**没有App的百亿估值AI助理Instinct​ | AI助理 | 个人智能体 | Noah Shinn | 智能代理 | 红杉资本 | AI信任机制 | 邀请制增长 | Reflexion论文 — Transcript & Summary | SozAI**
Source: https://sozai.app/transcript/ai-assistant-instinct-billion-valuation/

Explore Instinct, a $10B AI assistant with no app, offering autonomous personal and social agent capabilities, backed by top VCs.

## Key Takeaways

- Instinct redefines AI assistants by eliminating the need for a traditional app interface.
- Autonomous agents can handle complex personal and social tasks with high privacy and trust.
- Invite-only growth and no advertising monetization reflect a user-first, sustainable business model.
- Trusted Personal Network enables secure, tiered access to personal data, mirroring real-world social dynamics.
- The future of internet commerce may be transformed by AI agents enabling bilateral intent matching.

## What the video covers

- Instinct is a revolutionary AI assistant valued at over $10 billion, founded by Noah Shinn, who dropped out of Northeastern University.
- The product operates without a standalone app, using phone, text, email, and social intelligence to assist users.
- Instinct can autonomously manage tasks such as digital wardrobe planning, shopping, subscription cancellations, and scheduling.
- It features a Trusted Personal Network enabling secure, tiered trust-based calendar and data sharing among users.
- The AI respects social boundaries, proactively reaching out only for urgent matters and maintaining user privacy.
- Instinct aims to rewrite internet commerce by enabling bilateral intent matching, improving booking and transaction efficiency.
- The company has a small, highly technical team with no sales or marketing staff, relying on invite-only growth.
- Revenue is primarily generated through transaction commissions, especially in travel, avoiding advertising-based monetization.
- Instinct’s compute demand grows exponentially, reflecting its increasing user capabilities and real-time processing ambitions.
- The product philosophy emphasizes minimalism, privacy, no ads, and no emotional manipulation, focusing on practical utility.

## Chapters

1. 00:00 Introduction to Instinct and its valuation
2. 01:15 Founding story and team background
3. 02:15 Product overview and unique interface
4. 03:14 User cases: wardrobe and subscription management
5. 04:17 Instinct-to-Instinct network and trusted personal network
6. 05:29 Impact on internet and commerce
7. 06:38 User trust and privacy considerations
8. 07:50 Business model and monetization strategy
9. 10:14 Technical challenges and future capabilities
10. 14:56 Philosophy and product vision

Answers

## Questions about this video

What makes Instinct different from other AI assistants?

Instinct operates without a standalone app, using phone calls, texts, and emails to assist users autonomously. It respects social boundaries and has a virtual computer behind it capable of performing any internet task.

How does Instinct ensure user privacy and trust?

Instinct uses a tiered Trusted Personal Network where users control access levels to their data. It monitors for trust abuses and mirrors real-world social constraints to maintain privacy and trust.

What are some practical use cases of Instinct?

Users utilize Instinct for digital wardrobe management with photorealistic try-ons, automatic subscription cancellations, proactive scheduling, and collaborative social planning with friends.

## Full Transcript — Download SRT & Markdown

00:00

Speaker A

Hello everyone, this is Best Partner. I'm Da Fei. On eBay, an invite code for a private beta has been bid up to $300. It’s not for limited-edition sneakers or concert tickets, but for an AI assistant invitation. The company is

00:13

Speaker A

called Instinct. Founded just a year ago, its valuation has surpassed $10 billion, with Sequoia, Benchmark, and a16z co-leading a recent $1 billion funding round. What’s even crazier is that it has no app, no app store icon,

00:24

Speaker A

and they’ve even claimed they never intend to build a standalone app. Today's content comes from the top Silicon Valley business podcast "Invest Like the Best," where Patrick O'Shaughnessy had a 90-minute deep dive with Instinct founder Noah, marking his

00:38

Speaker A

first systematic public discussion about his company. Let’s start with Noah. He is 23 years old, studied chemistry and computer science at Northeastern University, and dropped out in 2023 to start his business. As an author, he published the paper

00:51

Speaker A

"Reflexion," proposing a mechanism for large models to self-correct through a cycle of reflection and memory improvement. This paper has been cited hundreds of times by Google and DeepMind, and was featured at NeurIPS 2023. Afterward, he joined the Silicon

01:03

Speaker A

Valley AI agent unicorn, Sierra, as one of its earliest members to participate in the implementation of enterprise-level AI agents. After gaining experience at Sierra, he spotted a gap in the mass market for fully autonomous agents, so he founded

01:15

Speaker A

Spear Street Technology in 2023, and launched the product called Instinct. The core team has only 14 people, all with technical backgrounds, with no dedicated sales or marketing staff.

01:25

Speaker A

Most core members followed him out of the Sierra system. Alright, let’s talk about the product. What exactly is Instinct? Noah’s answer is very direct: it’s a human assistant; no need for fancy buzzwords to wrap it.

01:36

Speaker A

But its form is unlike any other AI product on the market. It doesn't even have a standalone app. It has its own phone number and computer. You can text or call it, and it can proactively call you. It has its own email address and

01:47

Speaker A

possesses social intelligence and awareness, collaborating with you just like you would with any other person.

01:52

Speaker A

Noah has been using it for months, and Instinct has only called him about three times. It only proactively reaches out when there is a very urgent and non-negotiable deadline. For example, it might say, "I don't want to disturb you, but this document is due

02:04

Speaker A

by 3:00 PM. It's already 2:55 PM, could you please take a look at it?" "The email is in your inbox, I can send another to bump it to the top." This sense of social boundaries is a core part of the product design. Noah

02:15

Speaker A

specifically emphasizes not to mistake a minimalist interface for limited capability; it has an entire virtual computer behind it. Theoretically, anything you can do on the internet, it can do too. What are users actually doing with it? Noah shared a few

02:27

Speaker A

real-life cases that left the deepest impression on him. A group of users has mastered digital wardrobe management.

02:32

Speaker A

They walk into their closets, use their phones to scan every shirt, pair of pants, socks, and shoes, while also scanning their face shape, body, and proportions, then have Instinct plan their outfits for the whole week. The AI's feedback isn't just a cold list of

02:44

Speaker A

items; it generates a photorealistic full-body image of the user wearing those clothes, letting you see how you'd look heading out in that outfit today. This capability also extends to shopping. Users can simply say "shop the web for clothes," and Instinct will

02:57

Speaker A

search thousands of outfit combinations from major e-commerce platforms, render the try-on effect for the user, and place the order for delivery once satisfied. Some have even set up daily tasks, having it push three brand-new full-body outfit suggestions every

03:08

Speaker A

morning that can be purchased with a single click. Another goal-oriented use case really illustrates the point.

03:14

Speaker A

After users link their bank accounts to Instinct, it scans all transaction records and subscription bills, proactively asking, "Are you still actually using this service?" Many users realize only then that they've been getting charged, without even knowing they had subscribed. And

03:26

Speaker A

Instinct doesn't just suggest you cancel like traditional finance apps; it runs the entire process fully automatically from end to end. It opens the website, logs in, and if an email verification code is needed, it uses its email access permissions to

03:37

Speaker A

retrieve it. It clicks all the way to the deepest cancellation page to complete the unsubscription, finally sending you a brief summary: "Done.

03:43

Speaker A

You've saved $2,000 this month." Any product relying on user inertia or forgetfulness will disappear in the face of such a full agent. After describing these user scenarios, Noah also mentioned a feature that just launched ten days ago: the

03:55

Speaker A

Instinct-to-Instinct network. Its starting point isn't to create a new feature, but to solve a real pain point. Professionals spend all day in meetings, and scheduling one requires going back and forth in messages for hours: "Is Wednesday afternoon okay?"

04:05

Speaker A

"No, Thursday works, but I'll be away on business on Thursday." Both parties actually just want to find a common opening, but if two people's Instincts could directly connect and cross-reference their free time to automatically sync their calendars, all

04:17

Speaker A

that back-and-forth would disappear. The user only needs to input one intent: "I want to meet so-and-so this week, prioritize these time slots and get it done," and the Instincts on both sides can coordinate the rest. The core of

04:28

Speaker A

this network is called a "Trusted Personal Network." You only connect with people you truly trust, and trust is tiered: spouses might share everything, while colleagues might only have access to work calendars and parts of your inbox. It’s not a flat map of

04:41

Speaker A

friends, but a weighted trust network with varying levels of access. Even more interesting is that it mirrors real-world social constraints; if you gain calendar access but try to snoop on other data, the other person's Instinct will immediately sound the

04:53

Speaker A

alarm, like "Patrick is looking for this type of information." Once trust is abused, real-life interpersonal relationships will also start to show cracks. Noah also shared a case of a group of six friends who let their respective Instincts collaboratively

05:05

Speaker A

plan weekend activities every week, automatically syncing their schedules, considering everyone's preferences and dietary restrictions, and even linking Spotify listening history to analyze shared musical tastes for recommendations. When heading out, it only calls one Uber; the system automatically calculates the optimal

05:18

Speaker A

route and notifies each person in sequence, "I'll be at your building in ten minutes." After shedding the tedious costs of social coordination, many previously unimagined uses for Instinct have emerged. Next, the two discussed a bigger topic: when everyone

05:29

Speaker A

has an around-the-clock super-agent, what will the internet and commerce become? Meng said he believes the internet will be completely rewritten in the coming years. Take restaurant reservations as an example: in the past, you had to go to a website yourself

05:41

Speaker A

to grab a spot, first come first served, and if you missed it, you were out of luck. But what if an agent could check every five seconds across every restaurant worldwide? Going a step further, it can achieve bilateral

05:50

Speaker A

intent matching: the user side can communicate, "It's my spouse's 30th birthday," and the restaurant side can say, "We prioritize major anniversaries." Demands on both sides are precisely matched, which completely breaks the mechanical model of traditional traffic funnels and speed-based booking.

06:02

Speaker A

Changes in the travel and transportation sector are even more direct, as 50% of the transaction volume on the Instinct platform comes from travel. Although it is still invite-only with a small user base, the annualized transaction volume is already approaching

06:14

Speaker A

gave a specific scenario: you just need to send a voice message saying, "I must be in New York tonight," and Instinct will retrieve your location, travel preferences, habits, flight seat selections, and linked credit cards. It issues the tickets directly, seamlessly

06:28

Speaker A

switches to the hotel module, places an order based on your past stay preferences, and integrates the entire itinerary while arranging your transport to the airport and from the airport to the hotel. From the user's perspective, they only said one

06:38

Speaker A

sentence, and all the complex steps behind the scenes were smoothed over. This raises a very practical question: why should users trust an AI agent that has access to their email, calendar, and credit cards? Noah observed an interesting pattern in the data:

06:50

Speaker A

building trust takes a few weeks. The figure seen a few days ago was that by the third week, 40%of users will proactively hand over their credit cards to Instinct. This 40%includes all data, even users who churned mid-process. He treats the time of the

07:04

Speaker A

first card linking as a proxy metric for the level of trust, taking it extremely seriously. Once a user connects at least one piece of sensitive information, the retention rate for those who truly begin to trust Instinct is 80%—a staggering number

07:15

Speaker A

in consumer tech products. Regarding security architecture, Noah split it into two layers: the first is encrypted storage of sensitive information, which is a tedious task with mature industry solutions; the second is a brand-new challenge—the new attack surface brought by an agent that can convince

07:27

Speaker A

40%of users to hand over their credit cards and full email access within three weeks. They designed a security system completely separate from the architecture of Instinct’s core agent . Any external input must first pass through a firewall to intercept

07:39

Speaker A

malicious content. Every action the agent attempts to execute, and even every chain of thought, is monitored in real-time by an independent system that can pause, intercept, approve, or veto actions before they occur. Simply put, it uses a watchdog not bound by the

07:50

Speaker A

same incentive mechanism to keep an eye on the clever "protagonist." Business models and alignment are another unavoidable topic. Noah's stance is very clear: Instinct must never monetize through advertising. His logic is sound: if an agent smarter than the

08:03

Speaker A

user makes money by pushing them to buy things they don't need, using its intelligence to persuade you, that is an extremely dangerous world. Most consumer internet giants today, such as Google, TikTok, Instagram, and Snap, are essentially free to use, stuffing

08:16

Speaker A

you with ads instead; as the saying goes, if you don't pay, you are the product. So, Instinct isn't just a task executor; it pursues a higher-dimensional goal: learning to build trust with users, providing peace of mind, and serving as a safety net

08:29

Speaker A

when things go wrong. So, how does it make money? Its reference model is Apple Pay or American Express, where the user enjoys a seamless, free experience, and merchants pay a reasonable channel fee to gain distribution and reach. In a digital

08:42

Speaker A

transaction chain, there are about 40 intermediary nodes taking a cut, but they are only splitting a 2%to 2.5% payment fee; this pie is quite small.

08:50

Speaker A

Noah is looking at a different curve: the comprehensive commission rates of major global platforms. Shopify takes 2 %to 3%, Amazon can take over 10%, and Apple takes 30%on in-app purchases. It doesn't mean Instinct has to charge 30% , but on the curve of distribution

09:04

Speaker A

power versus commission rates, there is still huge room for exploration regarding where it will land. In the travel category, some boutique hotels pay up to 30%per transaction, so Noah's idea is to scale up the travel segment first before expanding into other

09:16

Speaker A

industries. Here, there is a very subtle game of interests. When AI fully takes over the process of user requirement screening, comparison, and decision-making, the user's consumption decision-making power is effectively handed over to the agent. When multiple merchants offer homogeneous solutions

09:30

Speaker A

that meet the user's needs, should the AI choose the one with the higher commission or the one with the lower one? Noah explicitly stated in an interview that there is no one-size-fits-all technical solution; the team can set strict rules to

09:41

Speaker A

prohibit commission-based ranking, but real-world business scenarios contain a great deal of ambiguity. It doesn't seek to completely eliminate contradictions, but rather uses the compounding effect of trust to hedge against short-term commercial temptations. User trust is a continuously compounding asset; every

09:54

Speaker A

accurate execution and neutral decision by the AI strengthens it, but this asset is extremely fragile—one algorithmic bias toward commercial interests could cause it to collapse entirely. So, when will the established giants start to fight back? For example , when will companies like Uber start

10:08

Speaker A

to view AI agents as their enemies? Noah conducted an interesting deduction , mapping the digital services of various industries onto a single line.

10:14

Speaker A

Look at what proportion of its revenue comes from user attention on the app, and what proportion comes from the underlying goods or services actually delivered. A simple view is that 70%of advertising revenue will be cut, but there is another possibility. Take

10:26

Speaker A

DoorDash as an example: after shortening the number of clicks for user checkout, transaction volume actually rose because transaction friction decreased. Instinct takes this to the extreme, making friction almost zero, even taking predictive autonomous action—when you land late at night,

10:38

Speaker A

it knows you haven't eaten and messages you directly to ask if you're hungry. Would you like the same meal you ordered last night? This way, transaction volume won't drop but will instead rise. For businesses driven by transaction volume, the era is a

10:50

Speaker A

tailwind, but for companies that monetize thin user attention, it is indeed a dangerous moment. As for how companies should prepare, Noah doesn't advise a one-size-fits-all approach, but rather small-scale experiments: open a capability to 1%of users, run A/ B tests to see if they like it more and

11:05

Speaker A

if transaction volume rises, then dilute the risk with data. On the level of product philosophy, Noah proposed a guideline: don't focus on AI capabilities, focus only on user understandability. How well do users understand what is happening? How well

11:17

Speaker A

can they predict what will happen after they take an action? This concept even extends to the layout of a text message ; when users look at large blocks of text, they follow a banner-style scanning pattern—reading 80%of the first line, 50%of the second, and then

11:30

Speaker A

declining. Instinct therefore puts core information in the first 30%of the layout, minimizing cognitive load.

11:36

Speaker A

These seemingly "soft" skills are the real infrastructure investment and what he believes will be the source of future differentiation. When asked how much of these details come from personal taste versus data-driven decisions, Noah said it's both. Soft metrics like trust cannot be quantified

11:50

Speaker A

through short-term A/B testing, so their strategy is a phased canary release: test it internally for two days, then push it to the team, then to early user groups, and finally roll it out completely. In terms of growth, Instinct's marketing spend to date is

12:03

Speaker A

zero. Initially, it was sent to 200 friends and family, with five invitation cards each; the second day only added five people, followed by 1% to 2%daily growth. Once it reached a few thousand users, people started sharing tips online. Growth surged to 6

12:16

Speaker A

%, 7%, 8%, 9%, and is now around 10%to 11%daily. This means 10%of users are willing to give away their precious invite codes, with some even spending $ 300 online to buy one. The invite system isn't to create hype; it's

12:28

Speaker A

because compute power is limited. We can't wake up one day to find users have doubled, only for 80%of them to be unable to use the service. Compute power is the issue Noah spends 40%of his time worrying about. For

12:38

Speaker A

traditional consumer products, you just double your servers to handle double the users. But Instinct is different; the underlying compute demand grows 10% every day, doubling every single week.

12:47

Speaker A

Buying double the compute gets used up in a week, five times lasts less than three weeks, and ten times only buys you two months. Even more fatal is the months-long lead time for hardware delivery. Last-minute procurement costs three to four times the premium. Even

12:58

Speaker A

if growth slows from 10%to 5-8%, the compounding effect over a three-to-four-month lead time still reaches the scale of hundreds of millions of users. This is an all-or-nothing bet. In terms of cost optimization, Instinct has a secret weapon: they can run at a cost far

13:10

Speaker A

lower than previous models while achieving accuracy and performance comparable to GPT-4. The reason is that a large portion of an agentic product's work doesn't need to be done in milliseconds; it can be processed in minutes or even hours. Handling these

13:21

Speaker A

batch tasks with customized inference deployments can yield three, five, or even eight times the efficiency. By saving across different stages and improving efficiency fivefold while cutting costs by 10%, the cumulative effect allows for service at an extremely low cost. Regarding the

13:33

Speaker A

evolution of user experience, Noah believes the future will be even simpler; a group of users already sends over 90%of their messages via voice. He mapped the phone's side action button to Instinct. Just press it and say, " Tell Patrick I'll be there," and it's

13:45

Speaker A

sent—no need to unlock the phone, open an app, or type a single word. Looking ahead, with real-time voice and audio processing, when you’re walking or cycling with standard AirPods, it can summarize messages, schedule items in your calendar, and filter out

13:57

Speaker A

unimportant tasks. In the long term, interfaces will become extremely simple , but a simple interface doesn't mean less capability. The more radical trend is that interaction is shifting from handling single tasks to autonomously pursuing long-term goals. Users no

14:10

Speaker A

longer say "help me track this workout, " but rather "accompany me over the next three to four months to ensure I hit these goals," like gaining weight, losing weight, or hitting a specific running pace. Some small businesses are

14:20

Speaker A

even running their entire back-end operations on Instinct, giving macro-instructions like "keep inventory within this range," and it autonomously coordinates tools to compare prices, procure items, and reconcile data. This represents a shift from individual task completion to higher-level goal

14:33

Speaker A

achievement. Regarding the analogies of personality and the movie "Her," Noah is clear that he does not want Instinct to establish that kind of emotional relationship with users. It is more like a socially aware operator; no matter which room it is in, it knows

14:44

Speaker A

how to act best for different people in the moment, and learns how to do so better over time. It could become the most customizable thing in history, not by offering you a few fixed personality sets, but by evolving alongside you. As

14:56

Speaker A

for why it is called Instinct, Noah says he deliberately avoids anthropomorphizing it with a specific name. What exactly is Instinct? Even he feels it gradually taking shape through usage, and it’s fresh for every user; they have no preconceived notions. The

15:09

Speaker A

final point discussed was, facing agents destined to be more powerful in the future, who are the rivals and who are the allies? Noah says he doesn't agonize over whether it’s a messaging app or a WhatsApp-like application.

15:19

Speaker A

Because over 50%of traffic doesn't run on messaging apps at all. He starts from first principles: what are the interfaces users trust and are most familiar with today? Deliver the experience within those channels.

15:30

Speaker A

Regarding capital requirements, the latest round was about $ 1 billion in funding at a valuation of around $ 10 billion, led by Sequoia and Benchmark.

15:36

Speaker A

Noah says doing this is extremely capital-intensive; distributing compute at low cost to billions of people worldwide is expensive. He could easily charge a $ 100 monthly subscription per person, but the purpose of venture capital is to give them the confidence

15:47

Speaker A

to bypass the local optimum of a subscription model and prove that cross-industry transaction fees and experiences are the right path. Looking back at this interview, a 23-year-old leading a 14-person team has created a $ 10 billion-valued AI assistant

15:59

Speaker A

without an app. What makes Instinct impressive might not be how powerful the technology is, but the series of very restrained choices made in its product philosophy. No apps, no ads, no task executors, and no pursuit of emotional relationships with users.

16:13

Speaker A

Behind these choices is a unifying theme: managing user trust as the most important asset. As AI capabilities grow stronger, how to ensure it always stays on the user's side is an unavoidable question for the entire intelligent agent industry. Noah's

16:25

Speaker A

answer is to use business model design to align and constrain interests. By avoiding ad revenue and only charging transaction commissions, this model itself holds a contradiction: when your decisions and the platform’s profit potential point in the same direction,

16:38

Speaker A

why should you trust that it’s on your side? This question may need time to answer, and we'll see.

Topics: Instinct AI AI assistant Noah Shinn personal AI agent trusted personal network autonomous AI Silicon Valley AI AI commerce Reflexion paper invite-only growth

Study this video

- [Flashcards from this video](https://sozai.app/tools/ai-flashcard-generator/?from=transcript&id=76569)
- [Quiz on this video](https://sozai.app/tools/quiz-generator/?from=transcript&id=76569)

Free, made by AI from this transcript. No signup.


---
This is the markdown twin of https://sozai.app/transcript/ai-assistant-instinct-billion-valuation/ — the same content, without the markup.
Published by SozAI (https://sozai.app). Reuse and quotation are allowed with attribution and a link back.
Machine-readable index: https://sozai.app/llms.txt · data API: https://sozai.app/api/