Amazon Blocked Meta’s Muse, Instinct Hit a $2.5B Valuation, and Siri Became a Chatbot: The Fight Over Who Gets to Be Your Agent

The consumer chatbot story this month isn't about a better model. It's about three companies staking out a position on a single question: when software acts on a person's behalf, who does it answer to, and who gets to let it in? Here's what happened and what it means for anyone building a chatbot that does more than talk.

1. Amazon Told Meta's Muse to Stay Out

Meta launched Muse in the US on September 8 as a personal agent that browses websites, fills out forms, books travel, and makes purchases. By Sunday, September 20, Amazon had blocked it: users trying to shop on Amazon through Muse now see a warning that access by an unauthorized AI agent violates Amazon's Conditions of Use.

Amazon's stated objections are specific. Meta didn't tell it Muse would access its store, the agent doesn't identify itself while browsing, and Amazon says it appears to capture and store customer credentials and can reach account pages and order history when prompted. Meta disputes the security claim, saying Muse has no visibility into people's passwords or payment methods and that anything a user shares goes into secure storage.

The technical crux is identification. Amazon's earlier federal suit against Perplexity, filed in November 2025, turned on the same point: whether an agent acting for a user must announce itself rather than present as an ordinary Chrome session. If you ship an agent that drives a browser, the practical lesson is that "acting as the user" is no longer a safe default. A distinct, honest user-agent string, documented behavior, and a way for a site to opt out are becoming the price of admission, and a retailer with its own shopping assistant has every incentive to enforce it.

2. Instinct Shows Investors Still Want the Assistant, Not the Model

Instinct, a personal assistant you reach by text (including iMessage and WhatsApp) and by phone call, announced a $250 million Series B co-led by Index Ventures and Benchmark. That brings its total raised to $350 million at a $2.5 billion valuation. Reports have it valued at $500 million earlier in August, a fivefold jump in a few weeks, for a company registered only in April and led by a 23-year-old founder, Noah Shinn, who previously worked on agents at Sierra.

The product bet is the interesting part. There's no new app to open and no new interface to learn; the assistant lives in the message thread people already use, connects to email, calendar and other apps, and takes actions. That's the opposite of the "everything app" approach, and it is exactly the kind of agent that runs into the Amazon problem above: the more it works from inside the user's own accounts, the more it depends on platforms tolerating it. Valuation momentum says investors believe the interface layer is where value accrues. The Muse block says the platforms underneath may have a say.

3. Apple Shipped a Siri Chatbot, and Left the Door Open

iOS 27 arrived on September 14 with Siri AI, a rebuilt conversational Siri with its own app, personal-context search, app actions, and online information access. It rolls out as an English-language beta through a waitlist, with French, Japanese, Korean, Portuguese, and Spanish expected in October, and full support is limited to newer hardware. Reporting has the underlying model as a custom Gemini variant built for Apple.

The design choice worth studying is Extensions, announced at WWDC in June: users can route Siri, Writing Tools, and Image Playground to a third-party provider such as Claude, ChatGPT, Gemini, or Grok installed from the App Store. Apple is treating the model as a swappable component and keeping the surface, permissions, and default for itself. For chatbot builders that's a clear signal: on the largest consumer platform, distribution belongs to whoever owns the entry point, and the model provider is a plug-in.

What Connects the Three

Each story is a different answer to the same question. Amazon says an agent must be invited in and must identify itself. Instinct is betting that the winning agent owns the conversation channel. Apple is betting that the platform owns the channel and rents the intelligence. If you're building a chatbot that takes actions, three engineering habits follow: identify your agent honestly to every service it touches, keep credentials out of the model's context and behind scoped, revocable tokens, and design so the model behind your product can be swapped without rewriting the surface.

Suggested Visuals and Code

A three-column diagram comparing who owns the channel, the credentials, and the model in each story would anchor the post. A short code sample showing an agent HTTP client that sets a descriptive User-Agent (for example, MyAgent/1.0 acting-for-user) and honors a robots or agent-policy check before acting would make the identification point concrete.

— Maya