Instinct's text-message agent keeps pace with Muse and Dots on household chores

The first thing Instinct did for The Verge's reviewer was guess wrong. A Cincinnati area code was enough for it to send a link to the Enquirer's website and flight results between CVG and SFO, to someone who no longer lives in Cincinnati ^[4]. After a couple of weeks of heavier tasks, the review says, the agent has impressed ^[5].

The user here is specific. The reviewer's household has a preschool-age kid and two adults with full-time jobs, and the chores that pile up are swim lesson sign-ups and appointment confirmations ^[2]. Each of the three agents chats like a chatbot but also runs on a virtual machine and can plug into personal email and calendar ^[14].

Noah Shinn, an early employee of Sierra before he founded Instinct ^[11], pitches focus. "AI assistants are all we know," he told The Verge over email ^[9]. "We have been solely focused on building a personal assistant for everyday life that just gets it," he said ^[10]. On the work itself, the review has the three agents close ^[1]. The reviewer put the limit this way: "It's as effective as I think we can reasonably expect an AI agent to be right now" ^[6].

Muse has a cuddly bear mascot and Dots has anthropomorphized blobs, and the reviewer took both as obvious attempts to reach a wider consumer audience ^[7]. With Instinct, users send a text and get a short reply, plus a "secure vault" link when a site needs a login ^[16]. The thread follows them onto CarPlay or a watch ^[17]. The reviewer called that "either ideal or creepy" and settled on a little of both ^[17].

Shinn's case against the bigger companies rests on that channel. "While AI platforms excel at answering questions and prompts, Instinct seamlessly integrates into people's lives to turn conversation into action," he said ^[19]. "Just like you would text or email your personal assistant, you can do the same with Instinct," Shinn said ^[20]. The review does not test whether Meta or OpenAI could move their agents into a text thread, and it reports no plan by either to try. It does test chores, and on chores Instinct is level ^[1]. I think the lead Instinct can defend is the one that kind of testing measures: errands completed without the user stepping in.

Investors have already priced a lead. The valuation came within about two months of the August launch, after Muse was already out ^[12][21].

For anyone choosing among these agents, two questions sort them. One is whether a chore finished with no correction from the user. The other is whether the agent lives in a channel the user already checks. An agent that finishes and lives in the texts is what Shinn is selling. Finishing inside a separate app costs the user a tap, and the mascot is decoration. Failing inside the texts is the worst outcome, because a wrong guess like the Cincinnati flight search lands where the user reads everything else ^[4]. A failure inside an app is easy to ignore and easy to drop. For a household that already runs on text threads, I'd start with Instinct and accept that its mistakes arrive in the same place as the family's messages. To compare the three, count completions without follow-up, on the same few chores, over two weeks.