People who follow AI news closely and tend to judge strength by benchmarks and leaderboards; people who want to step back and understand AI trends through the lens of industry structure.
A thought experiment that reframes the landscape: think of model companies as generals, and companies that control access points as the king dispatching orders from the rear. See where that logic leads.
One analytical frame: models may become commoditized; access points may become scarcer. This is not a conclusion to accept outright, but an additional angle for observing AI trends.
A Slightly Counterintuitive Idea
Right now, almost everyone watching AI is looking at the same thing: whose model is strongest. Today Claude leads by a version, tomorrow OpenAI catches up, the day after Grok drops something dramatic. The battlefield is thick with smoke, and everyone is competing on intelligence, benchmark scores, and parameter counts.
Watching all this, I kept running into an idea I am still working to verify. The companies fighting hardest in the open field are Claude and OpenAI. But there is something worth sitting with: kings, as a rule, do not fight on the battlefield themselves.
The king stays in the rear. He does not swing a sword himself. He commands generals, decides which battle each one is sent to, and assigns each task accordingly. Following that analogy forward, Apple and Google, two companies whose models are often criticized today, may actually occupy something closer to the king's position. This is one reading. I am laying out the reasoning so you can stress-test it yourself.
Generals on the Field, King Dispatching from the Rear
Here is the division of labor I see. What model companies are doing right now looks more like being a general. Claude excels at one kind of battle, OpenAI at another, Grok at yet another. Each is pushing to the extreme within its lane. This is critically important work, and it burns enormous capital. I am not dismissing any of these generals.
But even the strongest general needs someone to decide when he steps onto the field. That decision-maker occupies something closer to the king's position. And what determines who dispatches whom ultimately depends on what kind of user you are.
Type One: Power Users
Creators, engineers, people who rely on AI professionally
- Prefer to specify the model themselves
- Call Claude for document reasoning
- Hand automation workflows to ChatGPT
- Use Grok to read international sentiment
- They are their own task dispatcher
Type Two: General Consumers
Have no interest in knowing which company is running in the background
- The whole division of labor is too complex
- Do not care about document reasoning or automation
- Want something simple enough to say in one sentence
- "Siri, do this for me."
- Does not matter who handles it, as long as there is one access point
I am a Type One person. Dispatching tasks myself is both a pleasure and a professional skill for me. But the vast majority of the market is Type Two. They do not care which AI company is doing the work under the hood. They just want to face one access point and say "handle this."
If That Is True, the Access Point Becomes Critical
Following the needs of Type Two users, something like Siri has always been about one thing: being that single access point. It stands in front of all the generals, catches the user's "handle this," and routes the task downward. Those who excel at fighting go fight. Siri does the dispatching. It does not necessarily need to be smarter than Claude or OpenAI.
If this path holds, what Apple needs to win is not the model itself but the position of access point for all AI. Access points come in roughly two forms.
Apple looks more like a hardware access point. However capable AI becomes, you always need a physical device to reach it. The iPhone, the Mac. That is the gate Apple guards. Google looks more like a cloud access point. Almost everything involving search, data, or anything that requires an internet connection flows through that layer. This is why I bring up both Apple and Google together: their models are frequently criticized right now, but each guards a gate that is very hard for others to bypass.
Why Apple May Be Able to Hold That Gate
Saying "the access point matters" is not enough. The real question is what makes it defensible. Here is an observation I find credible.
Apple is a peculiar company. It is the one firm that emerged from free-market capitalism with the most distinctly centrally planned characteristics. A typical Windows PC is a free market: the GPU comes from one company, the CPU from another, the RAM from yet another. Each makes its own component as strong as possible, then they are assembled. The upside is freedom. The cost is that every additional connection point is another potential failure point.
Apple works the opposite way. It brings hardware, operating system, chip, thermal design, and privacy under one roof, converging everything toward a single experience. This is extraordinarily difficult, difficult enough that most companies cannot do it. So you see an interesting pattern: many companies study Apple's products, its design, its website, yet struggle to replicate Apple as a company, because its real capability is that central coordination capacity. That capacity may be amplified in the AI era. Here are three concrete reasons.
The more useful AI becomes, the more it needs to know about you: your emails, documents, calendar, photos. These happen to be the most sensitive data. Reasoning that involves personal data naturally belongs on-device or in architectures you can trust. In the AI era, privacy has graduated from a brand promise to a product capability.
Once AI enters a workflow, the chain is long. Any unstable link in the middle and the whole thing collapses. Apple's hardware and software are co-designed, absorbing complexity into the system so that the tools simply work when you use them. That stability is exactly what you need when weaving AI into everyday life.
People used to laugh at Apple's unified memory pricing. But running large models locally changes that calculation: even a flagship GPU only has 32 GB of VRAM, whereas Apple's unified memory doubles as GPU memory, which is actually efficient for local inference. Apple is betting on personal devices and on-device reasoning.
None of this guarantees Apple wins. You could argue Apple's privacy track record has gaps, and pointing to a company that clearly does better is not easy either. What I am saying is that the cards in Apple's hand may carry different weight in the AI era than they did before.
Models May Get Cheaper; Access Points May Get Scarcer
Putting this together, here is my tentative conclusion. Feel free to push back on it.
Models themselves may grow increasingly commoditized. They are infrastructure, capital-intensive by nature. Lead by one version today and someone catches up tomorrow through any number of routes. For general consumers, when the gap in model capability narrows, they mostly do not care which underlying company is being called. Claude unavailable? Switch to GPT. GPT unavailable? Switch to Gemini. They all work well enough for a conversation.
If models commoditize, what becomes relatively scarce is the access point. When every model is fighting to get into your daily life, because there is no other way to monetize without entering it, whoever controls the device you open every day does not necessarily need to build the strongest model. They just need to select models: route complex tasks to cloud models, handle simpler things locally, and control the interface. That is enough.
What I Actually Want to Say
The most visible figure on a battlefield is always the general fighting hardest in the open. But what decides the outcome is sometimes the king who never steps onto the field and dispatches from the rear. Apple and Google's models draw criticism right now, and I am not going to dispute that. But each guards a gate. When there are so many AI options that it becomes overwhelming, what you actually need may be very simple: one "handle this," and someone in the background who routes the right task to the right agent.
This article is not asking you to believe Apple will win. I just want to remind myself, and you, to look at more than benchmark scores when reading AI development. Model strength is one track. Access points, privacy, stability, and who controls the device you pick up every day are several other tracks. Reading them together gives you a more three-dimensional picture of where things are heading.
Two free talks every month.
Join the LINE Community ↗