In 1994, Pizza Hut faced skepticism when launching PizzaNet, an early online ordering system. Customers hesitated to enter credit card details due to concerns about security. Today, we comfortably shop online, thanks to a robust infrastructure that safeguards transactions. We find ourselves at a similar crossroads with AI agents, where trust and security are paramount. The recent Clawdbot hack highlighted these risks, transforming a helpful AI into a harmful tool due to a supply chain attack. This incident underscores the urgent need to establish a secure framework for AI agents. The article outlines several essential layers for achieving security in AI, paralleling the evolution of e-commerce. Key components include supply chain security, prompt injection defense, and identity management. Each layer addresses specific vulnerabilities, like ensuring packages disclose what they do or implementing scoped permissions to limit access. Other important aspects are the ability to reverse actions, maintain audit trails for accountability, and contain breaches to minimize damage. Reputation systems for AI tools also matter, as they can help establish trust, although the Clawdbot incident revealed how easily that trust can be manipulated. The author emphasizes the need for a comprehensive security strategy, akin to the layered approach that made online commerce safe. Fast-moving solutions like better AI models must be supported by slower, foundational changes such as industry standards and legal frameworks. As companies work to build these layers—identity verification, auditing, containment—they are not just enhancing security but also creating significant market opportunities. Just as companies like PayPal and Stripe facilitated safe online transactions, similar innovations in AI could create substantial value in the emerging landscape.