OpenAI and Anthropic have both introduced features allowing AI to learn tasks by observing a single demonstration, signifying a move beyond solely relying on prompts. These new functionalities, Record & Replay and Record a Skill, enable users to demonstrate a workflow and have the AI create a reusable skill. This development suggests that prompting alone is insufficient for AI to reach its full potential. The author recalls working on Siri, where the primary challenge was not linguistics but context awareness and retaining user preferences across interactions. Human understanding often relies on tacit knowledge, which is difficult to articulate and rarely documented. Demonstrating a task to an AI captures nuanced details and decision-making processes that verbal prompts miss. These recorded skills can then operate autonomously, saving users from repeatedly re-explaining tasks. The future involves building libraries of these demonstrated workflows, which can be personalized and improved over time, similar to open-source software. The author emphasizes that AI learning through demonstration represents a significant leap, comparing it to training a new employee, implying a one-time effort for long-term efficiency.