Apple posted a job listing last Monday for a “Senior Engineering Manager, Agentic AI – Proactive” and quietly pulled it by Thursday. The listing described a team building “general-purpose computer-use agents” that can browse the web, write code, control apps, navigate devices, and coordinate specialized subagents. Bloomberg reported the story on November 19. The job posting is gone, but the signal is loud: Apple is building its own computer-use agent, targeting a 2027 release.
The opening line of the job description: “Imagine an AI agent that can use a computer the way a capable person does.” That is Apple going on record with an ambition that puts it squarely in competition with Claude Computer Use, OpenAI’s ChatGPT Agent, and Google’s Gemini browser automation — except on Apple hardware, with Apple’s privacy architecture underneath it.
What the Agent Can Do
The job listing is unusually specific about technical scope. The agent Apple is building will plan and carry out multistep tasks across apps, browsers, command lines, code, and APIs. It can coordinate specialized subagents along the way. The engineering manager they were seeking needed hands-on experience with LLM-based agents covering tool use, planning, long-horizon task execution, and multi-agent systems — plus experience with foundation model evaluation and post-training.
This is not a Siri upgrade. The existing Siri AI, which launched in September, works through App Intents: structured, semantic hooks that app developers implement explicitly. The new agent is different. It operates like Claude Computer Use — it can work on anything visible on screen, with or without developer cooperation. App Intents remain the preferred path, but the new agent closes the gap for apps that never implement them.
What This Means for Developers Right Now
Apple will likely ship before you’re ready if you wait for the agent to arrive before thinking about it. The window to act is now.
In iOS 27, SiriKit was deprecated. App Intents are the only way Siri can call into a third-party app. When Apple’s computer-use agent ships, apps with well-implemented App Intents will almost certainly get better results — faster, more reliable task execution — compared to apps the agent has to infer through screenshots. This is not speculation; it mirrors the pattern of every agentic platform: structured input outperforms visual inference.
The minimum you should do today: expose your app’s core actions as App Intents. Keep them composable and focused on single tasks.
struct SearchProductsIntent: AppIntent {
static var title: LocalizedStringResource = "Search Products"
@Parameter(title: "Query")
var query: String
func perform() async throws -> some ReturnsValue<[Product]> {
let results = await ProductCatalog.shared.search(query: query)
return .result(value: results)
}
}
The agent you’re building for doesn’t exist yet, but the framework it will use does. MCP server support is already rolling out in macOS betas. Apple is opening agent surfaces faster than it historically has.
Apple’s Actual Advantage
It is not model capability — frontier models from Anthropic and OpenAI will stay ahead for a while. Apple’s real advantage is the trust architecture.
Claude Computer Use hits 72.5% on OSWorld benchmarks, approaching the human baseline of around 72%. OpenAI’s ChatGPT Agent operates primarily in the cloud. Both send your screen state to a server. Apple has Private Cloud Compute for tasks that require server-side processing, but the company’s privacy commitment and the Apple Silicon Neural Engine make on-device inference viable for tasks that shouldn’t leave your machine — financial data, medical records, proprietary code.
The job listing is explicit: “safety, privacy, and security need to be central to the design.” For enterprise customers and privacy-conscious developers, that is a credible differentiator, not a marketing line.
The Open Question
Will Apple provide a developer API for computer-use automation? Face ID and the Secure Enclave are still closed. But the Foundation Models API, App Intents expansion, and MCP support in OS betas all suggest Apple is in an opening phase with its AI infrastructure. The most likely outcome: developers get structured access through App Intents and MCP, and the raw pixel-level computer-use layer stays internal — consistent with how Apple handles all hardware-adjacent capabilities.
The 2027 timeline should be treated as aspirational. Apple has slipped AI features before — several iOS 27 Siri capabilities missed their original dates. But the team is real, the technical scope is clear, and the preparation work is the same regardless of when it ships. MacRumors has the full job listing breakdown if you want to read the original requirements. Start on App Intents today.

