Apple is building an in-house AI agent to automate computer work, positioning itself against OpenAI, Anthropic, and Meta in a race to control how users delegate tasks to artificial intelligence.

A recent Apple job posting for a senior engineering manager reveals the scope: an agent capable of browsing the web, operating across applications, executing commands, writing and debugging code, and calling APIs. The agent will select the appropriate tool for each task while maintaining privacy and Apple's design standards.

The system is designed to move fluidly between tools and perform multi-step workflows across applications, browsers, command lines, and code—capabilities that mirror offerings from Anthropic's Claude and OpenAI's agents.

Apple's timeline is aggressive but uncertain. The company targets a late 2027 launch, though development remains in early stages.

Simultaneously, Apple is tightening the gates. The upcoming macOS 27.3 will add warnings that discourage users from granting full system access to third-party agents. In October, Apple told developers it would introduce additional controls requiring explicit user action before any application gains extraordinary access to a machine.

This dual move reflects Apple's core strategy: build proprietary AI capabilities while making it harder for competitors' agents to gain traction on its hardware. By controlling both the agent and the access layer, Apple can direct user workflows toward its own tools while maintaining the ecosystem lock that has long underpinned its margins.