AI assistant Instinct, founded by a 23-year-old college dropout, saw its valuation rise from $2.5 billion to $10 billion in two months

[[**An AI assistant called Instinct saw its valuation soar from $2.5 billion to $10 billion in two months. It was founded by 23-year-old dropout Noah Shinn, but its excessive demands for user permissions have sparked serious privacy controversy.**

**Key Points**

**1. Valuation quadruples in two months as capital bets wildly**

In August 2026, Instinct completed a $250 million Series B round (at a $2.5 billion valuation). Less than two months later, its Series C valuation reached $10 billion, led by Sequoia, Benchmark and others. With only 14 full-time employees, it has become a new wealth-creation legend of the post-2000 generation.

**2. Product core: executing real-life decisions on users' behalf**

Instinct can be granted access to email, calendars, credit cards and more, and through text-message or phone-call instructions it completes tasks on users' behalf such as booking tickets, shopping, canceling subscriptions and handling insurance negotiations. Users spend an average of more than $1,300 per month through it, with high-ticket scenarios accounting for a significant share.

**3. Founder's academic and industry background**

23-year-old Noah Shinn dropped out of his undergraduate program. He published the Agent milestone paper Reflexion at NeurIPS as first author, then joined the AI customer-service company Sierra as a research scientist, and founded Instinct after building up enterprise-grade agent experience.

**4. Permission disputes and security risks stand out**

In early 2026, an AI assistant called Instinct opened an invite-only private Beta test. No one expected that this summer it would go viral in Silicon Valley venture circles. A large number of investors and entrepreneurs spontaneously posted about their orders, and beta users quickly surpassed 100,000.

The product has still not been publicly launched and has disclosed no revenue, yet capital has already rushed in.

In August this year, Instinct had just completed a $250 million Series B round at a post-money valuation of $2.5 billion (about 16.7 billion yuan), co-led by Index Ventures and Benchmark, with existing shareholders such as Conviction and Greenoaks participating.

At the time, Instinct was already mired in huge controversy over its reliance on excessive authorization. But that could not stop capital's enthusiasm for betting on the next generation of AI applications.

Just less than two months later, Instinct announced the completion of its Series C round at a $10 billion valuation (about 67 billion yuan), led by Sequoia Capital, Benchmark and Coatue.

In just two months, the valuation quadrupled. With only 14 full-time employees, here is yet another wealth-creation legend of the post-2000 generation in the AI wave.

In early September, Meta also announced the release of a new agent called Muse, and within less than two weeks of launch, Muse topped the free app charts on the US App Store and Google Play. Because of this app, Meta's market value increased by hundreds of billions of dollars.

By comparison, even with its valuation quadrupling in two months, Instinct looks quite "cheap." With Muse and Instinct as a foundation, it seems the fire of AI applications is about to burn again.

Instinct was launched by Spear Street Technology, and its founder is Noah Shinn, who is only 23.

According to the company, Instinct can "efficiently manage users' lives": users grant it permissions for email, calendar, text messages, location and more, then send instructions via iMessage, WhatsApp or a direct phone call, and leave the rest to it.

Specifically, it can buy groceries, book concert tickets, reserve restaurants, handle insurance negotiations, cancel subscriptions and view apartments; it can even independently create a separate email account to communicate externally on your behalf.

"I'm very excited about everything early users have done with Instinct," Shinn wrote on Twitter. "They used Instinct to plan cross-country road trips, buy weekly groceries and concert tickets, and cancel subscriptions worth hundreds of dollars. Some even used Instinct to plan their weddings."

It seems AI assistants are finally ready to help you manage your life. You just need to trust them and hand over your email login information, contacts and credit card information.

At the end of August, Instinct had integrated Stripe's Link payment. Users can set amount limits and authorize one-time virtual cards so that Instinct can complete online payments on their behalf. According to Shinn on social media, users who have already shopped through Instinct spend an average of more than $1,300 per person per month through it, covering scenarios such as travel, daily necessities, yoga classes, doctor visits and car repairs.

Behind this figure are two intriguing details: first, high-ticket scenarios account for no small share, showing that what users entrust is not just trivial errands but life decisions that require real money; second, Instinct's breakthrough is not in the "agentic e-commerce" narrative that has been hyped for two years, but precisely in transactions such as bargaining, appointments, cancellations and negotiations that cannot be classified by the traditional e-commerce funnel. This is the true landing point of Agent value.

A 23-year-old dropout starts a business

Shinn did not appear out of nowhere.

While studying as an undergraduate at Northeastern University in the United States, he worked on type prediction and code generation in the university's programming languages research group, and also participated in interdisciplinary work on machine learning and programming languages at MIT.

What truly made his name in academic circles was a paper. In 2023, as first author, he published Reflexion at NeurIPS, proposing that AI agents, after a task fails, review their errors in natural language, store them in memory, and reuse that experience in subsequent tasks. The paper is regarded as a landmark work in the AI Agent field, achieving a 91% pass rate on the HumanEval code benchmark, surpassing GPT-4 of the same period.

That year, he chose to drop out, and in May joined Sierra, an AI customer service company co-founded by Bret Taylor (former Salesforce co-CEO and OpenAI board member) and Clay Bavor (former Google executive), as a research scientist. During his two years at Sierra, he participated in developing the company's core enterprise-grade agents. That experience also made him realize: enterprise-side AI Agents have already landed, but personal autonomous AI assistants for ordinary people remain a huge blank space.

In the second half of 2025, Shinn founded Spear Street Technology and launched Instinct, a "personal agent with no new interface."

Instinct stands on a lineage six decades long.

In 1966, MIT's ELIZA used keyword substitution to simulate conversations with a psychotherapist, marking the first time humans poured their hearts out to a machine; in 1995, ALICE became the first widely commercialized chatbot; over the following two decades, SmarterChild and the Xiao i robots flooded into MSN and QQ, yet remained trapped in the shell of "question-answering machines."

In 2011, Siri launched with the iPhone 4S, opening the era of voice assistants, and Google Now, Alexa and Cortana followed in droves. In 2014, Microsoft launched Xiaoice, which did not emphasize task execution but focused on emotional companionship. But there was still a huge gap between "understanding" and "getting it done": at the time, speech recognition error rates were as high as 15%-20%, intent understanding could only cover a small number of vertical scenarios, task completion depended on third-party apps opening their interfaces, and cross-app orchestration was nearly impossible. Assistants generally degenerated into toys for "asking about the weather" and "telling jokes."

Around 2017, Tmall Genie, Xiaodu and Xiaoai Tongxue spread smart speakers across China through subsidy wars, but then fell into homogenization and price wars, and the entrance narrative was hard to monetize. The industry gradually reached a consensus: the ceiling of an assistant's capabilities is determined by the underlying language understanding technology, not by the form of the entrance.

In November 2022, ChatGPT launched, and for the first time conversational assistants had open-domain knowledge Q&A, writing and reasoning capabilities. In 2023, multimodal large models such as GPT-4, Claude and Gemini further connected text, images and audio. Then the Agent era arrived. With large models as the brain, they autonomously plan, invoke browsers, code and APIs, and complete multi-step real tasks.

From OpenAI's Operator to Coding Agent, and then to personal Agents like Instinct that have no new interface, the form of AI assistants has evolved once again, and this time, it can truly get things done for you.

Building a "general-purpose AI assistant" to become everyone's life assistant is the vision of countless companies and entrepreneurs. Although OpenClaw has already become popular in tech circles earlier this year, products like Instinct are just beginning to emerge and may be poised to take up that key that has waited sixty years.

Permissions, privacy and loss of control

The excitement Instinct has stirred up arrived almost simultaneously with its controversy.

The focus of the controversy is precisely its biggest selling point: permissions. Want it to book a flight for you? It needs your credit card and airline account. Want it to organize your inbox? It needs access to your email. The utility of an AI agent depends on the level of permissions you grant it, and the permissions Instinct asks for are dauntingly broad.

AI agents need built-in safety safeguards. Without these protections, they will run out of control, because they are specifically designed to complete tasks, no matter what means are used to complete them.

This is not alarmism. According to The Information, in July this year, an internal OpenAI model escaped its development sandbox and breached Hugging Face's internal systems; Meta's Muse agent exhibited various bad behaviors during testing: sending unauthorized emails and even attempting to sabotage a competitor app the user was developing.

Meta, Google, OpenAI and Anthropic are all setting up safeguards for agents: limiting the apps, accounts and data they can access, requiring users to approve high-risk operations such as purchases and sending emails item by item, and scanning web pages, emails and files for hidden instructions.

But the risk has not disappeared. Sam Altman recently told Fortune: "We haven't solved alignment. Our research isn't done. I don't believe any lab has solved alignment."

For sixty years, the product form of each generation of AI assistants has been determined by the technological ceiling of the time. The failure of each generation of products almost stems from the same misalignment—capabilities fall short of being irreplaceable, yet the business model demands that they be indispensable. Instinct is betting on large models crossing this threshold for the first time. The capital market voted yes with a valuation of 60 billion, and the real test has only just begun. AI industry signal channel: Frontier Tech. This content is published with the author's authorization; the views represent only the author and do not represent Huxiu's position. If you have objections or complaints about this manuscript, please contact tougao@huxiu.com. Everyone who is changing and wants to change the world is on the Huxiu APP[ ]]]