Instinct quadruples its valuation to $10 billion while still invite-only

Someone who has forgotten about a subscription types a few words to Instinct on a PC, or says them into a phone. The agent then cancels it on its own ^[5]. It works on a virtual phone and computer of its own, loading web pages and apps and using them directly ^[6]. The company says Instinct can "use devices the same way humans do" ^[7].

"We're building Instinct to be the best personal agent that can handle the deeply personal nuances of everyday life," founder Noah Shinn said in a statement ^[15]. In practice, Instinct operates products that other companies already built. With this design, an order for weekly groceries goes through a grocer's site or app, which the agent loads and clicks through ^[5][6].

This design bears on the argument that do-it-for-you agents threaten apps built around a single task. As published, the single-task app stays in the loop as the thing the agent operates. The person no longer looks at the screen before confirming. According to SiliconANGLE, Instinct is one of roughly a dozen prominent agent products competing to be a persistent assistant ^[14]. Its closest startup rivals are Wajo, Ollie and Poke, and the wider field includes agentic versions of ChatGPT, Grok and Siri ^[14].

Some are questioning how much personal information they have to hand over to get these features, and TechCrunch wrote that Instinct's first privacy policy "was particularly worrisome due to its overreach" before the company updated it ^[12]. Investors added $7.5 billion of valuation in a month ^[2]. Instinct has not shared user numbers or any growth metrics ^[11]. It has no mobile app and talks to users by SMS ^[10].

If I ran a single-task app, I would not rebuild around agents because of this round. The usage figures that would justify a rebuild have not been published. I would check whether the tasks a user can state in a few words still complete cleanly when software drives the screen, because that is where a user who delegates would show up first. The tradeoff sits in the same spot. If an agent can finish your flow easily, it also skips whatever you show people before they confirm.

The 2x2 I'd use has two axes. The first is whether a user can state your core task in a few words, the way Instinct describes canceling forgotten subscriptions or ordering weekly groceries ^[5]. The second is whether your revenue comes from the completed transaction or from what the user sees on your screen before completing it. A few-word task with transaction revenue turns agent traffic into sales, and the job is making sure those sales go through. A few-word task with screen revenue is the most exposed box, because the agent finishes the task without looking at what you sell around it. Tasks that involve choosing, like planning a cross-country road trip ^[5], keep the user looking at options whichever way the revenue comes in. I'd expect exposure there to be lower.