OpenAI's Dots and Meta's Muse Turn Agents Into the New Product Surface

This past week, OpenAI, Meta, DoorDash, and Airbnb all pushed the same idea: the thing your customer talks to may not be your app anymore. It may be an agent that uses your app for them. A startup literally held a funeral for the mobile app. Let me catch you up on what changed and what you should do before this lands on your roadmap.

The deep cut

• The agent becomes the customer, not the user. DoorDash now takes orders by text, so your interface is a chat thread someone else designed.
• Agents need a kit, not a chatbot wrapper. Chesky told Altman an app store without an SDK fails, and ChatGPT's first one did.
• A paid agent and a free agent serve different masters. OpenAI's $100 Dot works for you; Meta's free Muse will sell you jackets.

The front door moves to someone else's house

DoorDash just launched an AI agent you text to order food over Apple Messages. Say "order my usual" and it knows your Friday night cart. No app open, no menu scroll. The restaurant's brand, the DoorDash layout, the upsell banners, all of it happens somewhere you do not control.

This is the shift under all the noise this week. Your product used to be the screen people looked at. Now it can be a backend that an agent queries on the customer's behalf. Ben Thompson's read is blunt: agents are the ultimate Aggregators, and they turn apps into "a means, not an ends." If that holds, the agent owns the customer relationship and you supply the goods.

Why a chat window is not the answer

Here is the useful pushback. Brian Chesky, who runs Airbnb, does not think a chatbot is the final form. He told TechCrunch that chatbots are bad for shopping because they hand you two options at a time and make you take turns to get anywhere. People enjoy browsing stays. You do not want to strip out the dreaming.

So Airbnb shipped AI search, not a bot. Type "baby" and it surfaces cribs and playgrounds as live filters. Chesky's line on the hard part: "Anyone can vibe-code a search function, but to do something that doesn't kill conversion rate, that's the hard part." The lesson for your team is that an agent-friendly product still needs rich controls, maps, comparison, messaging, the stuff a flat chat thread cannot do.

The handoff problem nobody solved yet

Chesky's sharper point is structural. He says the agents trying to be your one "quarterback" are building on nothing. He told Altman that ChatGPT's first app store flopped because it had no real SDK or operating system under it. His test of Airbnb inside agents like Muse and Instinct: "it doesn't work very well."

The gap is the handoff. If an agent books your stay but the user wants to message the host, compare options, or verify ID, either the agent does a clean handoff back to you or it fails. Chesky wants apps to become agents that talk to each other over MCP, the standard that lets agents connect to software. Until that plumbing exists, you get "apps that have agents inside them" instead of the full shift. That is the opening for your team to own the controls before someone else owns the thread.

Who the agent actually works for

Watch who pays. OpenAI's new Dot is gated to $100-per-month Pro accounts, and The Verge's reviewer found it felt like "workplace software that happens to be able to order you a burrito." Because you pay upfront, it is not under pressure to nag you. Meta's Muse is free, which means ads are coming. The Verge team watched Muse keep pushing a jacket they had looked at once.

That split matters for your brand. A free agent's incentive is to monetize the surface, so your product shows up next to whatever pays. And trust is still shaky: Dot got stuck on security checks, asked the reviewer to hold a button for it, and wanted a checking account number the user would not hand over. The intermediary is not neutral, and it is not yet reliable.

The funeral was a pitch, but the bet is real

Photon raised $4.5 million after holding an actual funeral, coffin and all, for the mobile app. Co-founder Daniel Tian says the end of the app era is "inevitable," and his thesis is simple: people do not want to install your app, they want an agent inside iMessage or WhatsApp they already use. The company signed 40,000 developers and grew revenue 10x in four months.

Tian's reach for the future is agent-to-agent-to-person: your travel agent pulls in a flight-booking agent, which pulls in a hotel one. Treat the funeral as marketing, but take the bet seriously. The distribution problem Photon solves, nobody discovering your app, is the same reason a leader should care which agents can reach your customer and on what terms.

Three questions for your team

• If a customer reaches us through someone else's agent, what do we control and what do we lose? Map the handoff points now, the way Chesky maps messaging, compare, and verify.
• Do we build agent access as a real SDK and set of UI controls, or a chat wrapper that kills conversion? Pick before a partner picks for you.
• Which agents can we afford to be intermediated by, and which turn us into a commodity next to a paid ad? Decide who the agent works for before it is handling our orders.