Iris Holm: Good to be back — how'd your week treat you?
Lila Soto: Oh, weird, actually — I saw something that I keep thinking about. There's a Meta product called Muse, a shopping agent, and Meta and Walmart publicly partnered so it could work on Walmart's site. And users are reporting it just gets blocked anyway. Walmart's own anti-bot systems flag it.
Iris Holm: Right. A contract doesn't touch infrastructure.
Lila Soto: Which, yeah — but also, Amazon didn't even try the partnership route. September 20, 2026, they just blocked Meta Muse. And before that, November 2025, they sued Perplexity over its Comet browser agent. A federal court ordered Comet off Amazon while the lawsuit proceeds. So we have actual legal precedent forming in real time.
Iris Holm: First major legal test of whether a platform can enforce its terms of service against an agent acting on a user's behalf. And the court said yes — at least provisionally.
Lila Soto: So the thing I want to figure out — if platforms can block agents even when users want them, and even when there's a deal on paper, what does the promise of AI shopping agents actually mean?
Iris Holm: Think about a mall that lets shoppers in freely but bans the personal shopper you hired yourself — while running its own personal-shopping service in every store. That's Amazon. They block Meta Muse, they sue Perplexity over Comet, and meanwhile they're running Rufus, Alexa for Shopping, Buy for Me — all first-party agents, all inside the walls.
Lila Soto: Wait, so Yelp is literally charging agents to get in?
Iris Holm: Explicitly. Paid data-licensing program. Access is a revenue line. Not a risk to manage — a product to sell. That's not a security posture. That's a toll booth.
Lila Soto: And eBay made it even more literal — they updated their user agreement in February 2026 to ban buy-for-me agents by name. Not 'unauthorized bots.' The actual category of thing.
Iris Holm: Named the threat. Which tells you they knew exactly what they were protecting. JPMorgan Chase and Charles Schwab say security — and I mean, fine, maybe — but that rationale also happens to lock a third party out of their customer relationships and transaction data. Delta Air Lines has zero agent deals as of October 2026. Zero. Not one. LinkedIn and Facebook block nearly every bot they haven't approved themselves. The pattern isn't security. The pattern is: we monetize the access.
Lila Soto: Huh — and even when a platform wants to let a specific agent in, there's apparently no real infrastructure to do that cleanly yet. Which sounds like it gets worse before it gets better.
Iris Holm: Much worse. But that's the next layer. The question right now is simpler: Shopify opened its storefronts to agents, so the permissive move is possible. Amazon chose not to. That's a business decision wearing a security badge.
Lila Soto: But Shopify *can* open the door, okay, so the permissive move is real. But picture someone on a Friday night, on their smart TV, asking a voice agent to book a last-minute Delta flight. The agent just... hits a wall. Not because Delta put out a press release saying no. Because Delta never built a pathway for agent-based booking at all. There's nothing to knock on.
Iris Holm: Delta has zero agent deals as of October 2026. Zero. Not one signed agreement. So the agent isn't being rejected — there's no handshake protocol to even attempt.
Lila Soto: Which is — I mean, that's a different problem than blocking, right? That's just absence.
Iris Holm: Absence that gets enforced by mature tools. CAPTCHA, rate limiting, IP-blocking of known AI infrastructure, advanced fingerprinting — those aren't new. They're already deployed widely. The blocking side is built. The credentialing side — the 'yes, this is a trusted agent, let it through' side — that's not.
Lila Soto: And Web Bot Auth was supposed to be that, kind of? Cloudflare, Visa, Mastercard announced it October 2025 — a standard for agents to prove their identity to platforms. But I guess... it's still a pilot? I haven't seen any confirmation it's actually been adopted broadly.
Iris Holm: Early-stage. And look — Cloudflare flipped to blocking AI crawlers by default for all new sites in July 2025. Three months later they're selling per-page crawler charging. So the company announcing the credentialing standard is also the company that just turned blocking into a baseline and access into a pricing tier. That's not a contradiction — that's a strategy.
Lila Soto: So even if Delta wanted to let a trusted agent in tomorrow — like, genuinely wanted to — there's no reliable way to do that yet. The tools to say yes don't exist at scale. Only the tools to say no do.
Iris Holm: That's actually where I land, yeah. The patchwork right now — Amazon blocks outright, Yelp charges per license, Shopify opens — that's probably not temporary chaos. That's probably the template. robots.txt eventually produced rough norms across the early web, but it took years and the stakes were just indexing. Here the stakes are transaction revenue, liability, customer ownership. The commercial pressure to *keep* the patchwork is orders of magnitude higher.
Lila Soto: So the question isn't whether agents will exist. It's whether your agent will work where you actually need it to — or only where someone already cut a deal.
Lila Soto: Kind of a strange place to end up, given how the whole thing was sold. Anyway — this was good. Genuinely.
