When marketplaces become the enemy: how AI agents are changing commerce

Amazon filed a federal lawsuit against Perplexity in November 2025. Legal theater, some said. On June 11, 2026, the Ninth Circuit Court of Appeals heard oral arguments in Seattle — and the question at the centre of the case has implications far beyond two companies fighting over shopping.

This isn't about violating rules. It's about how power works on the internet — and how it's changing.

How marketplaces built empires on control

When Amazon created their marketplace, they solved a real problem: people didn't want to navigate hundreds of separate stores. Aggregation made sense. Genius, even.

But over time, something shifted. Major platforms realised that controlling the entrance — controlling what users see — was a money-printing machine.

Instead of surfacing the best product, they began surfacing the most profitable one. Sponsored results now take up half the screen. Users arrive and see a curated selection, not an open market.

Amazon's advertising revenue hit $15.69 billion in Q2 2025 alone, growing 23% year-over-year. Not from shipping. Not from commissions. From selling placement. That was only possible because they controlled the only entrance.

That's where the power came from. And that's where the fear comes from now.

When an AI agent clicks on a product

Imagine you tell an AI agent: "Find me running shoes that are comfortable, within my budget, with two-day delivery."

The agent doesn't go to one marketplace. It doesn't visit one store. It searches across Amazon, Zappos, Nike, and smaller manufacturers you've never encountered. It compares on your actual criteria.

Here's what matters: the agent doesn't click on ads. Sponsored placement is invisible to it. It only processes criteria — quality, price, shipping. It has no incentive to show you a worse product.

Marketplaces don't lose traffic in this scenario. They lose power.

Why this is more than convenience

Marketplace dominance survived through concentration. When choice is limited to one platform, the platform can manipulate results, raise fees, and place counterfeit brands next to originals while profiting from both.

AI agents redistribute access. A small brand with better pricing can now be discovered as easily as a category leader. That's real competition — not competition for algorithm placement, but competition on the actual merits of the product. This is the shift that GEO research documents at the content level: generative engines evaluate quality directly, not platform authority.

For major platforms, that's existential.

Why platforms fight instead of adapt

Here's the uncomfortable truth: major platforms could build AI agents that operate in users' interests. They won't.

Because an agent that works honestly would destroy the advertising model. An agent that shows the best product instead of the sponsored one is economically suicidal for a company that earns $15 billion per quarter from selling placement.

So instead: lawsuits. Demands that AI agents identify themselves. Arguments about Terms of Service violations.

Amazon warned Perplexity at least five times starting in November 2024, then filed a federal lawsuit in November 2025 under the Computer Fraud and Abuse Act. The core claim: Perplexity's Comet browser agent accessed Amazon's systems without authorisation — even though users had explicitly given it permission to do so.

The real message was simpler: we understand the model is changing and we're losing our advantage.

The legal line that matters

On March 9, 2026, U.S. District Judge Maxine Chesney granted Amazon a preliminary injunction blocking Comet from accessing Amazon's logged-in pages. The key finding in the ruling: Comet accessed Amazon accounts "with the Amazon user's permission, but without authorization by Amazon."

That distinction — user permission versus platform authorisation — is the crux of the case. Perplexity argued Comet is an extension of the user, not an independent bot. Amazon argued its Terms of Service govern who is authorised to access its systems, regardless of what the user wants.

The injunction was stayed pending appeal. On June 11, 2026, a three-judge Ninth Circuit panel heard oral arguments in Seattle. Perplexity's attorney Chris Michel put it directly: "Amazon seeks to stretch criminal statutes that prohibit computer hacking to sue Perplexity simply because Amazon wanted its own customers to access its website its preferred way."

The breadth of support for Perplexity's position is notable. Mozilla, the Electronic Frontier Foundation, the ACLU, and other digital-rights groups filed amicus briefs arguing that a ruling for Amazon would expose developers and everyday users of browser extensions and automated tools to criminal or civil liability — and would deter journalism and public-interest research.

No decision was issued from the bench. The Ninth Circuit typically issues opinions in CFAA appeals on a multi-month timeline.

The outcome will determine whether platforms can legally close themselves off to AI agents acting on behalf of real users. Every retailer, marketplace, booking platform, and SaaS service that gates content behind a login is sitting behind the same legal question. It's the first major federal appellate test of how computer fraud statutes apply to agentic software — and the decision will be cited long after this particular dispute is settled.

What this means for fair competition

Much of the argument in favour of platforms sounds like: "They violate Terms of Service. They should identify themselves. They should pay for access."

These are all ways of saying: platforms should remain gatekeepers.

But Terms of Service aren't law. They're rules a platform wrote for itself. When AI agents serve the platform's interests — driving more purchases, more engagement — they're innovative. When they shift the balance of power toward users? They become a violation.

It's structurally similar to what happened when writers tried to restrict the printing press, or when ATMs threatened bank tellers. The system protects itself before accepting transformation.

How this looks in practice

The strategic question for businesses is concrete: how does an AI agent know your company exists?

Right now, most companies answer that question by registering on marketplaces. Because that's where the traffic is.

But when AI agents become the primary discovery channel for products and services, the answer changes. Katherine Black, who leads Kearney's research on agentic commerce, puts it directly: "The next retail arms race won't be for clicks or shelf space. It will be for algorithmic preference — being the brand an AI agent consistently selects".

That requires visibility to agents, not just presence on platforms. Structured, machine-readable data that independent agents can find and parse, regardless of which marketplace controls the front door. Regardless of who wins in court. What that looks like in practice is a separate question — but the direction is clear.