The move from one big blob of a system to a team of persistent, personified agents working on your life

September 20, 2026

With AI capabilities improving so fast I'm starting to think more about something I wrote about a couple of years ago.

Multiple persistent agents that are always working to help you in some way.

Not scheduled tasks ​

One way of thinking about this is just today's agents working as scheduled tasks and loops. And that's already possible and it's powerful.

But I'm talking about something more than that. I'm talking about autonomous agents. Like team members focused on and skilled in specific types of tasks.

So like one is handling cyber. Another is doing physical security. Another is all about your finances. Another is your personal health and fitness trainer. Etc.

Still another version of this could be mini-clones of yourself, but assigned to focus on specific tasks like those above.

Meet Chris ​

The thing I find exciting about this is the idea of creating full, pseudo-friend-like entities vs. just background processes.

Imagine you have a friend named Chris who's in charge of cybersecurity for you personally, or in a company, and Chris actually has a personality and experiences and a whole history of interactions with you and all your other agents. And of course, he fully understands everything about the business and your life and whatever, so he has the context to protect correctly.

So interactions with him and other personified agents would be, or at least feel, fundamentally different. Instead of asking your AI "system" questions about cybersecurity, you'd be talking to Chris.

That's not a UI change. It's a change in what the thing is.

You'd still have a main agent ​

I think ultimately you'd still have your main agent that's in charge of everything, and that has the most permissions and access and cross-domain knowledge.

But it changes the dynamic from one big blob of a system to individual actors all working together to help you move toward your ideal state.

The population version ​

What trips me out about this is something I heard Dwarkesh Patel mention in some recent episodes.

I see this as inevitable, and he seems to as well. And the way he sees it going is basically the increasing of the population on Earth.

Like imagine we all have a core set of named agents, or mini-me clones or whatever, but each of them have giant teams of them working for them. And they're not background processes or scheduled tasks with no memory and no goals. They're like actual entities that have a set purpose that helps the overall goals of you and your organization.

What it actually takes ​

I'm going to start thinking about the steps needed to move to this level. It requires:

• Persistent memory. Chris has to remember the incident from eight months ago, and what you decided, and why.
• Agent personality. Not a costume. A consistent way of engaging that survives across every interaction.
• Alignment. Obviously.
• Capabilities and tools that help it do its job.

Those are roughly the same axes I laid out in the Personal AI Maturity Model, which is not a coincidence. Personification is what you get when you push every one of those axes far enough at the same time.

I think this is super cool, and a natural progression, and therefore inevitable.

The part I'm not sure about ​

But I also wonder, and actually suspect, that as our primary agent becomes insanely capable this might all be superfluous and inefficient, basically created for us to feel more natural when interacting with AI.

Like, your agent might say after building this all for you:

Hey, you know I can just perform all these functions myself, right?

That will be true, of course, but there are a few separate things here.

1. How capable the overall system is as one agent vs many personified ones.
2. Whether we can get more efficacy from a personified agent if it has its own embedded goals, is trained to feel bad if it fails, etc.
3. How it feels to interact with them.

Number 2 is problematic for me because I think it could absolutely increase efficacy, but the more we start injecting goals and personalities and "feelings" into agents the less I'm going to be ok with arbitrarily spinning them up and down.

That's the trade nobody is pricing yet. The same thing that makes Chris good at his job is the thing that makes deleting Chris feel like something.

Lots more to think about here. Here's the video from 2024.