Instinct founder on an app-less personal Agent: How a 14-person team supports a $10 billion valuation

[[**A one-year-old AI personal agent company with only 14 employees, Instinct, has reached a valuation of $10 billion, and its founder is a post-2000s dropout. The product has no standalone app and relies on text messages and phone calls for interaction. Its annual transaction volume has already reached the $1 billion scale, and it achieves profitability through merchant commissions rather than user payments.**

**Key Points**

**1. Product form: no app, fully interactive via text messages and phone calls**

Instinct has no standalone app or website. Users contact it via text message, phone call, and email to handle tasks such as booking flights and canceling subscriptions. The first design principle is "understandability before capability": the first two lines of information must make the matter clear, otherwise attention will be scattered.

**2. Trust building: users need 2-3 weeks before binding core permissions, and retention reaches 80% after binding**

After using it for about three weeks, about 40% of users will proactively bind a credit card. Once they bind sensitive information and experience the value, the retention rate for this group is about 80%. Data is authorized by users at their own pace and can be revoked at any time. The risk-control mechanism is independent of the main decision-making system and reviews instructions before taking action.

**3. Business model: no advertising, charging commissions to merchants who profit from transactions**

Noah explicitly rejects the advertising model, believing that if an Agent serves advertising, it will induce users to buy things they do not need. The company chooses a transaction commission model similar to Apple Pay: free for users, with about half of transactions coming from travel, and high-end hotels willing to pay 20%-30% commission for valid orders.

**4. Growth bottleneck: 10% daily growth but compute cannot keep up, about 40% of energy spent coping with it**

During the rapid growth period, the daily increase reached 10%-11%, and compute demand nearly doubled every week, but cluster delivery cycles last several months. The team built its own hybrid inference system to reduce costs. Noah estimates that in the future more than 90% of token consumption will occur in the background that users cannot see.

**5. Interaction evolution: over 90% of users use voice, and tasks extend from one-off instructions to long-term goals**

This article is from the WeChat public account: Datawhale, author: Datawhale, original title: "The hottest post-2000s founder of the year is born! An in-depth exclusive interview with the Instinct founder is here", header image from: AI-generated

One year since founding, 14 employees, valuation reaching $10 billion.

Instinct has become popular in Silicon Valley, and behind it is a post-2000s founder. Founder Noah Shinn is 23 this year. He dropped out of Northeastern University in the United States in 2023, joined the AI unicorn Sierra, and then founded Instinct.

Not long ago, Instinct announced the completion of a new round of $1 billion financing, jointly led by Sequoia Capital, Benchmark, and Coatue. A month ago, the company's valuation was still $2.5 billion, and in the blink of an eye it had quadrupled.

What is more unusual is that this product has no standalone app. Users contact it via text message, phone call, and email, asking it to book flights, cancel subscriptions, and arrange gatherings. Until now, using it still requires finding an invite code, and some people have even spent $300 on eBay to buy a spot.

Recently, Noah appeared on the podcast "Invest Like The Best" and systematically described Instinct's product, business model, and the difficulties behind its rapid growth.

1. No app, yet it still gets things done for you

Instinct has no standalone app and no website. It has its own phone number, email address, and a cloud computer. Users send it text messages, call it, and forward emails just like contacting a real human assistant; for urgent matters, it will call directly.

Why not make an app? Noah's explanation is: the product being the app is a self-imposed limitation. AI can become the experience itself; there is no need for an icon on a phone screen to be tapped before there is a sense of security.

The interface is extremely simple, but that does not mean capability is limited. The first principle he set for design is understandability before capability: users must be able to anticipate what it is doing and what will happen next. It mainly communicates with users through text messages, so the first two lines must make the matter clear, otherwise attention will be scattered.

One of the most widely circulated user stories: a couple was caught on camera on the big screen at the US Open and wanted to find that video. They sent their seat numbers to Instinct, and it worked for 18 hours, emailing the US Open, looking for recordings, asking customer service, and finally connecting with the editing staff to get the video file back.

2. One voice message, and flights and hotels are all booked

According to Noah, Instinct's current annualized transaction volume has reached the $1 billion scale, with about half coming from travel.

The experience is like this. A user sends one sentence: "I need to be in New York tonight." It first checks the schedule, then arranges the flight, hotel, and pickup car in sequence, and even remembers seat preferences and meal requirements.

Canceling subscriptions is also a typical scenario. Traditional personal finance software usually can only list bills and remind users which services can be unsubscribed. Instinct continues further: it opens the website, logs into the account, obtains verification codes from the authorized email if needed, and completes the cancellation step by step.

In the end, the user may receive only a brief report: this month it helped you save $2,000.

Some people also give it wardrobe and body measurements so it can generate images of themselves trying on clothes, then choose outfits across multiple e-commerce platforms. If they like a certain set, they say "this set," and it proceeds to place the order.

Agents can also collaborate with one another. A group of six friends each let their Instinct automatically plan a gathering once a week. They coordinate the six people's free time, pick activities based on shared interests, and sometimes even call a car to arrange pickups along the way.

From travel to shopping, these scenarios are starting to involve real spending. After users state their needs, more and more tasks that previously required opening several apps and confirming repeatedly can be handled by the same assistant.

3. After three weeks of use, 40% of users link a credit card

For a personal Agent to be useful, it needs to understand the user. The more data there is, the easier it is for it to remember preferences and the easier it is to get things done.

But Noah stresses that Instinct will not push users to grant permissions. The data is controlled by users themselves; they can authorize at their own pace and revoke it at any time.

Trust takes time. He disclosed that ordinary users typically need two to three weeks before they open up core permissions such as bank cards and email. After about three weeks of use, about 40% of users will proactively link a personal credit card and authorize it to make payments on their behalf.

Once users have linked at least one piece of sensitive information and have genuinely experienced the value, this group's retention rate can reach around 80%.

Noah divides the problem into two parts. One part is the encryption and storage of sensitive data, which is relatively mature engineering and tests whether the team can execute strictly. The other part involves the Agent's autonomous actions: it can make purchases, pay, and call external tools, and any single mistake could affect real life.

To this end, Instinct has set up a risk-control mechanism independent of the main decision-making system, which reviews instructions before actions are executed, blocks them when necessary, or asks the user to confirm again.

Permission boundaries also apply to collaboration between Agents. A married couple can share more information, while colleagues can only see authorized schedules and project emails. Unauthorized access will be blocked.

In Noah's vision, connections between Agents should carry over the degree of trust between people in the real world.

4. Users pay nothing; merchants pay

Noah has a clear stance on the business model: no advertising.

His judgment is that future Agents may surpass ordinary users in emotional intelligence and persuasiveness. If such a system also had to serve advertising revenue, it would be capable of inducing users to buy things they do not need.

The host brought up the old Silicon Valley saying, "If you're not paying for the product, you are the product," and Noah agreed.

Instinct's chosen direction is transaction commissions. He compares it to Apple Pay and American Express: free for users, with the company charging merchants who benefit from transactions.

He also compared the fee levels of several platforms. By his account, Shopify's overall monetization rate for merchants is about 2.5% to 3%, Amazon's exceeds 10%, and Apple's App Store can reach 30%.

Instinct does not seek such high commissions. However, about half of its transactions come from business travel. In Noah's estimate, high-end hotels are willing to pay 20% to 30% commission for valid bookings, and the existing transaction model can become its source of revenue.

He also discussed the Agent's impact on traditional platforms. Platforms that rely on user dwell time and advertising to make money will face disruption; companies that make money by fulfilling goods and services may receive more orders.

This is his judgment from Instinct's standpoint. For traditional companies, he suggests first opening a small number of scenarios for testing, to observe whether Agents can actually bring more transactions and higher long-term user value.

5. Growing 10% a day, compute can't keep up

Instinct's growth initially came from 200 friends and family.

Each user has only 5 invitation slots. Existing users give their slots to people around them, and the product gradually spreads. According to figures disclosed by Noah, the daily growth rate during the rapid-growth phase reached 10%, even 11%, with zero marketing budget.

The reason is straightforward: compute can't keep up.

A personal Agent needs to work in the background around the clock. Even when the user has not sent a message, it may be checking email, verifying schedules, and handling unfinished tasks. As users grow by about 10% per day, compute demand nearly doubles every week.

Procurement, however, is not that fast. Delivery cycles for compute clusters are as long as several months, so you cannot wait until you run out to place an order.

Buying twice current demand in advance may be used up in about a week; stocking ten times the demand requires a large upfront investment. A wrong judgment may leave expensive idle compute, or cause the service to stall due to insufficient resources.

Noah says he spends about 40% of his energy thinking about how compute can keep up with growth.

The team is also driving down the cost of each task. Instinct has built its own hybrid inference system so that different tasks use different processing methods.

Tasks that need an immediate reply to the user prioritize speed; background information checks and data processing can run in batches and asynchronously. Through methods such as model distillation, computational efficiency is increased several times over.

This is also an important condition for it to remain free.

Noah also offered a longer-term judgment. Chatbots mainly run when users send messages, but personal Agents need to continuously handle background affairs. In the future, more than 90% of Token consumption may occur where users cannot see it.

6. From getting one thing done to managing a four-month goal

Instinct's entry point is still changing.

According to Noah, more than 90% of users now interact with it mainly by voice. Earphones and the side buttons on phones can all become ways to reach the assistant. When complex information needs to be presented, it temporarily generates a web page or mini-app for users to view and operate.

The tasks users assign are also starting to get longer.

Some want it to record the day's exercise, while others hand it their weight-loss and running goals for the next four months, letting it continuously coordinate training, diet, and schedules.

In business scenarios, some small businesses have already handed back-office operations to Instinct, letting it handle procurement, reconciliation, and other matters based on inventory and funding constraints.

From completing a single instruction to continuously following up on a goal, the Agent needs to do more, and it also tests whether users are willing to authorize it over the long term.

Speaking of Muse in the same track, Noah acknowledges that it is a good product but believes the two have chosen different directions. Muse focuses more on a feature-rich native App, while Instinct hopes to let more and more things be done directly in everyday communication.

Final thoughts

At the end of the interview, the host asked about the name Instinct.

Noah hopes users will see it as a partner they can entrust things to. It does not need to deliberately imitate a person's personality, nor does it need to create emotional dependence; it should earn users' trust by getting things done day after day.

After users hand over their credit cards, every payment and every booking must be worthy of that trust.

This article comes from the WeChat public account: Datawhale, author: Datawhale AI Venture Capital Daily Channel: Frontier Technology. 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[]