Start Here: From Personal Computers to Personal AI
I remember when computers lived in rooms.
You went to them. You waited your turn. They belonged to institutions, not people.
Then the personal computer arrived. And the computer came home.
I've spent 25 years in technology. Networks, IoT, enterprise sales. I watched the internet, mobile, and cloud each change how we work and live.
Lately, I keep coming back to one thought:
AI is at its "personal computer" moment.
How I got here
Today, most of us use the same few AI models. Shared. Central. The same for everyone.
It feels a lot like the mainframe era. Just with a friendlier face.
What comes next, I believe, is a personal model. One that knows your history, your people, your priorities. And on top of it, agents that act on your behalf. Booking, planning, researching, reminding.
So I started sketching what that might look like.
My first idea was simple. One agent for each role in my life. A husband agent. A father agent. A golf agent. A work agent.
Then I noticed a problem.
They would all want Saturday morning.
Something above them would have to decide. And that something is me.
An unexpected teacher
That question, who is the one deciding?, led me back to Ramana Maharshi.
Ramana taught that the "I-thought" is the root of all other thoughts. And that it is not the Self. The Self is the witness. The mind is a bundle of thoughts.
I wasn't looking for an architecture in a spiritual teaching. But there it was.
- The Self is the human. The final say stays with us.
- The orchestrator is like the I-thought. Central, useful, but in service.
- Agents are like thoughts. Many, useful, temporary. None of them is "you."
- Roles are lenses, not separate selves.
To be clear, I'm not saying AI has a Self. It doesn't. That seat belongs to us.
I'm using the teaching as a lens. And keeping only what turns into something I can build and test.
I wrote up the framework here: One Self, Many Agents.
What I'm attempting
Most technology is designed to take our attention.
I want to see if personal AI can do the opposite.
Can it give attention back?
So I'm going to build one. My own personal AI, on my own machine, one layer at a time:
- A model that runs privately, on my Mac
- A memory that belongs to me
- A short set of values it must follow
- One simple agent that does one useful thing
- And an honest look at what should never be handed off
I don't know exactly where this ends. That's part of the point.
How I'll share it
Two places, two purposes.
- Here, on the blog: the why. What I'm thinking, what surprised me, what I'm learning.
- On GitHub: the how. The setup, the steps, the code. For anyone who wants to build along.
Each post here will link to its build notes.
Why share it at all?
Because I learn best by doing. And I understand something better when I try to explain it.
And because many people, smart people, are watching AI from the outside. Unsure where to begin.
If that's you, follow along. Try a step. Tell me what you find.
Let's stay curious.
Next: Setting up the workbench.