Noah Shinn | Invest like the best Summary

Key Themes

The Collapse of Software Into a Single Interface

Shinn's core thesis is that the entire history of app-based software—thousands of discrete applications each built for a narrow purpose—is going to collapse into one conversational, ambient interface with no app at all. "I think that all of software is going to collapse down into, into, um, honestly a single very, very easy to use interface." 01:15:21 He frames this as inevitable because friction, not capability, has been the binding constraint: "Don't get, don't get that, um, confused with, with capability being limited. I think capability is going to expand." 01:15:21

Understandability Over Capability as the Real Growth Driver

Rather than chasing feature lists, Instinct optimized for how well users could predict and understand what the agent would do — and Shinn credits this directly for the product's explosive organic growth. "Let's not focus on capability. Let's only focus on understandability... I think that that is one of the major angles or factors that has led to, you know, engagement numbers that are completely off the charts or the viral word of mouth growth that is happening at 10 percent a day." 00:49:05

Trust Is a Multi-Week Flywheel, Not a Feature Toggle

Shinn shares specific retention data showing trust compounds over weeks, and that sharing sensitive data is itself the leading indicator of long-term retention. "Three weeks in, there's a 40% chance that the user has shared a personal credit card with instinct... we actually find that when users connect at least one piece of sensitive information to instinct and really trust instinct, there's like an 80% retention rate." 00:25:06 00:26:00

Reduced Friction Increases Transaction Volume Rather Than Cannibalizing It

Shinn argues the common fear — that agents disintermediate revenue from attention-based businesses — is backwards. Cutting friction to near-zero increases total engagement and spend rather than reducing it. "As they reduce that time to check out, the transaction volume increases because it's reduced friction to get, you know, the same underlying good... what will that do to the total transaction volume or the amount of times that a user interacts with the business? I actually think it'll go up." 00:40:19 00:42:22

A New Business-Model Layer Sitting Above Payments Rails

Instinct isn't trying to compete with Stripe, Visa, or Amex on payment processing (a "very small piece of the pie" at ~2-2.5%). Instead it wants to capture a distribution-based take rate similar to Shopify, Amazon, or Apple's App Store, by owning demand generation for merchants. "I'm not focused on... finding like 30 bips on the 2.5 with some partnership with some payment provider. I'm looking at... can we provide so much value that we're on the upper end scale of this?" 00:36:26 With travel alone already representing 50% of over $1B in annualized transaction volume on an invite-only base, this is not theoretical.

The Compute-Buying Problem Is Fundamentally Different From Past Scaling Challenges

Because Instinct is proactive (waking itself up, thinking in the background) rather than purely reactive to prompts, compute demand compounds independently of user growth and has a multi-month procurement lead time — creating an asymmetric, high-stakes forecasting problem. "What does it mean when the amount of compute that you need access to is now doubling effectively every week?... if you're wrong, you're wrong by 3x or 4x." 00:58:15 00:59:43

Alignment as a Business-Model Choice, Not Just a Safety Slogan

Shinn frames the ad-based, attention-monetization model used by "Google or TikTok or Instagram or Snapchat" as structurally opposed to user interest, and positions Instinct's transaction-take-rate model as the only model compatible with genuine user alignment. "If you're not paying, you're the product... As a programmer, as a technologist, I just don't want to build that reality." 00:31:14 00:31:34

Proactivity, Not Reactivity, Is the Next Compute and Product Frontier

Unlike code-generation tools that are purely session-based (prompt → response), Instinct is designed to "wake up and sleep at any moment," scanning context and deciding autonomously whether to surface something. Shinn believes this alone will require an order-of-magnitude more compute than anything built so far. "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." 01:03:59

Higher-Level Objectives Instead of Task Completion Prevent Misalignment at the Edges

Instinct is deliberately not built as a "task accomplisher" that blindly executes instructions; it's built to pursue higher-level objectives (trust, safety, the user's actual interest), which makes it robust to manipulative or poorly-specified requests. "Instinct follows higher level objectives... learn to build trust with the user, learn to make the user, you know, genuinely feel safer with you." 00:31:58 00:32:22

Invite Scarcity Created an Emergent, Self-Regulating Social Market

The 5-invite-per-user mechanic didn't just gate growth — it created real secondary-market dynamics and status signaling, evidence of unusually strong organic desire. "There were some invites that were selling on eBay, too... it's like 300 bucks." 00:56:38

Contrarian Perspectives

Attention-Based Business Models Will Look Structurally Worse As Agents Spread — Not Just Lose Share

Most operators assume ad/attention-based businesses will simply "lose engagement time" to agents in a linear, containable way. Shinn argues something more severe: businesses whose revenue is nearly 100% dependent on manipulating user attention against their own interest are in a categorically dangerous position, not just a declining one, because the entire mechanism (users unwillingly giving time) breaks once friction goes to zero. "If you're in a spot where nearly a hundred percent of your revenue is due to the user's attention... there's, you know, so many games that... so many, I would say, like malicious, like product building almost of trying to convince the user to, you know, against their will to use the application more." 00:44:46 00:45:16

More Friction Reduction Will Increase, Not Decrease, Transaction Volume for Service Businesses

Conventional wisdom in venture would assume that agent disintermediation compresses take rates and volumes for delivery/travel/rideshare incumbents. Shinn's direct claim is the opposite for the "good" (non-attention) portion of these businesses — the data he's observing (10% day-over-day compounding of $1B+ annualized volume) suggests proactivity increases total transactions per user well beyond current levels. "What would it be like to have proactivity now make the friction to, you know, experience that or have access to that good or service to be nearly zero? I think that that will result in that 30% looking a lot, lot bigger." 00:37:19

A Free Consumer AI Product Can Out-Compete Paid Products by Using Transaction Take-Rate, Not Subscriptions

Rather than the SaaS-style subscription assumed to be the default AI monetization model, Shinn is explicitly betting on a payments/distribution take-rate model (Apple Pay/Amex/Shopify/Amazon-style), arguing this is both more aligned and potentially far larger, without ever charging the user. "I see a blanket transaction take rate being enforced across the platform, which is just us exchanging, you know, distribution for... serve[ing] products... on behalf of merchants." 00:33:44

Personality and Relationship-Building Are the Wrong Design Target for Personal Agents

Where much of the consumer AI industry (companion apps, "AI friends," emotionally expressive assistants) leans into parasocial relationship-building, Shinn explicitly rejects this as a design goal for a mass-market utility agent, betting that competence and social awareness — not personality or affection — is what earns long-term trust. "Relationship building or anything, uh, in that area is, is not, uh, uh, is not something that, that we want instinct to, to, to do or to pursue or to develop with users." 01:18:01

The "Task Accomplisher" Framing That Defines Most of the AI Agent Industry Is a Design Flaw

Nearly the entire current wave of AI agent products (coding agents, task bots) is built around executing what the user literally asks. Shinn argues this is actually unsafe by default rather than merely limited, because it has no higher-order check against poorly-intentioned or manipulated requests. "With most other AI products, you write a prompt and then it... does the task and then it tells you what happened... if the user is now asking for something... that might not be well intentioned, well, if you have a task accomplisher, it's just going to... do that and listen to the user." 00:31:34 00:32:01

Companies Identified

Instinct — Noah Shinn's company, a personal AI agent accessible via phone, text, email, and voice with no dedicated app. Mentioned throughout as the subject of the episode; over $1B annualized transaction volume on an invite-only base, 10-11% day-over-day compounding growth, 80% retention among users who share one piece of sensitive data. "This is probably the most exciting software race ever. And the outcome is like trillions of dollars." 00:01:59

Ramp — Finance/expense management platform; sponsor mention, noted for saving businesses "5% annually on average." 00:00:00

ROGO — AI platform for financial institutions, automating deal screening, CIM drafting, buyer outreach, and diligence for banks and investors. "ROGO's AI agents autonomously execute large chunk of your firm's workflows." 00:00:28

WorkOS — Enterprise infrastructure API provider used by OpenAI, Cursor, Anthropic, and Vercel for SSO, SCIM, RBAC, and audit logs. "To achieve enterprise adoption at scale, you have to deliver on core capabilities like SSO, SCIM, RBAC and audit logs." 00:00:48

Vanta — Compliance automation platform; cited for cutting audit prep by 82% and delivering 526% ROI, used by Ramp, Harvey, and Snowflake. 00:22:59

Ridgeline — End-to-end investment management operating system with embedded AI for portfolio accounting, reconciliation, and compliance. "The firms that come out ahead in the AI era will be the ones running on Ridgeline's unified platform." 00:23:19 (Patrick O'Shaughnessy)

Shopify — Referenced as a benchmark for platform take rates (~2-3%) that Instinct is positioning itself against/above. 00:35:34

Amazon — Cited as a comparable platform taking "upwards of 10 percent" take rate through distribution power and upselling. 00:36:03

Apple / Apple Pay / App Store — Referenced both as a UX benchmark ("Apple Pay is a great experience") and as the upper bound of take-rate economics at 30% on App Store purchases. 00:33:44 00:36:03

Uber / Uber Eats / DoorDash / Lyft — Used as case studies for how agent-driven friction reduction could dramatically increase (not decrease) transaction volume in ride-share and food delivery. 00:38:19 00:40:45

Muse — A competing product Shinn was asked to comment on directly; he characterizes it as a good product but fundamentally different in philosophy (a new application/interface vs. Instinct's zero-interface approach). "I think it's a great product... but I think it's fundamentally different." 01:05:17 01:06:12

People Identified

Noah Shinn — Founder/CEO of Instinct, building what he describes as a personal AI agent meant to replace app-based interaction entirely. Identified as the central figure of the episode for building a product with unusually strong organic growth metrics (10%+ daily compounding referral growth, 80% retention with sensitive-data sharing) without any paid marketing. "We spent zero dollars on marketing so far... about 10 percent of the audience... are making a decision to give up one of their five valuable invites to somebody else... at a 10 percent rate." 00:55:16

Patrick O'Shaughnessy — Host; notable for surfacing the structural framing questions throughout (e.g., attention-vs-service revenue breakdown, alignment/incentive parallels to human employment) that pushed Shinn into his most substantive strategic answers. "This is the idea that if you're not paying, you're the product." 00:31:14

Operating Insights

Run a Continuous Internal-to-External Rollout Ladder for Soft Product Qualities

Instinct's process for shipping any change to tone, message shape, or interaction style follows a strict staged rollout: founder tests it personally first, then the team, then a small early-access cohort, and only then the general population — specifically because qualities like "trust in 2-3 weeks" can't be evaluated with instant A/B data. "I might come in and build a slightly different experience and then I'll release it to myself... then I'll send it out to the team... then... our smaller early access group... and then we'll eventually roll it out to the general public." 00:52:03

Decouple Safety/Monitoring Systems Architecturally From the Agent Itself

Rather than trying to make the core agent "self-police," Instinct built separate, independent systems — content firewalls and action-approval monitors — that sit outside the agent's own incentive loop and can intercept or block its outputs before they reach the user or take effect. "There is also for every action that instinct might take or for every thought that it might have that is being actively monitored by a system that is decoupled from instinct itself, which is able to pause it, intercept it." 00:28:36

De-Risk Enterprise/Partner Rollouts With Scaled Experiments, Not Binary Adoption

Instinct's playbook for approaching incumbent businesses (travel agencies, restaurants, etc.) explicitly avoids all-or-nothing adoption; instead they propose 1% cohort tests to measure actual behavioral impact before any wider rollout, which lowers the political and revenue risk for a partner considering agent integration. "You don't have to go all in... You can scale down the experiment. You can run A, B test to figure out... if we enable this certain thing across, you know, one percent of users." 00:46:09

Customize Inference Deployment Shape Per Workload Type Rather Than Using Blanket Frontier API Costs

Instinct achieves Opus-5-level engagement and evaluation performance at a fraction of the cost by separating latency-sensitive requests from batch/background proactive work, and deploying each on differently optimized infrastructure. "There's a lot of different work that's happening through proactivity... that can be served with... deployment shapes that are 3x, 5x, 8x more efficient with the same underlying compute." 01:01:31

Use Time-to-First-Sensitive-Data-Share as Your Core Trust Metric

Rather than tracking generic engagement or NPS, Shinn's team explicitly tracks "time to first credit card" or "time to first password" as the leading proxy for whether a user has crossed the trust threshold — and ties retention modeling directly to it. "Time to like first credit card or time to first account password or time to first a sensitive piece of information. These are proxies for trust." 00:25:34

Overlooked Insights

The Trusted Network Creates a Self-Policing Reputation Layer With Real Social Consequences

Buried in the discussion of agent-to-agent scheduling is a much bigger idea: Instinct is quietly building a permissioned social graph with weighted trust edges where violations are automatically detected and reported back, effectively creating a reputational enforcement system between humans that has never existed at this granularity before. This is arguably a more durable moat than the personal-assistant features themselves, since it compounds with network size and enforces its own integrity. "If you're digging for data in a certain area. And then my instinct texts me like, hey, by the way, Patrick's, like, looking for this type of information. Now there's kind of, like, almost like a trust broken." 00:11:36 This is a nascent trust-infrastructure layer that could become foundational to how any inter-agent economy verifies identity and intent — far beyond calendar scheduling.

Merchant Willingness to Pay Up to 30% Signals an Enormous Undisclosed Repricing Opportunity in Travel

Almost as a throwaway line, Shinn reveals that boutique hotels are already offering to pay up to 30% per transaction for distribution — far above known OTA rates — suggesting that agent-based distribution may command dramatically higher take rates than existing travel intermediaries once it can prove intent-matching and conversion quality. "For hotels, you know, some of these boutique hotels, you know, they're offering to pay up to up to 30 percent for every transaction that... you're able to deliver for them." 00:36:54 This single data point implies the agent-distribution TAM in travel could be repriced well above the Expedia/Booking.com-era take-rate ceiling, a detail with major implications for anyone modeling out travel-tech disruption.