cost of agent traffic. The lesson for PMs: a platform's stated reason for blocking a competitor (security, terms of service) is worth checking against its actual revenue structure, since the real motive here appears tied to protecting a $76 billion ad business that agents do not engage with.

Shopify's partnership with Muse

Shopify announced it would give Muse direct backend access for improved search, plus support inside its own agentic checkout product, Shoppay, across every store on its platform, mirroring an existing partnership it has with OpenAI. Meta's chief AI officer and Mark Zuckerberg both publicly framed the deal as making shopping and checkout easier inside Muse, with more partnerships to follow. The lesson: a platform without an ad business to protect can treat agent access as a distribution advantage rather than a threat, and use it to compete against a much larger rival.

Tools & Products

Notable Quotes

"The most insightful thing I read about technology 11 years ago was Ben Thompson's aggregation theory. Amazon blocking Muse is another perfect example. Customers want one agent that knows them and can get things done. The platforms being aggregated want the opposite." (Nicholas Bustamante)

"Amazon blocks Muse. Sure, they will blame it on security, but Amazon made $76 billion in the last 12 months on advertising. Agents don't look at ads. A new age of agentic battle has begun." (Joseph Carlson)

"I hate it as a consumer, but it's the right move for Amazon. Agents want to replace the storefront and level the playing field, but Amazon is big enough to say no." (Shilmonut)