It’s the Context Stupid!

The New York Times ran an opinion piece today about the growing backlash against artificial intelligence. I recognized immediately that the solution had been designed and articulated months ago. People are not simply angry about data centers. They are disillusioned at the power demands, job threats, education problems and the same familiar Silicon Valley arrogance, and they are asking a basic question: who is this actually for?

The industry still does not have a good answer. The public was told that AI would transform work, education, health care, business, and ordinary life. What people have seen instead is a strange mix of impressive demos, weak usefulness, high costs, social harm, and a constant demand for more trust from companies that already spent decades burning through it. The problem is that the future being offered to them still does not make enough sense in their actual lives.

Months ago, in Meaning Before Applications, I made the architectural argument underneath what is now becoming the public argument. The failure is not the AI model. The failure is the architecture underneath it. Current AI is being asked to produce coherent intelligence from fragmented data, broken context, and applications that were never designed to share meaning with each other.

The file-and-folder model was built for a world where computers could store information but could not understand it. A folder has no meaning. It is a place where a person decides to put something, usually at one moment in time, for one purpose, under one label. That made sense when the computer was only a digital cabinet. It does not make sense when Ai is being asked to reason across a person’s work, obligations, relationships, history, and decisions. The old model forces meaning to live inside the containers of files and folders. AI makes that backwards.

In an AI-native architecture, the question is no longer where the file sits. The question is what the information means, who it involves, what it connects to, what decision it supports, what commitment it creates, what source it came from, and what action it now requires. A contract, an email, a calendar event, a phone call, an invoice, and a note may all be separate files or records in the old system, but in the real world they may be one matter, one customer, one promise, or one open obligation. The better structure is not a better folder tree. It is a persistent semantic layer that organizes information around entities, relationships, time, commitments, and source. That is better because AI no longer has to reconstruct the work every time. It can operate from the meaning that already exists.

The industry keeps answering this with bigger models and more compute. That misses the point. A bigger model does not fix missing context. A larger data center does not make scattered meaning coherent. A connector does not change the fact that the work itself is still trapped inside application silos.

Most work is not a prompt. It is a living context. It has history, state, authority, and direction. Current AI does not hold that structure, so every interaction becomes another attempt to rebuild the work instead of acting from an understanding that already exists. Everyone now understands the context window in a chatbot. PSL/i turns that idea into permanent, user-owned context storage for your data, your meaning, and your life. The difference is that the context does not disappear when the chat ends. It persists, compounds, and becomes the foundation AI works from. And you own it.

The Times piece uses the plumber example because it gets AI out of the fantasy world and back into real work. A plumber does not need a chatbot that gives plumbing advice. He needs a system that already understands the shape of his day. Who he is seeing. What was promised. What has to be brought. What still needs to be billed. What call has to become the next appointment.

That is not really a chatbot problem. It is a data organization problem. More exactly, it is a meaning problem. The system has to organize information the way the work actually happens, not the way software happens to store it. That is where AI is the strongest, but only if the structure underneath gives it meaning to work from.

The same is true for a lawyer, a doctor, a contractor, a small business owner, or a family trying to manage ordinary life. Their work does not live in one app. Their meaning does not live in one database. Their obligations do not sit neatly inside a single workflow. The information is spread across systems, and every system sees only its own slice.

That is what PSL/i was designed to solve. The system does not begin with the application. It begins with meaning. It organizes around the real domains of human activity: people and relationships, communications, commitments and decisions, financial activity, work and projects, knowledge and documents, time and events. Applications become views into that foundation instead of prisons around the data.

That is the turn the AI industry still has not made. It keeps treating the model as the center of the stack, then trying to attach the model to everything else. PSL/i makes the Personal Semantic Layer the foundation, with the Navigator interpreting intent, Lenses performing domain work, Human Approval controlling execution, and the system improving as meaning compounds over time. The model is still important, but it is no longer being asked to perform magic over broken context.

This is also why trust cannot be solved with a safety memo or another public statement about responsible AI. Trust has to be built into the structure. The user has to own the context. The system has to know where information came from. It has to ask for approval when intent is unresolved. It has to preserve meaning over time instead of forcing the person to rebuild it every session.

That is the difference between assistance and experimentation. People know when technology is being done to them. They may not use architectural language, but they understand the feeling of being forced into systems that do not understand them and then being told to trust the result. That is what the public is rejecting now.

The AI companies are spending enormous sums building capacity before they have solved usefulness. They are building data centers before they have earned trust. They are building models before they have built the missing layer between intelligence and the actual life of the person using it. That order is backwards.

You cannot reliably automate work before you have properly represented work. You cannot make jobs better if the system does not understand the job. You cannot make ordinary life easier if the architecture still treats a person’s life as a pile of disconnected messages, appointments, files, transactions, and tasks. Intelligence needs something real to work with.

PSL/i begins with representation before automation. It holds meaning before it acts. That makes the system safer, cheaper, more useful, and more aligned with the person using it. The point is not to make AI sound more human. The point is to make AI work from the actual structure of a person’s life.

Economics follows the same logic. Current AI has to keep rebuilding context, which means higher compute cost and weaker output. The persistent semantic layer does that work once and holds it. It does not need to copy every piece of raw data into a new system. It needs to know what the data means, where it came from, how it connects, and what role it plays in the user’s reality.

That is why this is an infrastructure argument, not a feature argument. A better chatbot is still a chatbot. A better workflow tool is still trapped inside the workflow it was built for. A personal deterministic semantic layer changes the foundation underneath all of them. Once meaning is persistent, the system becomes modular where every domain specific Lens becomes more useful because it is drawing from the same living structure.

The public backlash is not the end of AI; it is the market telling the industry that the current answer is not good enough. The market is rejecting extraction, arrogance, waste, and systems that demand trust before they have earned it. They are rejecting a future where the benefits flow upward while the costs are pushed outward and down.

AI needs persistent deterministic context, user-owned meaning, source traceability, human approval, and domain-specific execution that broadens AI capability and allows AI to work to its potential with humans or in some cases without  That is what I architected months ago, and that is what the current moment is now making plain.

Meaning before applications was never a slogan. It was the foundation AI was missing.