Cover image | Produced by Pencil News

In mid-September, personal AI assistant Instinct was reported to be in talks for a funding round of approximately $1 billion, targeting a valuation of around $10 billion (approximately RMB 67 billion). Sequoia Capital and Benchmark are participating in the discussions.

On August 26, it had just completed a $250 million Series B round at a $2.5 billion valuation. The valuation quadrupled in a month.

Why is Instinct worth so much? Because it's frighteningly powerful.

Simply tell Instinct: 'Take my parents on a trip during National Day for a few days with a budget of $20,000, and don't make it too tiring.' The rest—searching, price comparison, ticket purchasing, communication, reservations, and even payments—are all handled by it.

Instinct now has 100,000 users. At a $10 billion valuation, each user is worth $100,000.

Li Di, founder of Nextie and 'father of Xiaoice,' told Pencil News that capital is already betting on the 'first right of control over personal intentions.'

- 01 - The Strongest Gen Z Dropout

Instinct's founder, Noah Shinn, is 23 years old. He dropped out of Northeastern University in 2023 after conducting machine learning and programming language research at MIT and Northeastern.

What established his reputation in the AI circle was a 2023 NeurIPS paper titled Reflexion, of which he was the first author. The paper proposed: after an agent fails a task, it summarizes the error into text, stores it in memory, and retrieves this memory during the next execution—one of the earliest blueprints for 'enabling agents to learn from trial and error.'

Later, he participated in developing τ-bench, a benchmark specifically designed to test whether a model can converse with users, invoke tools, and adhere to business rules.

Both projects addressed the same question: how to enable a software agent to independently complete a multi-step task like a human.

After leaving school, he joined Sierra, an enterprise customer service agent company, as one of the earliest research scientists. Founded by Bret Taylor, former chairman of OpenAI and former co-CEO of Salesforce, Sierra develops AI to handle customer service processes for enterprises. After working there for a while, Shinn identified a gap in the market: while agents were already working in enterprises, they were absent from ordinary people's lives.

In 2025, he founded Spear Street Technology in San Francisco, with Instinct as its product.

Shinn's requirement for Instinct was: don't tell users what to do—just do it.

In February of this year, Instinct opened for invitations. By early August, users began spontaneously sharing their experiences on social media, triggering a Word of mouth (word-of-mouth) explosion: Silicon Valley investors and tech practitioners started intensively sharing their usage experiences.

Some used it to negotiate with insurance customer service, others had it buy groceries weekly, purchase concert tickets, find apartments, buy home insurance, or customize wedding supplies. It could even make phone calls to reserve restaurants that couldn't be booked online and queue for its owner at the dentist. Some even used it to buy a house.

Instinct is powerful but also highly controversial. To truly act on behalf of users, an AI agent must obtain permissions rarely granted simultaneously to software in the past: it needs access to your schedule, emails, contacts, multiple accounts, and even the ability to send messages and place orders on your behalf. For example, investor Jason Yeh merely asked Instinct to 'find' a restaurant with available seats, but it directly booked a seat with a $200 cancellation fee for him.

Despite the potential for mishaps, Silicon Valley remains extremely excited, with Instinct's valuation jumping almost monthly.

Around April, Instinct began its seed round, raising $17 million at a $150 million valuation. In early August, it completed a $75 million Series A round at a $500 million valuation. By late August, its Series B round raised $250 million (led by Benchmark), reaching a $2.5 billion valuation—a fivefold increase.

Now, it's in talks for a $10 billion valuation with a $1 billion funding round.

Investor Ben South said, 'If you believe personal agents are the ultimate battleground for consumer software, then Instinct is currently the only ticket with ultra-high VC return potential.'

- 02 - Reasons for the Valuation Surge

Li Di believes: whether AI assistants truly cross the commercialization threshold depends on when users dare to entrust them with the three words, 'You decide.'""You decide' is precisely why Instinct's valuation has surged.

'Help me search for the highest-rated restaurant nearby' is essentially search; 'Help me book a certain restaurant at 7 PM' is execution. But when a user says, 'Dinner tonight with a long-lost friend—find a quiet place. You decide,' the AI begins making judgments on behalf of the user.

Price, distance, taste, and past pitfalls must all be factored into the same decision. Li Di calls this the delegation of 'cognitive discretion.'""Instinct has already started handling money. In late August, it integrated with Stripe's Link. Users can set spending limits and authorize one-time payment cards, allowing Instinct to complete online payments on their behalf without directly sharing their real bank card information with merchants. Shinn revealed that users who have made purchases through Instinct spend an average of over $1,300 monthly via the agent, covering scenarios like international travel, groceries, yoga classes, and car repairs.

$1,300 represents transaction volume, not Instinct's revenue. But it's far more significant than 'AI can book restaurants': someone truly dares to hand their wallet to AI. At this rate, an active transaction user would spend over $15,000 annually via the agent. Even if future revenue is based on a mere 1% improvement in transaction efficiency, that corresponds to approximately $156 annually per user (1% is a projection, not actual commission).""While ChatGPT and Claude focus on monthly membership fees of tens or hundreds of dollars, an agent capable of booking flights, purchasing appliances, and renewing insurance taps into a consumer pool of thousands to tens of thousands of dollars per person annually. Li Di predicts two future revenue streams: loyalty premiums paid by users for 'it truly understands me,' and transaction fees shared by merchants from compressed customer acquisition chains—when users stop searching and simply say, 'Help me choose one,' merchants will still pay for conversions, just shifting from search ads to AI.

Payment giants also recognize that AI managing money is the future. In June, Visa partnered with OpenAI to integrate its payment network and authorization system into agent shopping. In September, Mastercard launched Agent Connect, allowing merchants to directly provide products, prices, and inventory to AI.

- 03 - A Fragile Valuation

Whether Instinct's valuation holds depends on three factors.

First is memory. Errand-running capabilities like replying to emails or booking restaurants will soon become standard for agents. What truly differentiates them is long-term personalized memory: the agent must gradually understand a user's decision-making criteria. In September, Meta launched Muse, and Perplexity launched Brain—both vying for this space. After three years with an assistant who knows your boss's style, your child's schedule, and your travel preferences, switching feels like replacing a trusted assistant. That's stickiness.

Second is computational costs. Chatting ends once; an agent must search the web, compare results, invoke tools, and retry if it fails—each step burning tokens. Gartner predicted in August that by 2028, the reasoning cost of a single agent workflow could rise to over five times current levels. After exceeding 100,000 users, Instinct has warned that its services are nearing full capacity. Whether ten additional uses per user lead to greater losses or harder-to-replicate personal memories determines whether it becomes more valuable with scale or increasingly unprofitable.

Third is trust—the easiest factor to self-destruct. Shinn said Instinct won't charge users for now and mentioned advertising. Li Di argues that a personal agent's greatest value lies in users believing it acts in their best interest. If it recommends hotels based on who pays the most for ads, that trust evaporates. Currently, Instinct's terms allow retaining user data for model training. Security testers have demonstrated that it could be tricked by phishing emails, and some users found it retained email copies even after disconnecting from Google. Shinn responded that the team has tightened autonomous operations and is addressing these issues.

Another criticism of Instinct isn't its business itself but that it, along with VCs, is manufacturing valuations.

For example, lead investors first inject most of the funds at a lower valuation (e.g., $2.5 billion), then add a smaller portion at a much higher valuation (e.g., $10 billion). Other eager investors follow at the latter price. The highest figure is often the one publicly announced.

If Instinct's new round truly closes near a $10 billion valuation, Benchmark's shares, bought at a $2.5 billion post-money valuation less than a month ago, would theoretically be worth nearly four times as much under the latest pricing system.

Tech Business summarized: These companies share common traits: young founders, extremely short existence, astronomically high valuations, and highly immature products. VCs no longer price products, revenue, or even teams. They price the 'speed' of valuation itself.

This model succeeds by exploiting a fear of missing out in VC circles: what if it's the next OpenAI or Anthropic?

This article does not constitute any investment advice.