Apple Is Turning Siri Into an Agent That Does the Work for You

AI Agents

Apple Is Rebuilding Siri as an Autonomous Agent — and Rewriting How It Ships iOS

What happened: Apple plans to turn Siri AI into a true “autonomous agent” that takes actions on a user’s behalf, according to Bloomberg’s Mark Gurman. The goal is to match the “do-it-for-me” rivals — OpenAI’s Dots, Meta’s Muse, and Grok — rather than just answer queries. Today’s Siri can draft messages and set reminders; the new vision would let iPhone, iPad, and Mac users hand off repeated, tedious tasks and have Siri monitor apps for changes in the background.

Why it matters: This is the clearest signal yet that “agents” are becoming the default interface for the two billion people who carry an iPhone. When Apple ships a background agent baked into its operating system, the agent pattern stops being a startup category and becomes table stakes. For anyone building AI Employees, that’s a double-edged sword: the concept gets instant mainstream legitimacy, but the bar for what an agent must do — securely, reliably, across apps — just jumped. Apple is even restructuring its engineering into eight-week “M1/M2/M3” cycles specifically to accommodate agent development, a rare admission that shipping agents is a different kind of build.

What’s next: Gurman reports the expanded capabilities could arrive “as soon as next year.” Watch for Apple to lean hard on its on-device privacy advantage — a genuine differentiator against cloud-centric rivals, and a preview of how the agent wars will actually be fought: on trust and permissions, not just capability. Nothing here is confirmed, so treat it as a strong directional signal rather than a roadmap.

Source: ETV Bharat (via Bloomberg’s Mark Gurman)

Quick hit — Hyundai is testing NVIDIA’s “sandboxed” agent runtime: Hyundai Motor Group is exploring NVIDIA’s NemoClaw blueprints and OpenShell runtime to run enterprise AI agents under policy-based controls that restrict what an agent can access and do. It’s described as an exploration, not a deployment — but it’s a real signal that governance, not capability, is becoming the gating factor for enterprise agents. Source

Quick hit — Big AI labs agree on a single “don’t train on this” signal: Google, Microsoft, OpenAI, and Anthropic will all honor one robots.txt line (`__usage_ai-training`) telling crawlers whether content can be used for training. It’s a real convenience win for webmasters — but it governs use, not crawling, doesn’t cover already-scraped content, and Meta hasn’t committed. Source

AI News

NVIDIA declares “the golden age of physical AI” in Seoul: At AI Day Seoul, NVIDIA showed Samsung’s semiconductor simulation running ~1,600× faster (18 hours to 41 seconds), LG’s CLOiD robot handling manufacturing tasks via Isaac GR00T, and a GPU–NPU partnership with SK Telecom and Rebellions for sovereign AI. The through-line: AI is moving decisively from chatbots to chips, robots, and national infrastructure. Source

Microsoft employees push back on mandatory AI adoption: A group calling itself “Concerned Microsoft Workers” circulated an open letter criticizing the company’s AI-first push — including pressure to use AI tools daily without proper training — days after Microsoft announced voluntary buyouts covering roughly 60,000 employees. The friction points to a real management question: should AI adoption be measured by tool usage or by better work? Source

NVIDIA backs four Korean AI startups: NVIDIA invested in Point2 and plans stakes in Motif Technologies, RLWRLD, and Upstage (joining Upstage’s Series C), extending its Korean footprint across AI data-center and model work. It’s another data point in NVIDIA’s quiet push to seed startups that run on its stack. Source