Imagine asking your AI assistant to book a flight, schedule a meeting, or even create a playlist—all using the apps on your phone. While this sounds like the future, how close are we really? Researchers have developed ShortcutsBench, a groundbreaking benchmark that tests an AI's ability to use real-world APIs, the building blocks of your favorite apps. Instead of focusing on theoretical tasks, ShortcutsBench uses actual APIs from Apple's operating systems and real user requests taken from the Shortcuts app. The results? While AI has made impressive strides, it still struggles with complex, multi-step actions. Current AI agents, even those powered by cutting-edge large language models (LLMs) like Gemini and GPT, excel at simple tasks like "Check the weather and tell me." However, they stumble when faced with scenarios requiring multiple app interactions or intricate parameter settings. The research shows that the biggest hurdle for AI isn't selecting the right app, but rather figuring out *how* to use it. Specifically, extracting the right information from your request and plugging it into the correct parameters within the API call proves surprisingly difficult. Another significant challenge is the AI’s awareness of missing information. Often, a user request doesn't provide every detail an app needs to function. For instance, asking AI to "book a table" requires it to realize it needs to know *where*, *when*, and for *how many people*. The current generation of AI often overlooks these implicit requirements. ShortcutsBench’s innovative design, using real-world APIs and genuine user queries, gives us a far clearer picture of AI's true capabilities. While there’s still work to be done, this research paves the way for truly helpful, app-savvy AI assistants in the not-so-distant future.