Apple CEO John Ternus urges software chief to accelerate AI

The meeting was private, and the instruction was blunt. According to Bloomberg's Mark Gurman, Apple chief executive John Ternus sat down with software chief Craig Federighi in recent weeks. The instruction was that Apple needs to accelerate its software road map. The phrase Gurman used is that Apple must move quickly beyond merely making the necessary improvements to Siri.

That sentence is the whole story in miniature. Fixing Siri is the minimum. John Ternus is asking for something harder. He wants a software organisation that ships artificial intelligence on the cadence of a company that intends to stay at the centre of consumer computing.

Apple has not commented on the report, which rests on Gurman's sources and appeared in his Power On newsletter on Sunday. That makes it a reported account rather than a confirmed plan. It is still the clearest signal yet of how the new chief executive intends to govern, and of where he thinks Apple has been too slow.

The distance between those two readings is wider than it sounds. A chief executive who asks for repairs is managing a product. A chief executive who asks for a new road map is managing a culture, and John Ternus has been in the job for just over two months.

What Gurman reported, and what it leaves open

The report describes a closed-door meeting between John Ternus and Craig Federighi in which the chief executive pressed for speed. Gurman also wrote that Apple could make Siri work more like an agent as early as next year. He said the company ultimately wants to go in that direction, and that the exact implementation has not been settled.

An agent, in this context, is an assistant that pursues a goal across applications rather than waiting for a step by step instruction. Ask it to book a table, move a meeting and message the people involved, and it does the sequence itself, including the parts that require access to other apps. Gurman pointed to work along those lines at OpenAI, Meta and xAI as the competitive reference.

Nothing in the reporting says Apple has a finished product. It says the chief executive has told his software leader that the current pace is not enough, which is a statement about management rather than about features. The instruction John Ternus gave is narrower than a product plan and broader than a bug list, which is why it landed as a directive rather than a deadline.

Why John Ternus wants more than Siri fixes

Siri is the most visible symptom of Apple's artificial intelligence problem, and it is also the easiest one to describe. The assistant was early to the market and then fell behind, and the promised personalised version slipped repeatedly, a delay that 9to5Mac has described as an embarrassing two year gap.

Fixing that is a defined engineering task with a defined end state. Moving beyond it is not. It requires Apple to decide what its assistant is for and how much system access it is willing to grant. It also requires a judgment about how much of the work happens on the device rather than in a data centre. Those are product and policy decisions as much as technical ones. They are also the reason the instruction from John Ternus lands on the software organisation rather than on any single team.

Apple's advantage in this race has never been the largest model. It is the integration: the operating system, the silicon, the privacy posture and the install base of more than two billion active devices. An agent that can act inside that stack is more useful than a chatbot that answers questions beside it. It is also riskier, because an assistant that takes actions can take wrong ones at scale.

Competitors have already moved into Apple's own ecosystem. ChatGPT plugs into iCloud mail and calendar, which means the assistant layer is being contested on Apple's home ground by companies that do not have to ship an operating system to reach Apple's customers.

That calculation is the one John Ternus appears to be making. An integrated agent inside iOS is the only version of this product Apple can ship without conceding the layer to somebody else. It is also the version that requires the most coordination between software, silicon and privacy teams.

The executive who has to deliver

Craig Federighi is Apple's senior vice president of software engineering, and since the departure of John Giannandrea he has also carried oversight of artificial intelligence. Apple made that arrangement explicit in December 2025, when it named Amar Subramanya as vice president of artificial intelligence reporting to Federighi. Subramanya is responsible for foundation models, machine learning research, and AI safety and evaluation.

Subramanya came from Microsoft and had spent sixteen years at Google, where he was head of engineering for the Gemini Assistant. That hire was designed to give Federighi a deeper bench. The reporting since then suggests the pressure has not eased.

Gurman describes Federighi as hands-on in meetings and major decisions, willing to argue a technical position hard, and possessed of enough command of detail to win those arguments. The same account says some software and design staff feel they have little influence over product direction as a result. That is a familiar description of a strong operator, and it is also the description of a bottleneck.

The detail that matters most for John Ternus is the planning horizon. Federighi reportedly works to a twelve to eighteen month frame, while hardware engineering, where Ternus spent his entire career, works to multi-year road maps. A chief executive who was trained to commit to a date years ahead is now asking a software leader who plans in quarters to compress that cycle. Neither approach is wrong on its own terms. They are simply different clocks.

Gurman also reported that there is no sign Federighi is leaving and that he has made succession planning a low priority. For a company that has just changed chief executive, that is a governance question as much as a personnel one.

For John Ternus the risk is asymmetric. If the software organisation speeds up, the credit goes to the chief executive who asked for it. If it does not, he has spent his first year pressing a leader he cannot easily replace, and the 2027 cycle becomes a verdict on both of them.

Rockwell, and the risk of losing the Siri team

The report adds a second name that matters, and gives John Ternus a personnel risk he did not create. Mike Rockwell, who built the Vision Pro and now leads Siri engineering, is reportedly still seriously considering stepping down, possibly next year.

If that happens, it would remove the person most closely associated with bringing the reworked Siri to market at the moment the chief executive is demanding that it arrive faster. Apple has not said anything about Rockwell's plans, and a decision to leave is not the same as a decision to leave badly. But a demanding deadline and an uncertain leadership bench are a difficult combination in any engineering organisation, and they are worse in one that has already been through a reorganisation.

This is the part of the story that the phrase accelerate tends to hide. Speed in software comes from stable teams that know the code, and reorganisation is the fastest way to lose that knowledge. Apple has not commented on Rockwell's plans, and John Ternus has not addressed the report in public.

The other lever: fewer managers, more shipping

The artificial intelligence instruction is only half of what John Ternus has been doing since he took over on 1 September. Since then he has moved on three fronts at once: the software road map, the management structure and the product calendar.

Bloomberg reported in late September that he was weighing an engineering led overhaul aimed at a faster, leaner company, with more frequent product launches and fewer management layers. The cuts have since become concrete, in an account summarised by Gizmodo. More than 200 positions went last month across teams working on Siri, the Vision Pro, and artificial intelligence features for Apple devices. In the past two weeks the hardware engineering division has begun cutting engineering program management roles, affecting several managers including roughly half a dozen directors. Affected employees were given a few weeks to find another position inside the company.

The organisational logic is coherent. Apple has long run on two launch moments a year, spring and autumn, and a company that wants to ship intelligence features continuously cannot put them all through two doors. Relying less on those cycles, and removing a layer of programme managers between decision and delivery, is how a hardware company becomes a faster software company.

The risk is just as coherent. Programme managers are the people who know why a schedule slipped the last time, and cutting them while asking for acceleration removes institutional memory at the moment it is most useful. Some employees reportedly believe more reductions are coming this year or early next.

The clock John Ternus brought with him

John Ternus joined Apple's product design team in 2001, became a vice president of hardware engineering in 2013, and joined the executive team as senior vice president of hardware engineering in 2021. He studied mechanical engineering at the University of Pennsylvania and worked as a mechanical engineer before Apple. He has spent twenty five years in a company that measures programmes in years, and he has said that he worked under Steve Jobs and counts Tim Cook as a mentor.

That background explains the instruction to Federighi better than any strategy document. An engineer who has shipped hardware generations on fixed dates tends to treat a road map as a commitment. He also tends to view a slipped software feature as a schedule failure rather than an open research question. Whether that instinct produces better software or simply more pressure is the central question of his tenure so far.

What John Ternus inherited from fifteen years of Tim Cook

The comparison is unavoidable, and it cuts both ways. Tim Cook became chief executive in 2011, when Apple was worth roughly $350 billion. He left the role with the company valued at about $4 trillion, annual revenue above $416 billion, and a services business larger than most Fortune 100 companies.

He also left the artificial intelligence question open. Apple Intelligence arrived late, Siri's promised upgrade slipped, and the company spent 2025 reorganising the teams responsible. Apple announced the handover in April 2026, and Cook moved to executive chairman on 1 September 2026, with responsibilities that Apple described as including engaging with policymakers around the world. Arthur Levinson became lead independent director.

That arrangement means John Ternus has a predecessor who remains in the building, on the board, and in the public conversation. It is a stable handover by design. It also means the new chief executive has less room to blame inherited problems for long, because the person who created them is now chairman.

What has already shipped, and what is coming

John Ternus has not been idle in public. His first major appearance as chief executive came on 9 September, when Apple introduced the iPhone 18 Pro, the iPhone 18 Pro Max, and the iPhone Duo, its first foldable phone.

The pipeline behind those launches is unusually dense. Reports point to camera equipped AirPods, Apple's first smart glasses, and a special iPhone for the phone's twentieth anniversary. On the computer side, Apple expects to double MacBook Neo sales in 2027, and it has been spreading its launches across more events rather than concentrating them.

A full hardware calendar is not the same as artificial intelligence leadership, and John Ternus appears to know it. The point of a denser schedule is to create more moments where intelligence features can reach customers, which is why the software demand and the hardware plan are the same argument.

What would actually count as success

Apple does not publish an artificial intelligence road map, so the test has to be inferred from what the company ships. John Ternus has set three visible ones, whether or not he would describe them that way.

The first is cadence. Features that arrive with point releases, rather than waiting for the single autumn launch, would show that the reorganisation changed something real. A company that ships intelligence features in October and again in February is telling its customers that the assistant is alive. That is an argument John Ternus can make without launching a new phone. The second is capability. An assistant that completes a task across two or three applications without being scripted step by step is a different product from one that answers a question and stops. The third is retention, which is hardest to measure and easiest to feel, because an assistant that fails twice is one that people stop asking.

By that standard, John Ternus is not asking Craig Federighi for a demonstration. He is asking for a shipping habit, which is a slower and less photogenic thing to build. It is the kind of habit John Ternus built in hardware, where annual chip generations became routine, and the kind he is now asking the software organisation to acquire.

There is a counterargument worth taking seriously. An agent that acts on a person's behalf can fail in ways a search result cannot, and Apple's brand rests on not putting its customers in that position. A cautious assistant that declines half its tasks is a poor product, and a reckless one that misfires is a worse brand risk. The pace John Ternus is asking for has to be achieved without moving the reliability line, which is precisely why the demand is difficult rather than unreasonable.

The case for patience, and the case against

The case for patience is that Apple rarely wins the first version of anything. It won music players, phones and tablets by arriving after the pioneers and integrating better. John Ternus inherited that position intact, along with the silicon, the operating system, the privacy framework and the distribution that make an agent useful in ways a standalone chatbot cannot be.

The case against is that this cycle is different in one important respect. Previous platform shifts rewarded integration at the point of sale. Agents reward trust in execution, and they improve with usage and with model capability, both of which compound for whoever ships first and iterates fastest. A slow, careful agent that refuses half the tasks it is given will not train its users to rely on it, and the competitor that does will set the expectation.

There is a middle path, and the reporting suggests Apple is trying to find it. Running a development pipeline that gives engineers more time to code and explore, and less time on integration, is a bet that better tooling can buy speed without buying risk. Apple is reportedly testing exactly that, with a target of fifty percent more coding and exploration time, on the argument that coding agents have made integration less of the job.

What to watch

• The 2027 software cycle, which is the earliest point at which an agentic Siri could appear, on Gurman's estimate.
• Mike Rockwell's decision, which would tell you more about the health of the Siri team than any road map.
• The next round of job cuts, and whether they stay concentrated in management and program management or reach engineering roles.
• Whether Federighi's organisation changes structure or merely pace, since the two produce very different results.
• Whether John Ternus addresses any of this in public before the 2027 cycle, or leaves the reporting to stand as the record.
• Any public comment from Apple, which has so far said nothing about the report.

Questions readers are asking

Is John Ternus replacing Craig Federighi?

No. Gurman reported that there is no sign Federighi is leaving, and Apple has made no announcement. John Ternus has not signalled any change to his leadership team, and the report describes pressure on the software organisation rather than a change at the top of it.

Who is running Apple now?

John Ternus has been chief executive since 1 September 2026. Tim Cook moved to executive chairman of the board, and Arthur Levinson became lead independent director. Craig Federighi remains senior vice president of software engineering and oversees artificial intelligence.

What does an agentic Siri actually change?

It moves the assistant from answering questions to completing tasks across applications. That requires deeper system access, clearer permissions, and a much higher standard of reliability, because a wrong answer is an inconvenience while a wrong action is a problem. For John Ternus, that is the part of the plan with the most downside and the least visible upside.

Does this affect iPhone buyers?

Not immediately. The hardware schedule for the next year is already set, including the iPhone 18 line and the iPhone Duo. The part that could change is how quickly new intelligence features reach existing devices, and how many of them require the newest chips.

The bottom line

The instruction that John Ternus gave Craig Federighi is easy to report and hard to execute, because it asks a hardware company to run software at a speed its culture has never needed. The layoffs, the product calendar and the agentic Siri plan are all the same message delivered through different channels.

What makes this more than a management story is the timing. Apple is worth about $4 trillion, which means the market is already paying for artificial intelligence leadership it has not seen. John Ternus has roughly a year before the 2027 software cycle answers the question the report raises, and the person he has asked to answer it plans in quarters, not years. Whether the chief executive gets the cadence he wants is now the most consequential open question in consumer technology.