In April 1986, Ford engineer Jim Moylan had a frustrating experience at a gas station that led him to propose a simple yet impactful idea: an arrow indicating the side of the vehicle where the fuel fill door is located. This small change, implemented in the 1989 Ford Escort, became a standard feature in all cars, significantly improving convenience for drivers. Moylan's invention exemplifies what the author calls "epistemic infrastructure" — tools that provide timely information without requiring complicated coordination or training. These tools, like the Moylan Arrow, operate quietly in the background, enhancing user experience without drawing attention to themselves. The article contrasts two approaches to improving the world in the context of artificial intelligence governance. Will MacAskill suggests a top-down international treaty to regulate AI development, focusing on restricting dangerous capabilities. However, this approach faces significant hurdles, such as defining what needs regulation and achieving global consensus among governments with competing interests. In contrast, Owen Cotton-Barratt and others propose a series of practical AI tools that individuals could adopt without waiting for international agreements. These tools include features like community notes, rhetoric highlighting, and reflection scaffolding, all designed to enhance decision-making and critical thinking. The author argues that governance should be subtle and integrated into systems, much like the Moylan Arrow. Effective governance is often unnoticed because it operates through structural changes rather than mandates. Examples include automatic circuit breakers in stock markets and unemployment insurance that adjusts during recessions without direct intervention. The proposed AI tools align with this philosophy, providing immediate feedback and assistance without requiring extensive coordination. In stark contrast, MacAskill's treaty proposal is more about theoretical discussions than actionable solutions, complicating the path to effective governance in the rapidly evolving AI landscape.