Am I Talking to One AI or a Bunch of Different People?

What started as a question about too many ChatGPT conversations led me to question how we organize our relationship with AI.

Introduction

ChatGPT is organized around conversations. The interface even looks a little like text messaging. You start a conversation, talk about something, and it gets added to the list with all the other conversations.

That made sense to me at first. But as I started using ChatGPT for more things, the list of conversations grew like crazy. Projects helped organize them, but eventually I started wondering why I was organizing things this way in the first place.

I am not having conversations with a bunch of different people. I am talking to the same AI about different topics.

So why do I need to keep starting and managing separate conversations?

It seems like there should be one place where I talk to the AI. If I am working on a client, it should bring that information into the conversation. If I switch to something personal, it should switch with me. The separate conversations can still exist in the background so I can search them or go back and see what was said. I just shouldn't have to manage them in order to talk about something.

That was the question I started with. I thought I was asking about a better way to organize ChatGPT.

From My Experience

The problem became more obvious as I started using AI for more than just asking questions.

I use it for work, but “work” is not one thing. I may be working on a client issue, developing a report, researching something, writing an article or thinking through a business idea. I also use it for personal things. I might ask about tennis, something going on with one of my cars, a financial question or whatever else happens to come up.

Some of those things clearly belong to projects. Some don't. Some may start as a simple question and turn into something much larger.

That last part is important because I often don't know where a conversation belongs when I start it. I may ask a question because something caught my attention. The answer raises another question, which leads somewhere else, and eventually I realize I am working on something completely different from where I started.

That is actually how this article started.

I wasn't trying to write an article about AI. I was questioning why I had accumulated so many ChatGPT conversations and whether there was a better way to organize them. But once I kept following the question, the issue became less about organizing conversations and more about why I needed to organize them at all.

The more capable the AI becomes, the stranger that responsibility seems. If it already knows that I have clients, projects, ongoing responsibilities and other things I regularly work on, why should I have to tell it which organizational box to open before I can ask a question?

Maybe organizing the information should be part of the AI's job, not mine.

Once I started thinking about it that way, I began to picture one primary place where I interact with the AI. I might say I want to work on a particular client, then switch to something involving my business, check on a personal matter and later return to the client. I should not have to find the right conversation each time. The AI should be able to bring the right information forward as the subject changes.

That also means a conversation probably isn't the thing that should determine what gets remembered. Some things are ongoing projects. Some are clients. Some are people. Some are responsibilities that never really end. Others are temporary problems or questions that may never matter again. Trying to make all of those fit into the same container called a conversation doesn't make much sense.

At that point I started thinking about how this works with people.

If I had a Chief of Staff, I wouldn't hire a different Chief of Staff for every subject I deal with. I would have one person who understood the different parts of what I was doing. I might tell that person we need to work on a client issue, then ask about something else, then come back to the client. I would expect the Chief of Staff to keep track of the context.

That seemed much closer to what I actually wanted from AI.

The idea became even more interesting when I considered what AI can already do beyond having a conversation. I already use automated tasks that can monitor something and bring it to my attention when it matters. AI can search for information, work with email, use applications and perform increasingly specialized tasks. As those capabilities expand, there is no particular reason I should have to manage each one individually.

If I tell the AI what I am trying to accomplish, it should be able to determine what information it needs and which tools or specialized agents should do the work. I still set the objective and make the decisions that require my judgment, but I don't need to supervise every step used to get there.

That sounds less like an assistant and more like a Chief of Staff.

And if the AI can coordinate specialized agents, monitors and tools on my behalf, then those begin to look a lot like its staff.

That was about as far as I could take the idea just by thinking through how I was actually using AI. The obvious next question was whether anyone else was thinking about it the same way.

What I Found

Once I had worked through the idea for myself, I went looking to see if other people were thinking about AI this way.

They are.

I found everything from research papers to experimental projects and new products using terms like AI Chief of Staff, personal operating system and agent orchestration. The terminology varies, and the implementations are different, but a common idea kept showing up: instead of treating AI as a chatbot that waits for the next question, people are starting to build systems that maintain context, coordinate work and use other agents or tools on the user's behalf.

One of the more interesting examples came from the Brookings Institution. A 2026 paper on what the authors call “context-maxxing” describes advanced users adopting a Chief of Staff, or what the authors call a Chief Context Officer, that maintains broader context while delegating specialized work to other agents and bringing the results back to the human. That is remarkably close to the structure I had been imagining: the person deals primarily with one AI while that AI figures out what needs to happen underneath.

Then I found projects attacking the conversation problem almost directly. Zarviz describes its approach as “one conversation, forever.” Instead of making the user manually manage context, its system identifies the intent and project and assembles the relevant context for the model. Its developers make an important distinction: they are not simply trying to build a chatbot with a bigger memory. They are separating the stored information about the user from the particular slice of context the AI needs right now.

That distinction between memory and context matters. The AI does not need everything I have ever discussed loaded into every interaction. It needs to know what information exists and retrieve the part that matters for what I am doing now.

Another project called Fénix takes a similar approach. It maintains one continuous conversation while keeping separate context for different projects behind the scenes. It can also send work to background subagents. The developer describes the problem in almost exactly the terms that started me down this road: moving between projects without having to switch chats or lose track of work already underway.

I also found open-source projects explicitly describing themselves as personal operating systems. One AI Chief of Staff project uses persistent files and memory so the AI can begin a session already knowing what the person is working on and what was previously decided. Another, Personal OS, goes further and puts a Chief of Staff agent above specialized agents for planning, research, execution, review and other jobs. The Chief of Staff decides when those specialists are needed.

The more I looked, the more it became clear that I had not stumbled onto a unique idea. In fact, pieces of it are being developed in several different ways.

But that made the original question more interesting, not less.

These systems are approaching the problem from different directions, but many seem to be moving away from the same thing: the idea that our relationship with AI should be organized around a collection of isolated conversations.

Maybe the important shift isn't from a chatbot with poor memory to a chatbot with better memory. Maybe it is from a chatbot that participates in conversations to an AI that maintains an understanding of what I am doing.

If that is where this is going, then the conversation is still useful. It just may not be the thing everything else should be organized around.

Sidetracks

One of the things I kept coming back to was how people actually handle context.

Think about walking down the hall between meetings. You leave a meeting about one subject and run into someone who asks you about something completely different. While you are talking, you remember something else you need to tell them. Someone else walks up and asks a question about a third subject. Then you continue down the hall and walk into your next meeting.

If you ever watched The West Wing, you saw an exaggerated version of this constantly. People walked through the White House discussing one issue, changed subjects when someone joined them, handed off a task, made a decision and kept walking. Nobody stopped and announced that they were closing one conversation and opening another.

That is much closer to how my mind works than a list of separate chats.

It also made me think about meetings themselves. Companies have meetings all day long, but they don't normally organize the company around the meetings. The meeting is an activity. What comes out of it may belong to a customer, a project, an employee, a decision or an ongoing responsibility. We may keep minutes or a transcript so we know what happened, but the transcript isn't the organization.

Why should a conversation with AI be any different?

That question becomes even harder when I don't know where a conversation is going. I might ask whether it is going to rain before a tennis match. That sounds like a simple weather question. But maybe the answer makes me wonder how quickly a tennis court dries. That leads to how the court is constructed, which leads to drainage, which makes me wonder whether there is a better way to build one. Now my weather question has become an engineering discussion and possibly a business idea.

Where was I supposed to file that conversation before I started it?

The obvious answer is that I shouldn't have to.

That led me to another distinction I hadn't thought much about before: memory and context are not the same thing. A good Chief of Staff may know an enormous amount about the organization and the person they work for. That doesn't mean all of that information belongs in every discussion. The skill is knowing what matters now.

AI may need to work the same way. It can remember that a person, client, project or decision exists without dragging everything it knows about them into every conversation. When the subject changes, the relevant context changes with it.

And not everything deserves to be remembered forever. Some things are ongoing responsibilities. Some are projects with a beginning and an end. Some are temporary problems. Some are ideas worth coming back to. And some questions are just questions.

The more I thought about that, the less sense it made to look for one perfect container to replace the conversation. Maybe there isn't one.

The real problem may be expecting the person using the AI to manage the containers in the first place.

Soft Ending

I started with a pretty simple complaint. I had too many ChatGPT conversations and thought there ought to be a better way to organize them.

I don't think that is the problem anymore.

The strange part may be that I am being asked to organize my relationship with AI around conversations in the first place.

Conversations still matter. I want to be able to go back and see what was discussed, where an idea came from or why a decision was made. But that makes the conversation a record of what happened, not necessarily the place where everything related to that subject has to live.

The same may eventually be true of projects, agents and tools. I shouldn't need to know which agent to call, which application to open or which project contains the information before I can say what I am trying to accomplish. Those are implementation details.

If AI continues moving in the direction it appears to be moving, “assistant” may eventually be the wrong word for the role it plays. An assistant helps me do things. A Chief of Staff understands what I am trying to accomplish, keeps track of what is going on and coordinates the people and resources needed to get it done.

I still set the objectives. I still make the decisions. I still exercise the judgment.

But I shouldn't have to manage the staff meeting.

I should be able to say what I am trying to accomplish and let the AI figure out which staff meeting to call.