Apple is developing an AI agent and is taking on Meta and OpenAI
Apple wants an AI agent that controls your Mac like a human, but in the meantime, it is making ChatGPT a lot more difficult. Is Apple slowing down its rivals?
Apple is developing its own AI agent that independently performs tasks on its devices, from operating apps to writing code. This is evident from job vacancies that the iPhone maker posted online this week. Apple is thus entering a race late, in which Meta, OpenAI and Anthropic are already far ahead.
In one of its job postings, Apple is seeking a senior engineering manager for a proactive AI assistant. This person will be responsible for searching the web, operating apps, writing code, and navigating devices independently. “Imagine an AI agent that can use a computer like a capable person does,” Apple writes.
An AI agent therefore goes a step further than a chatbot. While a chatbot primarily provides answers, an agent carries out the task itself: it finds a restaurant and then also makes the reservation. According to the job posting, Apple's agent must plan tasks in several steps and, in doing so, manage smaller helper agents.
The job listing also mentions computer use, a technology that allows AI itself to operate a computer's mouse and keyboard. Such an agent clicks through the macOS menus as if a human were sitting behind the screen.
Bloomberg reported earlier this month that Apple was working on such a system. A launch could take place at the earliest at the end of next year, although development is still in its infancy. Meanwhile, the new CEO, John Ternus, is putting pressure on software head Craig Federighi to start working with AI more quickly and boldly.
Apple itself emphasizes security. “Agents that can execute code, use login credentials, and act on behalf of a user require robust permission models, sandboxing, and transparency,” the job posting states. Sandboxing confines software to a shielded environment, preventing it from accessing the rest of the system.
Apple is entering a rapidly growing market. In September, Meta launched Muse, an agent that makes restaurant reservations, books travel and compiles news summaries. The app has long been ranked second among free apps in the App Store, just behind TikTok.
The rest of the industry is also accelerating. OpenAI expanded the capabilities of its agent Dots, while Anthropic is making its agent Duet available to a larger group of testers. This week, Google updated its Tasks agent, which, like Elon Musk's Grok Bot, can make calls on behalf of users.
Yesterday, Google's Gemini gained another feature, as the chatbot can now control a Mac on its own. Agents have long been established among programmers, with tools such as OpenAI's Codex and Anthropic's Claude Code that write software independently. ChatGPT and Claude are also in the App Store's top five.
The news about Apple immediately pushed down the share prices of its competitors yesterday. Meta Platforms fell by almost 2 percent in a short period of time, while Anthropic lost more than 3 percent.
This is not entirely Apple's fault, as stock markets in the United States, Europe and Asia are generally under considerable pressure. AI stocks, in particular, are taking a heavy hit.
At the same time, Apple is making things more difficult for its rivals. In the test version of macOS 27.3, ChatGPT users must complete 14 steps to allow a remote agent to control their Mac, compared to 4 previously. That is 3.5 times as many steps.
Apple describes the new warning screens as a means of protecting customers, but some AI companies and users see them primarily as a hindrance to competition. This is a sensitive issue, as the U.S. Department of Justice has already sued Apple for blocking access to features and hardware.
For now, Apple is taking smaller steps. With iOS 27.3, Siri can work more closely with apps from other companies, for example, to book a taxi or a restaurant. We will have to wait at least another year for an agent that independently handles complex tasks across multiple apps.