Here’s What It Could Do That Siri Can’t Do Today
Your Train Is Cancelled. Which App Do You Open First?
The ticket is in your email. The meeting you might miss is in your calendar. The person waiting for you is one message away, but you need a new arrival time before you know what to say. You have one problem. Your phone makes you handle it in pieces.
Apple is building the foundations for an AI agent that could take on this kind of work. A job listing posted on 17 November 2026 describes software that could choose its own tools, work across apps, browse the web and write or debug code. In the train example, you might ask a single question: “Can I still make my meeting?” The agent could find the booking, check another service against your diary and prepare a message with a revised arrival time.
That journey is an illustration. Apple has described the engineering project and already opened a developer preview for deeper Siri integrations. As of 20 November, it has yet to announce a release date or supported devices for the broader agent. The interesting detail in its plans is how much of a task Apple thinks a computer could carry from one app to the next.
What Apple Is Actually Building
Apple’s recruitment material describes a team working on “general-purpose computer-use agents”. The wording matters because the proposed system has a choice about how to get something done. It might ask an app for information directly, navigate a website or run a command. Its job would include deciding which approach fits each step.
Take that question about the meeting. A calendar can supply a start time. A train operator can supply an expected arrival. Neither fact answers the question until someone checks which station the train reaches and how far it is from the meeting. Today, that someone is usually the passenger, standing on a platform and switching between screens. Doing that comparison on your behalf is where an AI agent could earn its place.
The job description also explains why this story extends beyond the iPhone. Apple wants the team to work with browsers, code editors and command lines. A developer could, in principle, ask an agent to investigate why a recent change broke a piece of software. It might inspect the relevant files, make a proposed correction and run a test. The useful result would be a change the developer could inspect, together with the outcome of the test. Apple’s listing sets that kind of work within the project’s scope. How reliably its software can do it remains an open question.
What Siri Can Already Do Across Apps
There is a nearer-term change with a firmer date. On 3 November, Apple opened a Siri AI developer preview in the 27.3 betas of iOS, iPadOS, macOS and visionOS. Any developer can use the preview to integrate an app with Siri AI. Apple says it has already started rolling out new capabilities in its own apps, including Mail, Maps, Calendar, Photos and Files.
One of Apple’s examples is finding a coffee shop you visited last week. That is a small but recognisable improvement. You might remember the visit perfectly well and still have no idea whether the name is in a message, a map search or a diary entry. Another example Apple gives is archiving files you have not opened for more than a year. Here, the assistant can act on a condition that would be tedious to check file by file.
Opening the preview to all developers gives the train example one important qualification. Your preferred travel app has to make useful information or actions available for Siri to use through that integration. An invitation to developers tells you who can build support. It tells you little about what the particular app on your phone supports today. Apple’s separate work on an agent that can navigate software itself is significant because it explores another way of reaching the information.
Who Could Feel the Biggest Difference?
Consider a parent reading a school email on the way to work. The class trip has moved from Wednesday to Thursday. There is already a calendar entry, a reminder to pack lunch and a message thread with the grandparent doing the pickup. Adding another event is easy. Remembering that the old reminder and pickup arrangement also need attention is the real work.
With permission to consult those apps, an agent could find the existing arrangements and ask one useful question: “Does Thursday’s pickup still work for Nan?” It could show the proposed calendar change and draft a message in the right thread. Even if the parent chose to send it personally, they would have one less loose end to remember that evening.
For someone who finds precise touchscreen movements difficult, the gain could be more personal. Imagine receiving a venue address as a screenshot and wanting directions to the entrance shown in it. Selecting small text, copying part of an address and moving between apps can turn an ordinary request into a task that needs another person’s help. Being able to say “Take me to the entrance in this picture” and hear the location read back could make that journey easier to manage independently. The read-back matters as much as the voice command, especially when a large venue has several entrances.
There is a quieter benefit at work, too. Someone preparing expenses might ask for the receipts from a particular trip. A useful agent could collect the documents and flag the hotel email that contains only a booking estimate. That missing receipt is precisely what the person needs to know about before submitting the claim. A polished total with the estimate silently included would leave them worse off.
When Should the Agent Stop and Ask?
Return to the passenger on the platform. The next train gets in on time, but the ticket is valid only on the cancelled service. The agent has answered the travel question and uncovered a spending decision. Should it buy a replacement, wait for the operator to confirm that the original ticket will be accepted, or tell the organiser you may be late?
This is where the design of a single confirmation could determine whether people trust the feature. “Continue?” tells the passenger almost nothing. “Buy a £46 ticket for the 10.42 to Euston?” gives them a decision they can actually make. The same principle already matters when people approve a purchase through an AI assistant.
Privacy raises a different question. Your calendar might say the meeting is a job interview. The train operator needs your route and ticket details. It has no reason to know whom you are meeting. An assistant with permission to read both records would still need to keep the appropriate details on each side. Apple’s recruitment material identifies permissions, privacy and visibility into the agent’s actions as engineering priorities. The eventual product will have to show how those choices work in practice.
What Comes Next for Apple’s AI Agent?
The first change to watch is which everyday apps add useful Siri actions through Apple’s developer preview. That could make a difference well before a general-purpose agent arrives. If the app you rely on can retrieve the right booking or update an existing reservation, one awkward part of a familiar job gets easier.
The bigger milestone would be a public demonstration in which something goes wrong halfway through. Suppose the agent finds a replacement train and drafts your message, but the booking fails. Does it recognise that you still have no ticket? Does the draft still say you will arrive at eleven? Can you see exactly where it stopped?
A successful booking would show that Apple’s agent can use several apps. Catching that unsent message, and telling you in time to do something about it, would show that it understands why you asked.
Sources
Apple Jobs — Senior Engineering Manager, Agentic AI – Proactive — Apple’s description of the team, the tools its agents would use and the role of permissions and privacy. Apple 工作 (香港)
Apple Developer — Siri AI is now open to all developers — The 3 November developer preview, supported beta platforms and Apple’s examples of Siri actions. Apple Developer
MacRumors — Apple is Building an AI Agent That Can Use Apps and Code — Coverage of the 19 November reporting and its distinction between Apple’s hiring plans and a future Siri release. macrumors.com
Next Reads
AI Agents Explained: What They Are, How They Work and Why They Could Change Everything — How an assistant turns a request into a sequence of actions.
ChatGPT Wallet Explained: How AI Can Shop for You Without Seeing Your Card Number — What you approve when an AI assistant reaches the checkout.
OpenAI Models Told Themselves To Hide Mistakes, Newly Published Report Reveals — Why a system’s account of a failed task can matter as much as the failure itself.