Apple wants to turn Siri into an AI agent. That’s an assistant you give a goal to, and it keeps working on it without you directing every step. On 8th November, Apple reporter Mark Gurman wrote that John Ternus is pressing software chief Craig Federighi to take a more ambitious approach to AI, with Siri as the target. Ternus took over from Tim Cook as Apple’s CEO at the start of September.
Apple hasn’t announced any of this, and the timing is Gurman’s own estimate. He says he’s been told Apple wants to go in this direction, and he believes it could make Siri “more like an agent as early as next year”.
So what would change for you? Siri AI, the rebuilt assistant Apple started rolling out in beta on 14th September, still works one request at a time. You ask, it does the job, and it stops. That includes the update Apple opened to developers last week. As we explained on Friday, each app describes its actions to Siri, and when you ask for something, Siri picks the action that fits.
An agent works the other way round. It holds on to what you want done, works out the next step and keeps going after you’ve stopped talking to it, then comes back when it needs a decision from you. Gurman points to three agents, all launched since August, which he says “can keep track of goals, determine what needs to happen next and continue pursuing objectives without the user initiating every individual step.”
What Siri would be up against
Grok Bot came first. SpaceXSI, SpaceX’s AI business, launched it in beta on 11th August. Bots have their own computer in the cloud. Each one signs into the apps and websites you already use and only comes back when something needs your approval. You can run several Bots at once and put them in a group chat, where they pass work between themselves. Grok Bot is included with every SuperGrok plan and with the Pro, Ultra and Teams plans for Cursor, a coding tool.
Meta launched Muse on 8th September. Each Muse runs on its own virtual machine, a private computer in Meta’s cloud that holds the agent and the logins you give it. Meta calls the machine Muse Secure VM. You message Muse in the Muse app or in WhatsApp. A second agent called the Sentinel runs on the same machine, and nothing Muse does reaches the internet unless the Sentinel approves it. Muse also checks with you before it sends an email or buys something.
OpenAI’s dots arrived in ChatGPT on 29th September. A dot also has its own cloud computer and browser, connects to more than 4,000 apps through ChatGPT’s plugins, and looks for ways to help while you’re not talking to it. OpenAI calls that “proactive research” and limits it to read-only tools, so a dot can’t send messages or change anything in your apps while it’s doing it. OpenAI’s own example is an early tester whose dot noticed he’d forgotten to invoice a publication, prepared the invoice and sent it once he approved.
All three are built the same way. The agent gets a computer of its own, logs into your accounts and works through websites the way you would. That’s why they keep going after you close the app. Each one also has approval rules, which makes sense when the agent is acting inside your email, your shopping accounts and sometimes your payments.
Why Apple’s version may look different
Apple already has its own route into your apps. Siri AI reaches them through App Intents, the actions each developer builds and describes for Siri, instead of clicking through screens. We think an agentic Siri would build on that and work through the actions an app has chosen to expose, though neither Apple nor Gurman has said so. If Apple goes that way, Siri could only do as much as developers open up, and App Store apps can’t add these Siri AI actions until submissions open in the spring.
Ternus spoke about agents on Apple’s results call on 2nd November, his first as CEO. He said: “As far as agentic capabilities, you know, we think these are promising tools for users. We just want to make sure that users are aware of what they’re giving access to.” He added that Apple wants it “done in a way that protects security and privacy”. That fits what Apple said last month when it tightened Full Disk Access on the Mac: the risks grow as agents become more capable and autonomous. So we’d expect an agentic Siri to ask for permission more often than its rivals do.
Apple has also changed how it’s building iOS 28, next year’s iPhone software. Gurman wrote that for the first stage, engineers got six weeks to write code and explore features and two weeks to bring the work together, instead of the usual four weeks for each.
Apple hasn’t said which iPhones would run an agentic Siri or what it would cost. Siri AI already has daily usage limits, as we explained last month, with more access promised “for a fee in the future”. We think an agent working in the background would make far more requests than an assistant you talk to now and then, so those limits would have to allow for it.
Which of these work in Kenya
Dots are open to Kenyan ChatGPT Pro subscribers. OpenAI’s rollout to ChatGPT Pro excludes the European Economic Area, Switzerland and the UK, and Kenya is on ChatGPT’s supported list, as we found at launch. The cheapest way in is Pro 100 at USD 100 a month, or about KES 13,000 at KES 129.94 to the dollar on 9th November 2026. Meta has rolled Muse out in the US and, since 27th October, Singapore. SpaceXSI’s Grok Bot announcements don’t list countries. As for Siri AI, Apple’s page only singles out the EU and mainland China. Going by Apple’s wording, an iPhone 15 Pro or newer set to English qualifies here.
Apple is starting late. Grok Bot, Muse and dots have been out since August and September, while Siri still handles one request at a time. Ternus says Apple wants users to be aware of what they’re giving access to, and we think that’s the right priority for software that would act on your behalf. If you want an agent today, Kenyan ChatGPT Pro subscribers can set up a dot. Siri, by Gurman’s estimate, could become more like an agent as early as next year.
Join the discussion