Apple is working with startup Anthropic to develop a new platform that uses AI to automate the writing, editing, and testing of software code. It is a modified version of Xcode, the software development environment (IDE) for OS X and iOS, with the integration of Anthropic’s Claude Sonnet Large Language Model (LLM). The platform is currently being implemented exclusively within the company, and a decision on its public launch has not yet been made.
Image source: Apple
The concept is similar to those being implemented by startups Windsurf and Anysphere, creators of AI assistants such as Cursor, which are popular among developers. These tools demonstrate how LLMs can integrate into the programming environment and act as an active participant in the development process, increasing its speed and efficiency. Apple has previously refrained from using AI in software development, but advances in LLM have made it change its mind.
In 2024, the company announced Swift Assist, its own code autogeneration tool for Xcode. Originally planned for release within a year, it never reached third-party developers. Apple developers complained that the system was prone to generating buggy code, which slowed down the app development process. The partnership with Anthropic was an acknowledgement of Apple’s need for outside technical support. The two systems could be used in parallel, sources said.
Anthropic’s Claude Sonnet is one of the most respected LLMs in programming. Sources say Apple’s new system will include an interface that allows developers to formulate requests to the AI in natural language. The system will be able to generate code and perform automated testing. It can also be used to detect and fix bugs in code, tasks that would typically be time-consuming without AI assistance.
The move reflects Apple’s growing willingness to partner with third parties after struggling to build its own AI solutions. Until recently, the company had avoided using third-party AI models, with the exception of a deal with OpenAI that allows ChatGPT to process queries in Siri. Google’s Gemini AI model is expected to be integrated as an alternative this year.
The rest of the Apple Intelligence platform, including custom emoji creation, text manipulation, and notification management, is powered by AI models built in-house. According to sources, Apple has already begun rolling out the new platform to its own divisions. If the results of its internal use are positive, the company may decide to make the system available to third-party developers.