Noah Shinn - Building Instinct: The Personal Agent - [Invest Like the Best, EP.493]

What Instinct actually is

Instinct is described by founder Noah Shinn as a personal assistant with its own phone number, email address, and computer — reachable by text, call, or email rather than through an app. It can call the user proactively (e.g., nudging about a document deadline at 2:55pm) and is designed around "understandability" rather than raw capability: an early product principle was to prioritize the user's ability to predict what will happen when they interact with it over adding flashy new features. Shinn frames this as different from prior AI launches because users already have an intuitive mental model — built since they first started using ChatGPT in 2023 — of what an acting AI assistant should feel like, so Instinct doesn't need to "sell" a new vision.

Real-world usage and the trusted network

Users are already deploying Instinct for wardrobe scanning and daily outfit generation tied to online shopping, personal-finance goal tracking, and full-cycle subscription cancellation (Instinct signs into accounts, cancels, and reports savings). A newer feature — the Instinct-to-Instinct trusted network, live only about ten days at recording — lets two users' agents coordinate directly to find meeting times without back-and-forth texting; spouses often grant each other's agents broad access while colleagues get narrower permissions. Shinn describes emergent social dynamics: if one person's agent detects another probing beyond granted access, it flags a "trust broken" signal between the users, and one friend group used the network to auto-coordinate a shared Uber pickup route for six people.

Reordering commerce and incumbents

Shinn argues traditional first-come-first-served systems (restaurant reservations) could be replaced by agent-to-agent negotiation that matches special occasions to available tables, benefiting both diners and restaurants. In travel — 50% of Instinct's transaction volume — a single voice command ("I need to be in New York tonight") can trigger end-to-end flight, hotel, and rideshare booking using learned preferences. On potential incumbent conflict, Shinn frames businesses along a spectrum of how much revenue depends on user attention/ads versus underlying service delivery; he expects reduced friction to increase transaction volume even for services like Uber and DoorDash, and advises companies to run scaled experiments (e.g., testing agent access with 1% of users) rather than going all-in.

Trust, privacy, and security architecture

Beyond the trust-building data (three-week timelines, 40%/80% figures), Shinn describes concrete safety infrastructure: content "firewalls" that intercept and block malicious inputs before they reach the agent, and a decoupled monitoring system that can pause or reject any action or "thought" before execution — explicitly built to catch hallucination-driven errors (e.g., fabricated proper nouns) before they trigger real-world actions. Shinn says an early version of the product lacked these systems; after identifying gaps, the team built systematic fixes rather than patches. He frames the user's ongoing control over their own data — the ability to grant or revoke access at any time — as a core design principle, distinct from historical backlash cycles that accompany new consumer tech.

Compute economics and monetization

Shinn says he spends about 40% of his time wrestling with compute-buying decisions, since demand can double weekly while hardware lead times run months; buying 10x ahead risks being "10x leverage" wrong if growth slows. He claims Instinct matches "Opus 5"-level frontier intelligence in engagement and evaluation metrics at a much lower serving cost, achieved through customized inference deployment shapes (batch versus real-time workloads) yielding compounding efficiency gains. On monetization, he rejects subscription-first thinking, aiming instead for a merchant take rate benchmarked against Shopify (~2.5-3%), Amazon (~10%), and Apple's App Store (30%), while stressing the product stays free for users. The host puts the latest round at roughly $1 billion at a $10 billion valuation, with Sequoia and Benchmark among the leaders; Shinn frames it as capital needed to absorb calculated risk rather than fund short-term subscription economics.