Apple to Replace Core ML with 'Core AI'
Apple is reportedly planning to replace its long-standing Core ML framework with a new 'Core AI' platform at WWDC 2026. The move signals a major strategic shift toward more advanced, on-device AI and ML capabilities, creating a new ecosystem for developers building for iPhones and other Apple hardware.
Apple's original Core ML, introduced in 2017, prioritized on-device inference to ensure user privacy and offline functionality. The framework was designed to optimize performance by leveraging the CPU, GPU, and the dedicated Neural Engine, setting a precedent for Apple's edge-first AI strategy. A significant evolution came with Core ML 3 in 2019, which enabled on-device model *updates* for the first time. This allowed apps to personalize machine learning models based on a user's local data, without that sensitive information ever needing to leave the device for cloud-based processing. While Apple focused on this private, on-device approach, the broader industry raced towards massive, cloud-based large language models. This led to criticism that Apple was falling behind, with its AI features seen as less complex and powerful than what competitors offered through cloud APIs. The shift to 'Core AI' appears to be Apple's answer, likely expanding on its new Foundation Models framework which gives developers direct access to on-device LLMs. This strategy maintains the privacy-first ethos while aiming to provide more powerful, generative capabilities that can be integrated with just a few lines of Swift code. For ML engineers, this trend underscores the critical need for production-oriented skills. Top tech companies prioritize engineers who can design, deploy, and maintain ML systems, focusing on MLOps, data engineering, and creating production-ready code, not just experimental models in notebooks. To build a standout portfolio, engineers should explore projects that utilize the modern AI stack, such as Retrieval-Augmented Generation (RAG) with vector databases. This approach, which gives models access to external knowledge, directly addresses the limitations of smaller, on-device models and is a highly sought-after skill. Preparing for ML system design interviews now requires a deep understanding of edge-versus-cloud trade-offs. Candidates should be ready to discuss model quantization, latency optimization, and managing the power consumption of continuous inference on mobile hardware.
