Who Owns the Intent?

Amazon v. Perplexity is the first shot in Big Tech vs. Big AI — and the real prize is a trust layer nobody has built yet.

Amazon has sued Perplexity. It won't be the last of these; expect a long run of Big Tech vs. Big AI lawsuits. On the surface, it's a dispute over agentic commerce. I think it's really a fight over the future architecture of the internet.

If you haven't used Perplexity's Comet browser (or OpenAI's near-identical Atlas), it's a genuinely different way of using the web: the AI agent browses for you. You watch it type, click, and move the cursor — the way IT does when it remote-controls your screen. It also compares and buys products for you, so no more clicks and tabs, just "buy me the softest toilet paper." I'm not sold on the experience yet, but your mileage may vary.

Amazon, unsurprisingly, isn't thrilled. Its suit accuses Perplexity of "trespassing" its site by disguising automated agent activity as ordinary human browsing. Perplexity fired back that users should be free to choose their own digital assistants — and that Amazon is bullying a startup to protect its own turf.

But this was never really a fight about shopping bots. Agentic commerce is coming, and Amazon knows it — the company launched its own "Buy for Me" agent earlier this year. (Whether agentic commerce actually ends up mattering is a separate question; see also: metaverse commerce, livestreaming commerce, voice-assistant commerce, IoT commerce.) The deeper tension is platform sovereignty vs. agentic autonomy. We've watched versions of this before — Google vs. publishers, Apple vs. developers, Facebook vs. everyone. Every platform wants to own not just the user and the transaction, but the intent that leads to it.

So here's what the lawsuit actually tells us:

Incumbents will not go quietly. Expect Apple, Meta, and Google to restrict third-party agents and lean hard into first-party, "compliant" assistants. Meta just moved to push third-party AI assistants off WhatsApp, leaving Meta AI as the only one standing. The open agent ecosystem was never going to last.

Agent authorization is the new KYC. Platforms will demand a way to verify that an AI genuinely represents an authorized, consenting user and is acting within the rules. That verification layer doesn't really exist yet.

Which means there's clear whitespace for trust infrastructure. Agentic commerce needs compliance rails — most likely middleware that manages agent identity and consent protocols. The startups that build and own that trust layer will quietly become the gatekeepers of agentic AI.