The First Great War Between Platforms and Protocols
The news did not get the attention it deserved.
A U.S. federal judge temporarily blocked Perplexity’s Comet shopping agent from using Amazon in the way Amazon says it had been operating: accessing protected parts of customer accounts and automating purchases through them. The injunction was granted on March 10, 2026, and the judge briefly stayed enforcement so Perplexity could appeal. Amazon’s lawsuit itself dates back to November 2025.
On the surface, this looks like a narrow legal dispute about unauthorized access, account security, and platform rules. That is the official frame, and it is not fake. Amazon argued that Perplexity’s system unlawfully accessed user accounts, disguised automated activity as human browsing, and created risks for customer data. The court said Amazon had shown “strong evidence” at this preliminary stage.
But if you strip away the legal packaging, this case is much bigger than one browser agent and one marketplace.
This is the first major visible battle in what I would call the great war of Platforms vs. Protocols.
What Amazon says this case is about
Amazon’s public position is straightforward: trust, security, and the integrity of the customer experience. In court, it argued that Perplexity’s agent entered password-protected parts of Amazon accounts without permission, ignored repeated requests to stop, and forced Amazon to spend money building countermeasures. Reuters reported that the court accepted Amazon’s argument that the company had already spent meaningfully more than $5,000 on defenses against this behavior.
Legally, that is enough to make the story sound mundane.
A platform says: “Someone automated access in a way we did not authorize.”
A court says: “Pause that while we sort it out.”
That is the official story.
It is not the whole story.
What the real economic conflict looks like
The deeper issue is not whether a bot clicked where a human should have clicked.
The deeper issue is who owns the decision layer in digital commerce.
Amazon is not just a warehouse with a checkout page. It is a full-stack commercial environment built around discovery, ranking, recommendation, sponsored placement, cross-sell mechanics, and monetized attention. Amazon’s advertising business alone generated $21.317 billion in Q4 2025. If you sum Amazon’s 2025 quarterly “Advertising services” figures from its own filings, the business reached about $68.6 billion for the year.
That number matters because it clarifies the real stakes.
If a human shopper comes to Amazon directly, Amazon gets to shape the journey: - what appears first, - what gets sponsored placement, - what gets bundled, - what gets cross-sold, - what gets seen, - and what gets ignored.
But if an external AI agent becomes the primary interface, that logic starts to break.
A strong shopping agent does not behave like a platform-optimized consumer. It behaves like a buyer-side optimizer. It does not care about glossy merchandising theater. It does not admire banners. It does not get hypnotized by “customers also bought.” It does not wander into the monetized maze. It tries to compress the entire journey into one cold question:
What is the best option for this user, under these constraints, right now?
That is wonderful for the user.
It is potentially devastating for the platform.
Because the moment an agent becomes the primary decision-maker, the marketplace risks being demoted from decision environment to fulfillment backend.
And that is the nightmare.
Why this is not really AI vs. retail
This is not a clean story about old retail fighting new AI.
Amazon itself is deeply invested in agentic systems. In its Q2 2025 results, Andy Jassy explicitly said Amazon’s own shopping agent was already being used by “many millions of customers.” In the same statement, he described a broader company-wide push into AI products and agentic infrastructure.
That detail is important.
Amazon is not rejecting agents in principle. It is rejecting a future in which someone else’s agent becomes the dominant customer interface over Amazon’s demand surface.
That is a very different conflict.
The issue is not automation as such.
The issue is control.
If Amazon owns the agent, the agent can preserve Amazon’s economics.
If Perplexity, OpenAI, Google, or another external layer owns the agent, then Amazon becomes one vendor among many in a machine-mediated comparison market.
That is the real threat.
This is where platform and protocol diverge
A platform wants you to stay inside its world.
A protocol wants to make the world interoperable.
A platform wants a captive interface, captive discovery, captive monetization, and captive loyalty loops. A protocol wants any compliant merchant, agent, and payment system to be able to talk to each other.
That is why this legal fight matters so much. It is not just about one company’s access methods. It is about whether digital commerce will remain centered on closed, high-control marketplaces, or whether it will migrate toward open agent-mediated transaction layers.
And the market is already showing both directions.
Walmart announced in October 2025 that it had partnered with OpenAI to let customers and members shop Walmart directly inside ChatGPT using Instant Checkout. Walmart framed this as the beginning of “AI-first shopping experiences” and explicitly described a move away from the old “search bar plus long list of items” model.
Shopify and Google went even more structural. In January 2026, they announced the Universal Commerce Protocol (UCP), an open standard designed so AI agents can discover merchant capabilities, negotiate supported actions, and complete transactions across merchants in a more interoperable way. Google described UCP as a common language for agentic commerce across discovery, purchase, and post-purchase support. Shopify described it as a way for agents to connect and transact with any merchant.
So while Amazon is using litigation to slow or contain one external agent path, other players are moving in the opposite direction: they are trying to formalize the agent layer.
That is why this is bigger than one injunction.
The innovator’s dilemma is now visible in commerce
Amazon is caught in a textbook strategic trap.
If it opens the doors
If Amazon fully opens to external agents, it risks losing control over the customer interface. The user might start with ChatGPT, Gemini, Perplexity, or another agent. The agent may compare Amazon against Walmart, Shopify merchants, brand sites, and specialized vertical sellers. In that world, Amazon still has logistics power, catalog scale, and trust — but it no longer fully controls discovery or merchandising.
That threatens advertising economics and weakens the strategic value of owning the storefront.
If it closes the doors
If Amazon closes the doors too aggressively, it risks becoming the company that defended the old UI while the market moved to a new one.
That is the dangerous part.
Because agentic commerce does not require every merchant to become a chatbot company. It only requires a sufficiently good protocol, enough merchant participation, trusted payment rails, and a strong enough user habit around delegating shopping decisions.
Once those conditions arrive, the center of gravity shifts.
And when the center of gravity shifts, the incumbent interface starts to look like legacy surface area.
Why the Perplexity case matters symbolically
Perplexity argued that Amazon was trying to block a model in which users choose their own AI layer — and, in Perplexity’s words, AI agents “don’t have eyeballs” for Amazon’s pervasive advertising. That line matters because it accidentally explains the whole conflict.
Agents do not value interfaces the way platforms do.
They value: - structured data, - capability exposure, - inventory access, - payment confidence, - delivery reliability, - policy clarity, - and machine-actionable workflows.
That is a protocol worldview.
Platforms, by contrast, have spent decades optimizing for visual attention, ranking power, UI persuasion, and monetized navigation.
That is why this clash feels so fundamental: both sides are rational, but they are rational in different economic grammars.
My take
The March 2026 injunction is not the final judgment on agentic commerce. It is not even close. It is an early legal win for platform sovereignty in one specific conflict.
But symbolically, it is much bigger than that.
It reveals the fault line that will likely define the next era of e-commerce:
- Platforms want to remain the place where decisions are shaped.
- Agents want to become the place where decisions are made.
- Protocols want to make the underlying merchant layer portable and interoperable.
- Users will increasingly prefer whichever option saves time, money, and cognitive load.
That is why I think this is the beginning of the first great war between Platforms and Protocols.
Amazon wants to keep you inside its carefully designed commercial universe.
External agents want Amazon to become just one vendor node in a wider machine-readable market.
The old battle in e-commerce was about traffic.
The next battle is about who gets to think on the buyer’s behalf.
And once that battle truly starts, the most valuable real estate in commerce will no longer be the homepage, the search result, or the sponsored slot.
It will be the agent.