I picked up my phone last week to check the weather and got into a small existential crisis before I made it to the forecast.
Three different AI assistants offered to help me. One was built into the operating system. One was built into the browser I had open. One was built into a productivity app I had not opened in two days but which had pushed a notification asking if I needed anything. None of them were the AI assistant I had actually chosen on purpose, which was open in a different tab on a different device.
I closed the phone and went outside to look at the actual sky.
It was not really about the weather. It was about a question I had been avoiding, which is: who is the one I am actually going to use, and who decided that for me?
That is the question this whole talk is about.
The Wrong Frame
Almost every AI conversation I hear in 2026 is built on the wrong frame.
The frame says the race is about which model is best. Which model has the highest benchmark score. Which model is the smartest, the cheapest, the fastest, the most accurate. People in this conversation argue about it the way they used to argue about phones in 2010, comparing specifications and waiting for the next release.
The frame is wrong because the race is not about the model. It never was. The race is about the layer the model lives inside.
The OS Layer
The history of platforms is the history of layers. Microsoft did not win the desktop because Windows was technically better than the alternatives. It won because it became the layer every other piece of software had to run on top of. Apple did not win mobile by selling the most phones. It won by becoming the layer the apps had to ship through. Google did not win search by having a marginally better ranking algorithm in 2003. It won by becoming the layer that the next twenty years of the internet was built on top of.
In every case, the winner was the company that got to be the substrate. Not the best individual product. The thing the other products had to plug into.
The AI version of that race is happening right now. It is largely invisible to people who are not paying close attention, because the race does not look like a race. It looks like a series of feature announcements that all sound vaguely the same.
Apple Intelligence. Microsoft Copilot+. Google Gemini integration. Amazon Alexa+. They are not feature releases. They are four different bets on becoming the AI layer that everything else runs on top of.
Every one of those companies has decided, independently, that the assistant is the new shell. That the place where the user spends their attention is going to shift from the app icon to the conversation, and that whoever owns the conversation owns the platform underneath it.
The Quiet Open Standard
There is a second race happening alongside the corporate one, and it is the part most people are not watching.
In late 2024, Anthropic released a thing called the Model Context Protocol (MCP). It is, in plain language, an open standard for connecting AI assistants to tools, data, and apps. It is to AI assistants what HTTP was to the early web. A neutral protocol. Anyone can implement it. Anyone can build a server that exposes their tool to any assistant that speaks the protocol.
In the eighteen months since, hundreds of MCP servers have appeared. Slack, Gmail, Notion, Linear, calendar tools, file systems, database connectors, CRM platforms. The pace of adoption looks like the early app store moment in 2008, except the unit being shipped is not an app. It is a connection.
What that means in practice is that the AI layer might not end up owned by any single company. It might end up looking more like the open web than the closed app store, with assistants from different vendors all able to talk to the same set of tools. That would be a very different outcome than the one the big platform players are betting on, and it is the second race nobody is naming out loud.
The honest position is that we do not know which race wins. We know both are happening. We know the closed-platform bets are bigger and more visible. We know the open-protocol bet is technically further along than most observers realize. And we know the SMB and education buyers making decisions in 2026 are mostly choosing without knowing they are choosing.
Same Story Across Every Buyer
For the small and mid-sized business owners I work with, this race is not abstract.
Every vendor selection in 2026 is quietly an AI substrate decision. Choosing Microsoft 365 with Copilot vs. Google Workspace with Gemini is not a productivity decision anymore. It is a decision about which AI assistant is going to be embedded in your team's daily work for the next several years. Choosing the company device platform is the same. Even the browser choice is the same now, because the AI assistant in the browser is going to be the place where a lot of the next-generation work happens.
These choices feel reversible. They are reversible in theory, the same way switching from Microsoft Office to Google Docs in 2015 was reversible in theory. The actual cost of switching, once a team has built workflows, memory, and habits around a particular assistant, is going to be high. Higher than most decision-makers are pricing in.
I talked to an SMB owner a few weeks ago who described the experience well. She said she felt like she was "voting in an election whose ballot was hidden." She was making a vendor decision. She was not making, in her own framing, a platform decision. The market was treating those as the same thing without telling her.
The Honest Part
I am going to be careful here, because predicting platform winners is the genre of writing most likely to embarrass its author in three years.
Every previous "X is the new operating system" prediction has been wrong about half the time and right about the other half, and which half each prediction ended up in was usually determined by some combination of regulatory action, developer adoption, and timing nobody could have called in advance. There is going to be an AI OS layer. Multiple players are racing for it. We do not yet know who wins, whether the win is winner-take-all or fragmented, whether it is open or closed, or whether the answer changes again in three years.
Anyone who tells you they know which company wins is selling something.What we can say with confidence is that the question is real, the race is happening, the choices being made today are going to matter more than they look, and the people making those choices mostly do not realize the time horizon they are committing to. Those four things are enough to act on without needing to know the winner.
What I Tell People To Do
When clients ask me how to think about this, I give them three questions in order. They are not predictions. They are decision hygiene.
Which AI assistant is currently embedded in the tools my team actually opens every day, and did I choose that on purpose?
If I had to switch AI substrates eighteen months from now, what would the cost actually be in retraining, lost memory, and broken workflows? Be honest, not optimistic.
Am I picking tools that lock me deeper into a single vendor's AI stack, or tools that give me optionality through open protocols like MCP?
Those three questions will not tell you which platform wins. They will tell you whether you are an active participant in the race or a passive one, and right now most buyers are passive without realizing it.
That is the part I want educators and execs to walk away with. Not a prediction about the OS layer. A clear-eyed recognition that the choices you make about everyday tools in 2026 are platform-level choices, and they deserve more thought than the average procurement cycle is giving them.
The election is happening. The ballot is hidden. The first move is to find out which one you are voting in.
If you want help running an AI stack audit on your own organization, that is one of the things bensaibrain.com is built for. Come say hi.
Sources
Created with ❤️ by humans + AI assistance 🤖