## Core Definition The **2000-Hour AI Trust Curve** is Steve Yegge's observation that it takes roughly 2,000 hours of LLM usage to develop genuine trust in AI systems. The trust is not a belief that the AI is perfect; it is the ability to **predict its behavior and failure modes**. ## The Trust Paradox | Common Assumption | Reality | |-------------------|---------| | Trust = AI is accurate | Trust = I know when AI will fail | | Trust grows with success | Trust grows with predictable failure patterns | | Trust = confidence | Trust = calibrated expectations | | Less trust = more checking | Right trust = knowing what to check | ## What Trust Looks Like - **Before trust (< 500 hours)**: over- or under-rely on AI, surprised by failures, check everything or nothing. - **Developing (500–2000 hours)**: start recognizing failure patterns, develop a "this feels wrong" intuition, learn which tasks AI is good and bad at. - **After trust (> 2000 hours)**: predict failures before they occur, know when to trust versus verify, hold calibrated confidence, and can explain the behavior to others. The 2,000-hour mark is roughly one year of full-time (or 2–3 years part-time) AI-assisted work — enough exposure to encounter diverse failure modes. It is where a user develops [[LLM Mechanical Sympathy]], the intuitive feel for how the machine behaves at its limits, like an F1 driver's relationship with the car. *Source: [[Steve Yegge on Vibe Coding]] — Steve Yegge, Latent Space podcast (swyx), Dec 2025, at 4:29 — https://www.youtube.com/watch?v=zuJyJP517Uw&t=269*