Personal agents won't cannibalize business revenue: lower friction drives more transactions
Instinct's founder says that after reducing checkout friction to nearly zero, transaction frequency between users and merchants rose instead of falling—the opposite of the old ad-driven model.
The video won't play here. Listen to the audio instead:
The argument · tap a timestamp to hear it
No app, only phone and computer
Instinct has no independent application; users interact with it through SMS, voice calls, and email. It has its own phone number and email address, and can even proactively call to send reminders—the founder mentioned that in urgent situations like signing documents, it will proactively tell users it's urgent and there are fewer than ten minutes left. The rationale behind this design choice is: building a new app adds another layer of learning cost for users, while the goal of a personal agent is to replicate ‘social intelligence’—interacting with it like you would with a familiar person, rather than opening yet another tool.
— Noah ShinnAgents can also ‘break off’ from each other
Instinct agents can negotiate directly to align schedules, eliminating inefficient back-and-forth between people, but access permissions are tiered: your spouse might have access to almost all information, while colleagues might only see your work calendar. More interestingly, there's an implicit trust mechanism—if the other party overreaches and accesses information they shouldn't, your Instinct will proactively alert you that they've been viewing that category of information, and this relationship's trust level will decrease accordingly. This ‘trusted person network’ is essentially a weighted relationship graph, not simply a friends list.
— Noah ShinnInvite-only with small sample volume: $1B+ transactions annually
Instinct remains an invite-only product with a small user base, but the platform's annual transaction volume has exceeded $1 billion, with half coming from travel bookings. The founder gives an example: a user just needs to say ‘I have to be in New York tonight,’ and Instinct automatically identifies their current city, compares airline preferences, seat types, and payment methods, books the flight, then books the hotel, and adds round-trip transportation to the calendar—the user never needs to open any app. He calls this type of end-to-end travel booking a redefinition of the traditional OTA business model.
— Noah ShinnAfter three weeks, 40% of users hand over their credit card
Trust builds not through promises but through time. Data shows that after three weeks of use, 40% of users will proactively provide Instinct with their credit card information; once users have shared at least one sensitive piece of information (credit card, password, etc.), retention jumps to 80%. The founder uses ‘the timing of first sharing sensitive information’ as a proxy metric for trust, and emphasizes that users can always revoke authorized data at any time—this is what he considers the key threshold for whether consumer AI products can be adopted at scale.
— Noah ShinnNot a ‘task executor’: rejecting the ad model
Noah explicitly states he doesn't want Instinct to follow the old ‘free plus ads’ path—if an agent smarter than users and more savvy socially convinces them to buy things they don't actually want, that's a dangerous reality. So Instinct isn't simply a tool for executing commands, but is trained to pursue higher-level objectives—building trust for users, making them feel cared for, even if that means giving up some short-term revenue from persuading them to purchase more. This is also why the business model ultimately landed on transaction-based take rates rather than ads or subscription.
— Noah ShinnFriction drops to zero, transactions actually increase
Intuitively, an agent taking over ordering should damage the interests of companies like Uber Eats and DoorDash that monetize user attention. But Noah points out the counterexample: throughout history, every reduction in checkout clicks has been followed by rising transaction volume, and Instinct can reduce friction to nearly zero—for instance, if your calendar shows you'll be late, the car will auto-arrange early; if you get home without eating, it proactively asks if you want that dish you ordered last time. He calls this a counterintuitive game: users spend less time in apps, but actual transaction frequency with merchants actually increases.
— Noah ShinnCompute must be provisioned months ahead; betting wrong costs three to four times more
Instinct's daily active demand itself doubles every week, and compute delivery cycles are measured in months—unlike other consumer products where you can scale resources in line with user growth. Noah describes the difficulty of this decision: buying 2x capacity gets consumed immediately, buying 5x runs out in three weeks, buying 10x carries high-leverage risk—if you judge wrong, excess compute gets forfeited at three to four times the price. He says 40% of his time is spent agonizing over exactly how far ahead to buy and how much to procure.
— Noah ShinnProactive agents need an order of magnitude more compute than code-writing products
Unlike programming agents like Cursor or Codex, where users ask questions, the model does work, and then returns results, Instinct's architecture allows it to wake up at any point during the day and decide on its own whether to bother the user—for example, scanning the day's schedule at dawn, finding something worth reminding about in the afternoon, then proactively reaching out to the user. Noah believes this native proactivity means the required compute is an order of magnitude higher than initially thought, and the industry's current compute buildout has barely scratched the surface of what proactive work demands.
— Noah ShinnIn their own words · checked verbatim
This is probably the most exciting software race ever. And the outcome is like trillions of dollars.
I'm not going to spice it up because it's really just a personal assistant.
Noah Shinn3:04
Man, any product that's dependent on consumer laziness or inertia is toast, huh?
Patrick O'Shaughnessy8:18
there's like an 80% retention rate, 80% if you share one piece of information.
Noah Shinn25:50
This is the idea that if you're not paying, you're the product.
Patrick O'Shaughnessy30:58
Do we buy compute, you know, 2x, 2x of what we have right now? Uh, well, we're going to consume that in a week, right? So then do you buy 5x?
Noah Shinn58:54
I think that the amount of compute that is going to be needed is going to be, honestly, orders of magnitude more than what we thought that we needed.
Noah Shinn1:03:59
Figures
| Travel share of transaction volume | 50% | 18:32 |
| Users sharing credit cards within three weeks | 40% | 25:50 |
| Retention after sharing at least one sensitive item | 80% | 25:50 |
| Daily sequential growth rate | 10%-11% | 54:51 |
| Invitation code resale price on eBay | ~$300 | 55:52 |
| Shopify take rate | ~2%-3% | 35:05 |
| Amazon take rate | 10%+ | 35:05 |
| Apple take rate | 30% | 35:05 |
Glossary
- trusted person network
- A tiered-permission social graph where Instinct agents can directly negotiate calendar alignment with access levels based on relationship (spouse vs. colleague) and trust history.
- take rate
- The percentage of transaction value a platform keeps as commission, similar to Apple Pay or American Express models.
- proactivity
- An agent's ability to wake up and act autonomously without user prompting, deciding when and how to intervene or remind.
How to listen
Founders and investors interested in consumer AI product design, transaction-based business models, and AI infrastructure compute purchasing decisions.
The opening segment defining ‘what is Instinct’ has low information density and can be skipped.