I saw a post on LinkedIn recently that claimed 99.7% of the people on Earth have never paid for AI.
The post estimated that only 24 million people were paying for a premium language model and only a few million were using more advanced tools such as agents, copilots, and automated workflows.
Those numbers are already out of date. OpenAI now says ChatGPT alone has more than 50 million subscribers. Other paid products exist too, and many people encounter AI inside software they already use without buying a separate ‘AI subscription.’
Still, the post points toward something important.
There is an enormous gap between the number of people hearing about AI and the number who have found it valuable enough to pay for directly. The technology industry talks as though AI has already reached everyone. In reality, we are still a relatively small group talking to one another about models, agents, context windows, and artificial general intelligence.
Most people are simply trying to get through their day.
Think about a leaking pipe.
Some people see the leak and immediately call a plumber. They may already have someone they trust.
Others put a bucket under it, drive to the hardware store, and try to repair it themselves. If the repair works, they are done. If it does not, they probably call the plumber.
Then there is a third group. They put down a towel or leave the bucket there. Maybe they cannot afford the repair. Maybe they do not have the time, tools, confidence, or knowledge to deal with it. Maybe the leak does not feel urgent enough yet.
The reason is less important than the behavior. Three people can encounter the same problem and respond in three completely different ways.
Technology works the same way.
A small group sees a problem and hires someone to solve it. Another group searches for a tool, learns how to use it, and tries to build a solution. A third group lives with the problem because the path from ‘this is frustrating’ to ‘technology could help me’ is not obvious enough.
Most AI products are still designed for that middle group.
A chatbot assumes the user can recognize the problem, describe it clearly, write a useful prompt, evaluate the answer, and figure out what to do next. More advanced agent tools ask that same person to connect accounts, grant permissions, define workflows, and decide how much authority the agent should have.
Those products can be incredibly powerful. I use them every day. But they still resemble a very well-stocked hardware store.
They are most useful to someone who already knows there is a leak and has decided to fix it.
That may explain why the chatbot has spread so quickly without yet becoming the final form of the product. Chat made AI approachable. It gave millions of people a simple place to begin. But making a technology accessible is not the same as finding the form that makes it indispensable.
A System That Goes With You
My first guess is that portability and continuity will matter more than another isolated chat window.
Today, our information is scattered across phones, computers, cars, televisions, smart speakers, calendars, inboxes, and dozens of applications. Each surface knows a little about us, but very few of them work together in a useful way.
I should be able to press the voice button in my car and ask, ‘Is my package arriving today?’
I should not have to remember which store shipped it, which email account received confirmation, which delivery company has it, or which application contains the tracking number. The system should understand what I mean, find the appropriate information, and give me a simple answer.
At home, I should be able to ask about my calendar or send someone a message without first remembering the exact command or navigating through a chain of menus. A conversation I begin on my phone should be able to continue my computer or television when that becomes the better place for it.
At work, I should not have to explain the same project from the beginning every time I move from email to Slack, Teams, or another application. The system should know enough about the work, within the permissions I have given it, to help me continue.
That does not require handing an autonomous agent for control of my life.
It requires continuity.
The useful product may not be an agent that does everything for us. It may begin as a system that remembers enough, across the right surfaces, to keep us from constantly starting over.
AI Needs More Than One Body
When I talk about embodiments, I do not mean every AI needs a humanoid robot.
An embodiment is simply the form through which intelligence becomes useful.
A report can be an embodiment. So can a project workspace, an email conversation, a voice inside a car, a smart speaker, a television application, or a small family robot sitting on a kitchen counter.
The right form depends on the work.
This idea is part of what I am exploring through both CoffeeBreak and Open Jibo Cloud.
Launch Check is the first focused CoffeeBreak application. It reviews an application and turns what it finds into something a builder can examine and act on. The point is not to give someone another empty chat window. The point is to create a useful result around a specific problem.
Mission Control is the next part of that direction. Instead of every useful interaction disappearing into a conversation, work can become a mission with a goal, state, history, approvals, and a visible next step.
Then there are Personas.
Persona can provide a consistent role, working style, knowledge boundary, and way of communicating. A Persona might help review work, coordinate a routine process, or take responsibility for a defined class of tasks. Over time, a mission could be handed to the right Persona based on the role needed.
That is very different from asking one giant agent to know everything, remember everything, and do everything.
People do not work that way. Businesses do not work that way. Useful systems probably should not work that way either.
Different Kinds of Memory
Continuity does not mean pouring every detail of a person’s life or a company’s work into one enormous memory.
We have been careful in CoffeeBreak to distinguish between different memory systems and different kinds of intelligence.
Some information belongs to a mission. Some belong to a workspace. Some belong to a role or Persona. Some may be useful in the moment but should never become a permanent memory at all.
Memory can also mature.
Something learned during a single task might remain temporary. If it proves useful repeatedly, it might become longer-term knowledge. Eventually, carefully selected knowledge could contribute to fine-tuning a model for a particular domain, workspace, role, or Persona.
Internally, we sometimes describe this as lasagna.
There is a foundation layer. Expert knowledge can be added above. Workspace-specific intelligence can add another layer. A Persona layer can shape not only what the system knows, but how it behaves, communicates, and approaches its work.
Different tasks may also require different levels of intelligence. A routine classification should not necessarily use the same expensive model as a difficult architecture decision. CoffeeBreak’s agency layer is being designed to route work based on what the task actually needs, while the Persona layer maintains the more specific role, memory, and behavior.
The goal is not to use the largest model for everything.
The goal is to use the appropriate intelligence, with the appropriate memory, in the appropriate place.
That is where AI starts to become a system instead of a feature.
From the Screen to the World
The longer-term part of CoffeeBreak is what we call Embodied.
The idea is that the useful parts of the system should be able to travel with you. The same underlying continuity might appear through Slack or Teams at work, email when that is the natural surface, SMS when you are away from a computer, or voice when your hands and eyes are occupied.
It could appear in your car, on your television, through a smart speaker, or inside a robot like Jibo.
The form can change without forcing you to begin again.
Jibo makes this especially interesting to me. Jibo was built as a family robot, but the original company shut down in 2018. I am now part of a group working to revive Jibo through Open Jibo Cloud.
A robot does not need to be the source of all the intelligence it uses. It can be a physical expression of a larger system. It can provide presence, movement, voice, and personality while drawing on memory and intelligence that can also be available somewhere else.
That raises some wonderfully strange questions.
Could the same AI system work through a modern phone, a Jibo robot, and my first computer, an Epson Equity I+ from 1987?
The Equity does not have the processing power to run a modern model locally. It was never designed for anything resembling today’s AI. But could it become another surface for the system? Could it send a request, receive a response, remember where I left off, and participate in the same continuous experience?
I bet it could.
That experiment would be fun, but it would also make a serious point. The future of AI may not be defined by one device, one model, one chatbot, or one all-powerful agent.
It may be a connected system of specialized intelligence, memory, tools, roles, and surfaces that follows the user without demanding constant attention.
The Product Is the Continuity
A chatbot is useful. An agent can be useful. A robot can be useful.
But none of those forms automatically becomes the product people need.
The real product may be the continuity among them.
It is the ability to ask a question in the car and see the supporting information when you reach your computer. It is the ability to assign work through Teams and review the result later through Mission Control. It is the ability for a Persona to understand its role without gaining access to every other part of your life or business.
It is intelligence appearing where it is useful, remembering what it should, forgetting what it should, and reducing the friction between a problem and the next reasonable action.
That kind of system could help all three people standing under the leaking pipe.
It could connect the first person to trusted help. It could guide the second person through the repair. It could help the third person recognize the problem, understand the options, and take a manageable next step.
That feels closer to broad utility than another blank chat box waiting for the perfect prompt.
Maybe the chatbot is not the product.
Maybe the product is a useful connection that does not disappear when you close the window.
