Inside the Personal AI Assistant Growing 10% a Day | Instinct Founder

Summary

• Instinct is approaching $1 billion a year in transaction volume on a very small, invite-only user base, growing about 10–11% day over day with $0 spent on marketing. Noah Shinn frames the compounding starkly: “That’s not just $1 billion flowing through the platform now. That’s $1.1 billion tomorrow,” then roughly $1.2-something and $1.3-something the next days. Patrick says he saw invites selling on eBay for about $300. Travel accounts for 50% of transaction volume.
• Shinn’s envisioned business model is a blanket transaction take rate, rather than ad-funded persuasion; he contrasts it with subscription as a shorter-term alternative. He benchmarks Shopify at 2–3%, Amazon at “upwards of 10%,” and Apple in-app purchases at 30%, and notes that boutique hotels have offered to pay up to 30% per delivered transaction. “I’m not focused on finding 30 bps on the 2.5%” of payment rails—the prize is Apple Pay-style distribution economics extended to “effectively every major industry.”
• Trust is associated with unusually strong retention. Three weeks in, about 40% of the user base has shared a personal credit card with Instinct. Shinn treats time to first credit card, password, or other sensitive information as a trust proxy; among users who connect at least one sensitive item and really trust Instinct, retention is about 80%.
• Compute is the existential constraint: demand can double “effectively every week,” while compute has a lead time of several months—so mis-sizing the buy can cost 3–4×. Shinn spends about 40% of his time on this. His math: even 5–8% daily compounding over a 3–4 month procurement lead time could imply 100 million users. “If you’re wrong, you’re very wrong.”
• On serving cost, Instinct claims performance comparable to Opus 5—matching engagement, A/B-test, and internal-evaluation results—at a low cost through customized inference deployments. Background workloads that can finish in minutes or hours run on deployment shapes “3×, 5×, or 8× more efficient,” contributing to Shinn’s personal goal of making the product free for life, though he will not commit to that.
• The incumbent thesis: businesses monetizing user attention are exposed; businesses monetizing the underlying service may see volume rise as agents reduce friction to nearly zero. Patrick says products dependent on consumer laziness or inertia are “toast,” while Shinn argues that proactive agents—a car always lined up or dinner offered as a user lands—could increase interaction with the underlying service.
• The endgame is interface collapse: “all of software is going to collapse down into honestly a single very easy-to-use interface.” Instinct has a phone, computer, and email address rather than a conventional app; more than 90% of messages from some users are sent by voice. The Instinct-to-Instinct “trusted person network” creates configurable access relationships around calendars, plans, and permissioned data.
• Patrick says the latest round was roughly $1 billion at about a $10 billion valuation, led by Zoya, Benchmark, and Kotu, and Shinn acknowledges the scale of the race. The opening framing is a months-long window against the biggest players, with Patrick calling it “probably the most exciting software race ever” and saying the outcome is trillions of dollars.

Deep dive

1. The grand game: a months-long window, trillions at stake

• Patrick opens by asking Shinn to name the game he’s in, and the answer is unusually candid: “You’re going up against the biggest players in the world. The window to do it is actually on the order of months… And you have no distribution advantage, but you have this massively growing viral product. What do you do?” Patrick’s own verdict: “This is probably the most exciting software race ever, and the outcome is trillions of dollars.”
• Shinn’s framing of why the moment differs from past consumer launches: this is not a founder asking users to trust a vision. “We’ve already had that idea since 2023 when we all started to use ChatGPT for the first time” — everyone already has an idea of what AI should act like; Instinct’s early traction comes from the fact that “it just works.”

2. Not an app — a person-shaped interface with a phone, a computer, and an email address

• Instinct, started about a year ago, deliberately requires no new application. “It has a phone and a computer. You can text it. You can call it. It can actually call you too” — it has its own email address, and Patrick confirms he has experienced its calls. Shinn’s example of its social awareness: it phones only when genuinely pressing — “you need to sign this document by 3 p.m. and it’s 2:55 right now.”
• The design principle: “there shouldn’t be any new application needed to be able to interact with AI.” That simple interface should never be confused with limited ability, because the computer lets it do “quite literally anything that you might want to do on the internet.”

3. Emergent use cases: wardrobes, subscriptions, and the death of dark patterns

• The wardrobe case, as told: users scan every item of clothing plus their own face, body, and proportions, have Instinct plan the week’s outfits — rendered as themselves wearing them, not a bullet list — and can ask for three creative new outfits each day. They can then extend the same capability to shopping: Instinct generates “thousands of outfits” across the internet, and the user points to one and says, “Can you send that to my house?”
• The personal-finance case goes end to end where popular consumer fintech products stop: Instinct scans connected bank accounts, surfaces forgotten subscriptions, signs into the sites, handles email confirmations, and cancels them — then reports, “You’re saving $2,000 this month.”
• Patrick’s takeaway, worth keeping verbatim: “Any product that’s dependent on consumer laziness or inertia is toast, huh? Like toast.” Shinn: “Yes.”

4. The trusted person network: agents negotiating for their humans

• Instinct-to-Instinct coordination, introduced only about a week or 10 days earlier, began as a scheduling fix for busy professionals: state an intention — “I want to meet with this person ideally by the end of the week” — and the two Instincts coordinate directly. “Nobody’s really playing games here. We’re just trying to find time.”
• Patrick frames the structure as unlike a simple friend graph, with weighted or configurable edges. Shinn describes the underlying principle: “I’m connected to you. I trust you. I’ve also configured it in this way such that there’s varying levels of access.” Spouses may share everything; colleagues might get a work calendar and selected inbox access. Trust violations can affect the relationship: if a connection starts “digging for data in a certain area,” the user’s Instinct can report it — “Patrick’s looking for this type of information” — creating an implied breach of trust.
• The friend-group example: six friends have Instinct brainstorm a weekly outing from their tastes, Spotify histories, and availability, then arrange one Uber to pick everyone up along an optimal route — removing much of the logistical burden from social coordination.

5. Rewriting reservations and travel — where the first billion flows

• The restaurant thesis: today’s first-come, first-served booking is an accident of internet design. With agents on both sides communicating context — “this is the person’s spouse’s 30th birthday” — restaurants can prioritize the special occasions they want to serve, and users can get tables for events that matter. Shinn says partnerships to “reinvent what the reservation system is” are coming.
• The headline numbers, dropped almost in passing: still invite-only, “we’re approaching $1 billion a year in transaction volume through the platform” on “a very small user base,” and “50% of that is travel alone.”
• His travel scenario is a single voice note — “Hey, I need to be in New York tonight” — from which Instinct could resolve the user’s location, preferred airline, seat, class, meal, and credit card, then find a suitable hotel, potentially one the user has stayed at repeatedly, book Ubers on both ends, and align everything with the calendar. Preferences compound: state a taste once and it can extrapolate that preference to future bookings.

6. Trust is earned in three weeks — and associated with 80% retention

• Shinn’s data on the trust flywheel: “Three weeks in, there’s a 40% chance that the user has shared a personal credit card with Instinct.” He treats time to first credit card, account password, or other sensitive information as a trust proxy — and says that when users connect at least one sensitive item and genuinely trust Instinct, retention is about 80%.
• The governing principle stated repeatedly: “The user should always be in control of their data… share at the rate at which they feel comfortable, and if they want to take it back, they can certainly take it back.” He explicitly does not mind that trust takes weeks to build.

7. Security architecture: firewalls, watchdogs, and decoupled incentives

• Shinn splits the problem in two: storage of sensitive data is “a tractable problem — just very hard work,” while agentic autonomy creates genuinely new surface area. Two systems he describes are “firewalls” that intercept or block malicious content before Instinct consumes it, and an action monitor “decoupled from Instinct itself” that can pause or intercept an action or thought for approval before execution.
• On hallucination, the same decoupling logic applies: a validator outside the agent’s incentive system can catch errors such as “this proper noun was just generated due to some sampling error in the underlying model.” He says an early version of the product lacked firewalls, active monitors, and related infrastructure; the company addressed those mistakes by building systematic defenses rather than merely patching individual cases.
• His stance on the inevitable-breach question: security is “the most important problem,” and history shows that new consumer experiences can receive immediate backlash because “different” gets conflated with unsafe.

8. Alignment through the business model: never the ad-funded persuasion machine

• Shinn’s core worry: “Imagine a world in which Instinct is generally smarter than the user… it would be a very dangerous world if Instinct were influencing the user’s behavior to purchase something that they don’t want.” He names Google, TikTok, Instagram, Snapchat, and other ad-supported platforms as examples of a model he does not want to build. Patrick: “This is the idea that if you’re not paying, you’re the product.” Shinn: “Exactly.”
• The architectural corollary: “Instinct is not a task accomplisher.” It pursues higher-level objectives — building trust, helping the user feel safer, watching over the user, and having their back — rather than merely following every request. When a request is well-intentioned, doing the task is one way to express those objectives; the higher-level framing is intended to make it more robust to edge cases.

9. The take-rate thesis: skip the payment rails, capture the distribution premium

• Pressed by Patrick on where Instinct fits alongside Stripe, Visa, Mastercard, and issuers, Shinn says he is not primarily interested in replicating what the payment rails do. Roughly 40 parties may share only about 2–2.5% of a digital transaction, depending on its source, and he says those providers offer real value. The model he envisions is Apple Pay-like: free and useful for users, while merchants pay for distribution.
• His placement exercise along the distribution-power-versus-take-rate curve: Shopify at “between 2, 2.5 and 3%,” Amazon at “upwards of 10%,” and Apple in-app purchases at 30%. “I’m not saying we’re going to be 30%… we still don’t know where along this curve we’re going to be” — but travel rates and boutique hotels offering to pay up to 30% per delivered booking show existing models that could be used initially, then extended to “effectively every major industry.” The condition he flags honestly: this only works “if there’s significant distribution power.”

10. The corporate agent war: attention revenue dies, service revenue may grow

• Shinn’s diagnostic exercise: plot digital businesses by what share of revenue comes from user attention versus from delivering the underlying good. The naive read is that a business with 70% of its economics tied to ads and 30% to the underlying good would lose the former. His counter-read, via Uber: if Instinct proactively has “a car always lined up… and you will never be late,” friction goes nearly to zero and interaction volume may rise — “that 30% will look a lot bigger.”
• The food-delivery vignette: you land late, Instinct — which booked the Uber, so it knows where you are — texts, “Are you hungry? Do you want the same thing you ordered 5 days ago?” and the user just says, “Yeah, go.” Counterintuitively, users may interact with the business more because the friction to do so is much lower.
• His advice to incumbents avoids blanket answers: run scaled experiments — enable agent access for 1% of users and test whether transaction volume, willingness to buy, and product discovery improve — rather than “go all in… and now you see 70% of your revenue going to zero.” The approach with partners is explicitly “not to come in hot” but to collaborate; innovative CEOs and executives are already acting as early partners.

11. Product philosophy: understandability over capability, feel over features

• The founding principle inverted the industry’s habit: “Let’s not focus on capability. Let’s only focus on understandability” — the user’s ability to understand and predict what will happen. Shinn credits this, not raw features, for “engagement numbers [that are] completely off the charts” and 10%/day word-of-mouth growth. Consumers, he argues, are fatigued by capability announcements and “we’re missing the point.”
• The obsession runs down to message shape: front-loading information because readers may scan “80% of the first line, then maybe 50% of the next line, and then it tapers off so it looks like a flag.” Instinct should craft messages for the lowest fatigue and least cognitive load.
• Asked whether this is taste or data, Shinn answers, “Honestly, it’s all of the above.” It is difficult to build evaluations for whether a user will trust Instinct in two or three weeks, so the process is staged rollout: Shinn first, then the team, then an early-access group, then the general public. He is “very opinionated” about the experience, but also uses user feedback and quantitative testing.

12. Growth mechanics and the compute nightmare

• The growth story as lived: 200 friends-and-family users, five invites each; the next day about five more users arrived, then the total reached 210 — followed by acceleration from 1–2% to “10 or 11% day over day” once use cases went viral, with $0 spent on marketing. The invite scarcity spawned its own social economy: people asking for invites, others bragging about having three left, and Patrick saying he saw invites “selling on eBay… like 300 bucks.”
• The problem consuming about 40% of Shinn’s time: unlike scaling Instagram, here “the underlying compute also needs to grow by 10% every day” — demand can double effectively every week against a compute lead time of several months. “Do you buy 2x? We’ll consume that in a week. 5x? Less than 3 weeks. 10x? Now you’re 10x leveraged” — and “if you’re wrong, you’re wrong by 3 or 4x.” His sobering math: even decelerating to 5–8% daily compounding over a 3–4 month lead time could imply 100 million users.
• The offsetting edge: across A/B tests, internal evaluations, and engagement, Instinct matches “Opus 5… frontier-level intelligence” at low cost by shaping deployments to the workload. Background work that can take minutes or hours runs on configurations “3x, 5x, or 8x more efficient,” and “all of those compound.” Separately, Shinn says the proactive workload — waking at 6 a.m. to prepare the day, deciding to sleep, and waking again at 4 p.m. — means the required token volume could be “orders of magnitude more than what we thought we needed.” He also says the growth rate is faster than that of Claude Code or Codex, which faced similar exponential-scaling questions.

13. Endgame: interfaces vanish, objectives replace tasks — and the $10B bet

• The interface trajectory runs toward less, not more: more than 90% of messages from some users are sent by voice; Shinn triggers Instinct from his phone’s action button; the vision is an ordinary AirPod, always on, with voice recognition and “the discretion to know when you’re addressing it,” triaging contracts and calendar while the user is on a hike. “All of software is going to collapse down into honestly a single very easy-to-use interface” — while capability expands, including a files feature that generates “full web applications” on the fly.
• The next capability frontier is standing objectives, not tasks: fitness goals pursued over months and, in Shinn’s vision, small businesses native to Instinct, with entire back offices running on the platform and autonomous slices such as “make sure this inventory level doesn’t drop below this.”
• On personality, Patrick invokes Her; Shinn draws the line — “relationship-building is not something that we want Instinct to pursue or develop with users.” It should be a “socially aware operator” with “mutual respect and trust,” and the non-human name is deliberate: every user arrives with “a fresh slate.” On Muse, he says, “I think it’s a great product… a different take,” but spends “very little of my time thinking about competition”; walk into any café and almost nobody is using AI the way they imagined.
• Patrick says the latest round was roughly $1 billion at about a $10 billion valuation, with Zoya, Benchmark, and Kotu as leaders. Shinn’s justification for raising capital is that the business is “just very capital intensive” and needs bootstrapping because he does not want to take the easy out — “we could very easily just say everybody on the platform needs to pay 100 bucks a month.” He frames capital as a way to take calculated risks, prove transaction volume, and escape local subscription-based optima. Free-for-life is “my personal goal,” but “I’m not going to commit to free for a lifetime for now.”