The assistant is becoming the errand layer

A $300 to $400 AI smart speaker is a strange flagship object in 2026. It sounds less like a gadget launch than a bet that the next consumer interface will be judged by how much it can do for you before you open another app.

That is the interesting part of OpenAI’s reported hardware push. TechCrunch reported that OpenAI’s new AI smart speaker will reportedly sell for between $300 and $400. The lazy read is that OpenAI wants to compete in smart speakers. I think the sharper read is that OpenAI wants a permanent place in daily routines because assistants are shifting from answer boxes into the errand layer.

Search gave us information. Apps gave us buttons. The next fight is over who gets trusted to press the buttons for us.

The assistant as transaction machine

Google Maps is the clearest example because the errands are literal. TechCrunch reported that Google Maps is adding agentic features, including food ordering and hotel bookings. That is a big shift in what Maps is for.

Maps used to answer questions like “where is the restaurant?” Now it is moving towards “sort dinner out”. Those are different products. The first is search plus directions. The second touches preference, availability, timing, and money. The value moves from finding options to collapsing the gap between intent and completion.

Economists have a useful phrase for this: transaction costs. The web lowered the cost of finding things. Mobile lowered the cost of accessing them anywhere. AI agents are trying to lower the cost of coordinating life’s small annoyances. Dinner, travel, tickets, scheduling and messages are not intellectually hard in isolation. They are tiring because they require context switching and judgement.

That is why TechCrunch’s report that ChatGPT is bringing unlimited text chats to free users matters beyond the pricing change. Basic chat starts to look less like a premium feature and more like the habit layer. You do not charge heavily for the doorway if the money is in what happens after people walk through it.

The smart speaker fits that logic. It is not aimed at people who want another dashboard. It is aimed at moments when typing is annoying: cooking, packing, leaving the house, or remembering that something still needs booking. The question is whether people will pay premium hardware prices for a better delegation surface.

Permission beats attention

The most revealing story may be dating, because dating is where delegation gets uncomfortable fast. TechCrunch reported that Gen Z dating apps like Ditto are ditching swiping in favour of AI matchmaking. That is not productivity software with a softer interface. It is AI stepping into taste, vulnerability, and social risk.

If users let an assistant filter potential partners, recommending a hotel starts to look tame.

For builders, this is the point: the next product question is not “can the model answer?” It is “which errand can this product safely own?” Food ordering has a visible failure mode. A bad hotel booking is more expensive. A bad match is personal. Each category needs different levels of confirmation, memory, reversibility, and user control.

OpenAI’s own framing, “From asking to doing”, captures the direction of travel. The old app economy fought for attention. The errand layer fights for permission. The winners will not be the assistants that talk the most; they will be the ones users trust to make small decisions without making life weird.

---