AI Killed Your SEO. That's Just the Start.
OpenAI, Google, and Apple take their current AI hegemony seriously enough that there is no relationship with them so valuable that they won’t sacrifice it. It’s not simply a fight over AI search or agents, it’s the effort to build the winning Everything Machines, similar in thinking to but carrying far greater significance than Amazon’s Everything Store.
OpenAI is betting that their stunning early market share lead, built as ruthlessly and leveraging as much hyperbole as tech has ever seen, and Operator agent services will make them the Everything Machine of choice. Google has already sacrificed their stunningly valuable SEM and SEO ecosystem to the effort; no individual customer within those could possibly count. Apple would prefer that their existing app providers fall into line very quickly with Apple Intelligence and thrive, but they will move on to your competitors fast if it benefits them. For those of us for whom iOS apps are the right distribution channel, the Apple opportunity is superior. Regardless, we either help one or more of these Kaiju build and grow their Everything Machines or lose our businesses to them – or both if we do not execute well.
The death of SEO, the wholesale abuse of copyright, and the further carbonization of the planet are nothing compared to the upcoming battle between flexible AI agents: OpenAI Operator, Google Gemini Astra, and (effectively) Apple Intelligence. Nerds that read this won’t like that I conflated Apple Intelligence with the two major formal agent infrastructures, but my worldview and consulting work are focused on the functional equivalencies of replacement technologies. With enough of a head start and the right prevailing winds, I’ll race my hot air balloon against your jet any day. For me, the last decade-plus was a successful portfolio bet on a “toy” ecommerce system called Shopify, which I was told repeatedly would never catch up with “real store software.”
Apple Intelligence’s “toy AI” has many of the same characteristics – OpenAI and Google need to take more margin from their third parties to perform financially. The thousand-dollar candy bar that Apple has in our pockets gives them both the opportunity and incentive to make room for more and healthier third parties. OpenAI’s and Google’s challenge is that their AI task is far more difficult. The core of their Everything Machines are much more massive – imagine spreadsheets a trillion-plus columns wide and long but related across tens of thousands of dimensions instead of the two dimensions on our screen. In terms of AI compute expense, that trillion-plus columns across are the challenge. They have to capture each structural weakness and exception in (at least) the English language, multiplied by all the other exceptions.
Contrarily, Apple already has a primitive Everything Machine in our pockets and on our wrists, which gives them a tremendous advantage here both in terms of timing and approach. Their Everything Machine is small enough to live on our phones – which we pay for upfront. Today, Siri’s AI performance clearly lags, and Apple is clearly abusing their ability to be late-to-market, but by the end of the decade, they will still turn out to be the Turtle compared to the other two Hares. Both Meta’s glasses and the Jony Ive|OpenAI device collaboration are wildcards in this calculus, but getting lost in that speculation will not help us mere mortals and our careers survive the growth of AI.
The upcoming AI agent battle is a fight to increase our productivity, first in our personal lives and then at work. Just as Amazon started with overturning book stores, the early giant AIs have quickly turned over search. That search doesn’t yet directly compete with iOS apps to stream licensed content or deliver meals and goods, but that will all start to happen before the end of 2025. Everything Machine agents will depend on their ecosystem participants to customize services and apps to make up for the limitations of their agent capabilities. It’s SEO all over again – Google’s search would suck if it were not for the millions upon millions of web publishers that work to optimize their sites for the legacy search crawlers and indexing infrastructure. Operator and Astra require similar cooperation, this time to navigate or evade commerce forms, captchas, and other human-centric features.
How the Everything Machines now deliver data – and will soon deliver dinner – to us is additionally compute-intensive and can most reasonably (queue the Angry Nerds again) be represented as three buckets. The least ambitious, yet still amazing, is Inference, in which the AI interpolates within its existing training set to surface historical statistics, facts, and similar. Next up is GenAI, during which the AI extrapolates from its training set to write our school essays or create images that no human has ever drawn. Third and last, so far at least, are the agents whose main function is to interact with third-party internet systems to execute on some productive set of actions. OpenAI Operator and Google Project Astra get all the press attention, but the Everything Machines’ web crawlers are already agents assigned tasks by the core AIs that are being operated at scale. Operator and Astra will represent radical increases in scope, but there are already impressive agent operations at work today.
There are three differentiated aspects of Apple's approach. First, Apple’s “main” model is three billion parameters, rather than OpenAI’s 1.7 trillion, and resides on the phone. They certainly have lots going on in data centers behind the scenes, but it's nothing compared to their AI competition. Apple’s Everything Machine will be able to update every night on our bedside tables while we sleep using computer chips that are a source of margin rather than capital expense. Those chips are significantly their own IP, not NVIDIA’s. Second, Apple need not crawl the internet gathering the world’s information, organizing it at incredible cost, and hoping to distill it in some way that satisfies some random AI consumer. We have all spent the last fifteen years downloading iOS apps, configuring them, and filling them with private data that Apple can uniquely keep in the individualized AI that will live in Apple Intelligence on each of those phones.
The third difference is both a Carrot and a Stick for Apple and iOS apps. The Carrot is obvious – iOS app providers share Apple’s great incentive to keep as much consumer engagement within the phone as possible. Consumers’ core smartphone usage is concentrated on so few apps per person that those iOS app providers can afford to do a lot more work for Apple than they can for OpenAI and Google. And they are doing so, or will start very soon (Message Prashant or me – our core practice is helping with this) or they will see Apple hand their app share to a more enthusiastic partner. The Stick is the result of a rare self-conscious decision by Apple to let a third party, ChatGPT, into one of its core consumer workflows. It shows just how big a deal this whole AI hegemony is. With Apple Intelligence, ChatGPT is currently the backfill for Siri. When Siri encounters questions it struggles with, it can pass it along to ChatGPT. For example, when asked for the current dew point, Siri politely suggests passing on that query to ChatGPT. Yet that information exists in Apple’s Weather app and every other weather app that may be installed on an iPhone. If Apple Intelligence, Siri, and iOS apps fail consistently, ChatGPT will be offered the consumer engagement and transactions that iOS and Apple Pay would otherwise capture. This drives Apple to put tremendous pressure on app providers to make their data available to Apple Intelligence and jointly fend off ChatGPT.
This world is consistently conquered by Faster, Better, Cheaper -- Pick any Two; Apple owns Cheaper and likely Better. They also know that they have let the wolf into the henhouse and what that means if they do not execute.