A Personal Assistant With No App — Notes on Noah Shinn and Instinct
Notes on Invest Like the Best EP.493 (2026-09-28): Noah Shinn on Instinct, an AI personal assistant with no app, how it turns trust into a measurable metric, and why a free-plus-take-rate business model decides whose side it is on. Educational, not investment advice.
Contents
The mind is the ruler of the body and the master of the spirit. It issues commands; it takes none.
—— Xunzi, “Dispelling Blindness” (Warring States period; translation mine)
On 28 September 2026, in episode 493 of Invest Like the Best, Instinct founder Noah Shinn talked at length for the first time about the company he started roughly a year ago: an AI personal assistant with no app, reached through text, phone and email. He offered numbers — over $1 billion a year in transaction volume running through the platform, half of it travel; three weeks in, 40% of users have handed it a personal credit card; users who share one piece of sensitive information retain at 80%; growth around 10% day over day on $0 of marketing. These come from the founder on a podcast, with no third-party verification, and the product is still invite-only and young. What stayed with me is how he turned trust into something he can measure.
What the episode is about
Patrick O’Shaughnessy opened by asking what game is actually being played here. Noah’s answer was blunt: you are up against the biggest companies in the world, the window is months long, and you have no distribution advantage — just a product that keeps spreading.
The product shape is odd in its simplicity. There is no application. It has a phone, a computer and its own email address. You text it, and it can call you. Noah said it has called him about three times in several months, once with: “I don’t want to bother you too much, but you need to sign this document by 3PM and it’s 2:55 right now. It’s in your inbox. I can send you another email to push it to the top.” He uses that story to explain what he is after — social awareness rather than a longer feature list.
It runs invite-only, five invites per user. The constraint produced behaviour he did not plan: people emailing to ask for a slot, and invites selling on eBay for around $300.
The points I wrote down
1. The interface is the one you already use. Noah’s view is that talking to AI should not require a new application. He is careful to separate a simple interface from limited ability — the thing has a computer, so it can do what you can do on the internet. More than half of their traffic does not run on iMessage; the company stays unpinned from any single channel and meets people where they already are.
2. Their early principle was understandability, not capability. Three years of launches have been about what AI can now do, and Noah thinks consumers are tired of that framing and that it misses the point. Their question became: does the user understand what is happening, and can they predict what it will do next? That extends to the shape of a message — he says people read about 80% of the first line, 50% of the next, tapering down like a flag, so the important part goes in the first third and the rest is optional. I replayed that part, because it turns a soft word (“good experience”) into something you can edit.
3. The trusted network, and the Uber that picked up six people. Two Instinct users can connect so their assistants negotiate a time directly, skipping the back-and-forth. The design point is graded access: a spouse might see everything, a colleague only the work calendar. What happens on a violation is interesting — if someone starts digging outside what you granted, your Instinct texts you that they are looking for that kind of information, and what breaks is the relationship between two people. Noah described a group of six friends who let their assistants plan something new each week around their music tastes; one time a single Uber routed around and collected all six in sequence.
4. Trust as a measured quantity. They track time to first credit card, time to first account password, time to first sensitive item, and treat those as proxies for trust. Forty percent at three weeks — with churned users still in the denominator. Users who share one sensitive item retain at 80%, a figure Patrick called rare for consumer technology. My reservation: early adopters self-select for willingness, and nobody knows what this ratio looks like a hundred times further out.
5. The business model decides whose side it takes. Noah spent real time on why not ads. Picture an assistant more socially aware and more capable than its user, pointed at convincing that user to buy something they do not want — he called that a dangerous world and said he does not want to build it. So: free for users, a take rate from merchants. He anchored the range against what exists — Shopify around 2.5% to 3%, Amazon upward of 10%, Apple’s in-app purchase at 30%, and some boutique hotels willing to pay up to 30% per booking delivered. He said outright they will not be at 30%, and that he does not yet know where on that curve they land. He also drew a design line: most AI products are task accomplishers that do what you ask, while Instinct follows higher-level objectives such as building trust and keeping the user safe, with doing the task being one way to serve those.
6. Pushing friction to zero may raise transaction volume. This is the counterintuitive one and the inference I found most worth unpacking. Start with the split: divide a company’s revenue into the part earned from your attention inside the app and the part earned from delivering the underlying good. The intuitive conclusion is that assistants delete the first part. Noah’s counter is that delivery platforms already know that every click removed at checkout raises volume, because the resistance to getting the same good drops. An assistant takes that resistance near zero, and with proactivity it can have the car waiting before you think to order it — so the underlying-service share grows. He did not claim this holds everywhere; his advice to incumbents is to run it across 1% of users and read the data.
7. Compute is an exponential problem. He says it takes about 40% of his time. Fast consumer growth is familiar; having the underlying compute grow 10% a day is not. Buy 2x and you consume it in a week; buy 5x and it is gone in under three; buy 10x and you are back in a few weeks. Lead times run months, and buying late costs three to four times. His arithmetic: even at 5% to 8% a day compounded over the three or four months it takes to bring capacity online, you arrive at something like 100 million users — so do you buy for 100 million? On the cost side, proactive work does not need to finish in hundreds of milliseconds; minutes or hours are fine, and that batch shape can be served on deployments three to eight times more efficient. He claims parity with Opus 5 on their A/B tests and internal evaluations at a much lower cost — his own number, unverified.
Growing 10% a day sounds like good news
Plenty of people read “10% day over day” and “$0 on marketing” and feel they are late to something. The episode left me with the opposite reflex: that growth is also a liability.
The reason sits on the supply side. A software company that adds ten times the users can spin up servers in days. Instinct has to order compute months ahead, and being wrong costs three to four times. That is why the invite-only gate exists — Noah said he does not want to wake up with ten times the users and 80% of them unable to get a response. The $300 invites on eBay are a symptom of that supply constraint, not a marketing tactic.
I use this as a measuring stick elsewhere: when a demand curve looks beautiful, ask how long the lead time is on the supply side, and what the pre-bought capacity becomes if demand stops. Longer lead times and harsher penalties for buying wrong mean the growth itself carries more risk. It applies to data centres, grid equipment and specialty materials, and it is the habit I most want to keep from this episode.
Who pays for the free thing
Everyone has heard “if you’re not paying, you’re the product,” and few people take it apart. Noah’s exercise gives you a way to: split a company’s revenue into the portion earned because you are staring at its screen and the portion earned because it delivered the underlying good.
There are two uses for that split. As a consumer: take the three apps you open most, and ask whether each still earns money from you if someone else operates it and you never look at the screen. The ones that do not are built to keep you there, which runs against your goal of finishing and leaving. Noah used the word “liberating” about social platforms — users unhappy on the app, unable to stop scrolling, because that is what the business is paid for.
As an investor: the same headline about AI assistants disrupting an industry resolves differently depending on which side of the line a company sits. Volume-driven businesses may get a tailwind from lower friction; attention-monetised ones need a separate calculation. It does not translate into a buy or a sell — it adds one question when you read the news: which part is this company’s money coming from? Worth noting that Instinct sits on the far side of that line itself, taking a cut of transactions, which gives it a reason to want you looking at your screen less. Whether that alignment survives scale is the thing I will keep watching.
One Thing to Take With You
The idea I am keeping: split the value of anything into the part that exists because you are here, and the part that exists because something was actually delivered. It works on companies and it works on your own job — how much of your pay comes from occupying the seat, and how much from delivering something nobody can take from you? As the cost of having errands run keeps falling, the first part shrinks and the second holds up longer.
Here is something I tried and you can do today: open your phone, look at the top three apps by screen time, and write one sentence for each — “if someone else used this for me and I never looked at it, would I still pay for it?” One or two answers will be no. Those are the ones living on your time. It has nothing to do with investing; it just tells you where your attention is going.
This article is an educational discussion of investment method. It is not advice to buy or sell any individual security, offers no target prices, and does not analyze any current holding. Investing carries risk; make your own decisions or consult a qualified professional.