Apple wants Siri to behave more like a proactive agent—closer to ChatGPT Dots than to the request-and-reply Siri most people use today. That direction comes from Mark Gurman’s Power On newsletter on November 8, 2026, summarized by outlets including 9to5Mac, 9to5Mac on Ternus and Federighi, and AppleInsider. Gurman writes that John Ternus is pressing Craig Federighi to pursue a more ambitious AI strategy. Apple has not confirmed the story; Bloomberg’s full newsletter is paywalled.
For builders on iPhone, the practical split is two timelines: near-term App Intents in iOS 27.3, and longer-horizon agent behavior Gurman says could land as early as 2027.
TL;DR — reported vs confirmed
| Question | What Gurman / relays say | What Apple has confirmed |
|---|---|---|
| Siri will pursue goals in the background | Yes — compared to Dots, Muse, Grok-style agents | No public statement on November 8, 2026 |
| Agentic Siri could ship as early as 2027 | Gurman’s belief, per relays | No |
| iOS 28 uses 6 weeks code + 2 weeks integration | Per relays citing employees | No |
| First-party App Intents widen in December 2026 | AppleInsider on Gurman | Betas already show pieces of this |
| Third-party App Intents preview spring 2027 | AppleInsider on Gurman | No |
| Ternus pressed Federighi on AI ambition | Headline theme of Bloomberg edition | No |
What “agentic Siri” means in this report
Today’s Siri AI in iOS 27 finally covers much of what Apple promised years ago: conversation, on-screen awareness, and a standalone Siri app. It is still mostly user-initiated. You ask; Siri answers or runs a command.
Gurman’s November 8 framing is about continuity of intent. Rivals have shipped assistants that:
- Hold a goal across many steps
- Decide what to do next without a fresh prompt each time
- Run work in the background and surface results later
OpenAI markets that shape with Dots; Meta with Muse; xAI with Grok Bot. Our four-way comparison of proactive agents walks through how each product handles permissions and trust. Gurman, as relayed by 9to5Mac, quotes this line of thinking:
Dots, Muse and Grok can keep track of goals, determine what needs to happen next and continue pursuing objectives without the user initiating every individual step. Apple ultimately wants to go in this direction … and could make Siri more like an agent as early as next year.
“Next year” in a November 2026 newsletter means 2027 in calendar terms. That is reporting, not a WWDC slide.
Apple has advantages if it goes there: App Intents, deep OS integration, and on-device models tied to Apple silicon. It also has constraints we have tracked all year—partner models for hard tasks, usage limits, and internal AI spending caps Gurman reported in early November. Agentic loops burn tokens and need strong guardrails; OpenAI’s own Dots unapproved-email incident showed why.
On-device assistant metaphor for Apple pushing Siri toward proactive agent behavior on iPhone
App Intents: the bridge before full agents
Gurman and AppleInsider agree on a staged rollout that looks like engineering reality, not a single flip of a switch.
iOS 27.3 and first-party apps
Developers on iOS 27.3 betas have found Siri can list what it can do inside Apple’s own apps, then execute those actions—archive mail, move messages, schedule sends, and similar tasks. AppleInsider says Apple could ship broader first-party support to the public in December 2026 with iOS 27.3.
That matters because it is the feature Apple delayed after WWDC 2024: precise voice control inside apps, not just “open Spotify.” It is still command-driven, not a background agent, but it trains users and developers on the same App Intents surface an agent would need.
We explained the delegation layer in how Siri can swap in Claude or ChatGPT on iOS 27. Agentic Siri, if it ships, would stack on top of that plumbing.
Spring 2027 and third-party apps
AppleInsider, citing Gurman, says third-party App Intents are on track for a preview in spring 2027—possibly with iOS 27.5 or 27.6, ahead of WWDC 2027.
For app teams, the checklist is familiar:
- Implement App Intents with clear, testable actions.
- Expose a “what can you do here?” surface Siri can read aloud.
- Plan for multi-step flows that stay inside Apple’s permission model.
If Gurman’s agent timeline slips, third-party Intents may still land on time as the developer-facing milestone.
iOS 28: six weeks to build, two weeks to merge
The same newsletter reportedly changes how Apple runs the OS factory.
Gurman writes that the first iOS 28 cycle gives engineers six weeks for coding and exploration and two weeks for integration—replacing a four-plus-four week split. Employees told Gurman that agentic coding tools made the old integration-heavy rhythm feel wasteful: more time writing and exploring, less time wrestling branches.
That is a process story with product consequences:
- Faster experiments in point releases if integration truly shrinks.
- More parallel AI-assisted prototypes before features harden.
- Risk if two integration weeks are tight for a platform as large as iOS.
Apple has not posted this schedule on developer.apple.com. Treat it as internal rhythm reporting until WWDC or release notes say otherwise.
Phone-sized on-device AI stack as Apple shifts iOS 28 engineering time toward coding and exploration
Ternus’s push fits the leadership story we tracked when Cook handed the CEO role to Ternus and when Ternus told Wired Apple would not ship AI for AI’s sake. A shorter integration window is one lever; a Dots-like Siri is another.
Research vs product (do not merge them)
On November 2, 2026, Apple Machine Learning Research posted work on an 8B UI navigation agent trained with online reinforcement learning on live iOS apps. We covered that paper and GiGPPO. It is evidence Apple’s researchers are serious about agents that tap and scroll.
Gurman’s November 8 piece is product politics and schedule, not a citation of that paper. Readers should keep the lanes separate until Apple links them on stage.
What developers are saying
The Hacker News thread on the Bloomberg link is thin: several commenters hit the paywall, and commenter scrlk shared an archive.is mirror in this reply. There is little hands-on technical detail yet—unsurprising for a newsletter summary. Worth revisiting if betas change Siri’s background behavior.
Privacy, permissions, and why Apple cannot copy Dots overnight
Dots-style agents ask for broad access: mail, calendars, files, purchases, sometimes browser sessions. Apple sells trust as a product. That is why AppleInsider’s summary stresses staged Intents instead of a overnight “do anything” mode.
Apple’s public pattern on agents matches what we wrote in AI agent privacy promises across Muse and Dots: vendors market convenience first; enterprises and careful users read data paths second. Federighi’s team must ship visible approval steps for anything that spends money, sends messages, or changes account settings—exactly where OpenAI tightened Dots after incidents.
Gurman’s reported six-week coding window may also be how Apple catches up on evaluation: more prototypes of permission flows, fewer rushed integrations that break in December point releases.
EU, China, and the slower markets
Apple Intelligence still rolls out unevenly worldwide. Our claimed-vs-verified piece on China in December notes features can ship without the newest Siri AI. Agentic behavior that depends on Private Cloud Compute or partner models will face the same geography gates.
If you ship globally, plan features per region and avoid assuming a U.S. beta equals a worldwide entitlement.
What this means if you build on Apple platforms
| If you… | Near term (2026–early 2027) | If agentic Siri ships |
|---|---|---|
| Ship an iOS app | Implement App Intents; test on 27.3 betas | Design idempotent actions an agent can chain |
| Use Siri + cloud models | Watch delegation and limits | Expect stricter approval UX for irreversible steps |
| Compete with Apple Intelligence | Lean on cross-platform agents today | Apple may bundle persistence with hardware trust |
Apple’s partnership pattern—Meta Muse on Mac Studio, Claude and ChatGPT behind Siri—suggests on-device orchestration plus partner models rather than a single cloud agent.
What people are asking after the newsletter
Will agentic Siri require a new subscription?
Gurman’s piece does not quote a price. Apple already signals that heavy server-side Siri work may sit behind iCloud+ or usage caps, per Ternus on the earnings call. Background agents burn compute continuously, so expect either tiering or per-task approvals that throttle cost—not a flat free upgrade.
Can Apple ship this without OpenAI or Anthropic?
Delegation code shows Apple can route hard tasks outward. Agentic Siri might still call partner models for open-ended reasoning while App Intents handle on-device actions. Our September 9 AI recap lists which features Apple framed as on-device versus cloud.
Does a six-week coding cycle mean more bugs?
Not automatically. It moves risk from integration bottlenecks to review discipline. Teams that used agentic coding to flood branches still need human release managers. Gurman’s report is about schedule shape, not quality guarantees.
Where does this leave macOS and visionOS?
This newsletter is iOS-centric. Mac power users should watch whether App Intents parity arrives on macOS Golden Gate the same quarter, and whether Visual Intelligence gains persistent tasks on AirPods camera leaks we have tracked separately.
Related reading
- Claude Duet vs ChatGPT Dots vs Meta Muse vs Grok Bot
- Apple Siri delegation on iOS 27
- Apple’s 8B iOS UI agent research
- Ternus on Siri server limits and paid access
- Internal AI spending caps at Apple
- ChatGPT Dots safety and unapproved actions
- Apple Intelligence China rollout timing
- Agent privacy promises across Muse and Dots
Specs, beta features, and Gurman-reported timelines reflect sources available as of November 8, 2026. Apple may change plans without notice.