A 23-Year-Old College Dropout Built a $10 Billion AI Assistant. The Story Is Better Than the Headline.

Noah Shinn’s Instinct went from private beta to a $10B valuation in months. It is a real product with real traction, real privacy risks, and a business model that still has to prove itself.

This is almost too perfect a Silicon Valley story. A 23-year-old student drops out of college. Builds an AI assistant.

Seven months later, investors are throwing billions at him. The startup is still invite-only. Most people cannot even use it. And a month after raising at a $2.5 billion valuation, Instinct raised another $1 billion at a $10 billion valuation.

Four times higher. In one month.

Reuters says Sequoia, Benchmark and Coatue participated in the latest round. The company has only 14 employees.

Somewhere, an entrepreneurship-course salesman has already made the thumbnail: HE DROPPED OUT AT 23 AND BUILT A $10B COMPANY. WHAT’S YOUR EXCUSE?

Fine. There is only one small problem with this version. Almost none of the interesting part happened overnight.

The “college dropout” part is technically true

Noah Shinn did leave Northeastern. But this was not someone discovering Claude Code on Friday night and vibe-coding a unicorn before Monday morning. By 2023, while still a student, Shinn was already the lead author of Reflexion, a NeurIPS paper on language agents that learn from feedback by storing verbal reflections in memory rather than updating model weights. His co-authors included researchers affiliated with MIT and Princeton.

Before Instinct, he worked as a research scientist at Sierra. Sierra’s own bio says his earlier work covered machine learning and programming-language research at Northeastern and MIT.

In other words, he spent years working on almost exactly the thing that later became valuable. This is less cinematic. It is also a much more useful lesson.

“Drop out and build something.” No.

Get unusually good at something before everyone suddenly decides it matters.

Then build something. Slightly harder course to sell.

And Instinct is not vaporware

I have used Instinct myself, and unfortunately for the easy skeptical take, it is actually good.

The basic idea is almost boring now: one persistent personal AI agent, reachable by text or phone, with access to a computer and parts of your digital life. But the execution is the product.

Instinct can connect to email, messaging, screen, audio, location and other context. It can follow up on things you forgot, arrange rides, make calls, book services and proactively return to unfinished tasks. The company’s own formulation is good: “there are no new interfaces.” You communicate with it the way you communicate with a human assistant.

You can say something like: I’m flying to Paris tomorrow. Sort the trip out.

A useful assistant should not just search “Paris hotels.” It should check when you land. Look at your meetings. Compare hotels and transport. Notice that the cheap hotel creates a stupid commute. Figure out whether you need an airport transfer. Maybe call somewhere. And ideally remember the whole thing next week.

This is what people wanted when they talked about AI agents for the last two years. Not another chat window with 14 buttons called Tools. A digital person who handles the annoying bits.

Users have reportedly used Instinct for travel, groceries, concert tickets, subscription cancellations and even wedding planning. More than half of transaction volume on the platform is currently travel-related, according to Shinn. The product remains invite-only, but reports in September put its user base above 100,000. So there is clearly something here.

The magical part is also the terrifying part

For an assistant to be genuinely useful, it needs context.

And “context” is a pleasantly harmless word for: your email; your calendar; your messages; your contacts; your location; your screen; possibly your credit card; and eventually most of your digital life.

The better it knows you, the more useful it becomes. Wonderful.

The better it knows you, the worse a failure becomes. Less wonderful.

Early Instinct users have already found some fairly serious edge cases.

One reported that the service retained Gmail information after access was disconnected. Another found that it continued summarizing email after disconnection. One tester demonstrated that an ordinary email could be used for a prompt-injection attack. Another user said Instinct sent an email on her behalf without asking first. The company subsequently added data-deletion controls and has said it takes the security issues seriously. This is not the same failure mode as ChatGPT hallucinating the population of Belgium.

If a chatbot says something stupid, you close the tab. If an agent has access to your inbox, financial credentials and the authority to transact on your behalf, “oops” becomes a much more expensive word. That is the central product problem for personal AI.

The useful version requires enormous trust. The dangerous version also requires enormous trust. Convenient.

And then there is the compute bill

Instinct is currently free. That sounds fantastic until you remember that an agent spending 20 minutes browsing websites, running a cloud computer, reading context, reasoning, calling tools and trying again is not exactly the cost profile of a weather app.

Instinct has already had periods where users reported slower responses and missed tasks as demand pushed against available compute. The company has kept access limited with a waitlist while scaling infrastructure.

This is particularly entertaining because the product’s central promise is: Give me the boring work. I’ll do it for you.

Which translates financially into: Please spend tokens continuously while the user is somewhere else.

That gets expensive rather quickly. And this is where the business story becomes more interesting than the valuation headline.

The monetization story has already changed

A few weeks ago, the obvious question was how Instinct intended to make money at all. Subscription? Advertising? Premium tier? Sell your soul to Marriott?

Shinn has now given a much clearer answer. His stated goal is to keep Instinct free for users for life. And, importantly, he argues against advertising because an agent paid to influence the user would create exactly the wrong incentive. If your personal AI understands you better than anyone else and then starts nudging you toward whoever pays it most, we have invented an extremely sophisticated salesman disguised as your best friend. That would be bad.

Instead, Shinn has described a transaction model: Instinct remains free, while merchants and services pay a take rate when transactions flow through the platform. Instinct is approaching $1 billion in annualized transaction volume, according to Shinn, although that is gross transaction volume, not Instinct revenue, and the exact calculation has not been disclosed. This is actually a much more interesting model than ads.

If your agent books the hotel, flight, restaurant, insurance, groceries and eventually half the things you buy, it becomes a distribution layer. Think Apple Pay.

Except Apple Pay does not know that you are exhausted, have three meetings tomorrow, hate Heathrow and promised your wife you would be home Thursday night.

Your agent might. There is potentially a very large business hiding there. There is also potentially an enormous conflict-of-interest problem hiding there.

The details matter. A lot.

So is $10 billion insane?

Probably. Maybe. Welcome to AI valuations.

Instinct was valued at $2.5 billion in late August after raising $250 million. Roughly a month later it raised $1 billion at $10 billion. The public product is not fully launched. The service is still constrained by compute. Security is still being hardened. The user business model is intentionally free. The merchant take-rate business model is, for now, mostly a thesis.

Under the financial logic we used for software companies ten years ago, this would be an entertaining investment committee meeting. Under 2026 AI logic, apparently this is conservative.

Safe Superintelligence, Ilya Sutskever’s research lab, reached a $32 billion valuation without a commercial product or revenue.

At least Instinct can book you a hotel. Progress.

But the investors may be betting on something bigger than Instinct

I think this is the part that makes the valuation easier to understand, even if it does not make $10 billion sensible.

The bet is not: Will this particular AI assistant generate enough subscription revenue?

The bet is:

Will a personal AI agent become one of the primary interfaces between humans and the economy?

If yes, the winner sits in an absurdly valuable position. It sees demand before Google does. It decides which products enter the shortlist. It negotiates with services. It executes transactions. It knows your schedule, preferences and constraints.

Eventually it may know what you need before you formulate the request.

We spent 20 years building businesses around owning the search box, the social feed, the app store and the marketplace. A persistent personal agent potentially sits above all of them.

That is what venture investors are buying an option on. Not a chatbot. An interface layer for a person’s digital life.

The less glamorous conclusion

The Noah Shinn story is genuinely impressive. Just not for the reason that will appear in motivational TikToks. He did not randomly drop out, discover AI and build a $10 billion startup in seven months.

He spent years doing technical work on AI agents, published important research, worked professionally in the field, assembled a very small but strong team, and then happened to launch when the market suddenly became obsessed with exactly that category.

The product also appears genuinely strong. I use it. I understand why people get hooked.

What has not been proven yet is the business. Can Instinct scale beyond an unusually technical early-user population? Can it afford the compute? Can it build a transaction business without compromising the agent’s loyalty to the user? Can it secure a product whose entire value proposition is basically “give me the keys to everything”? And can it do all of that while Meta, OpenAI, Google and everyone else concludes that the persistent personal agent is the next interface?

If yes, $10 billion may eventually look cheap. If not, we will have a very expensive case study.

Either way, the interesting part starts now. The fundraising round was the easy bit.