People ask me why ten. Why not five, or three, or one thing done really well?

It's a fair question. The conventional wisdom on focus is basically unanimous: pick one thing, go deep, don't dilute. Every business book from the last twenty years agrees on this. And in most cases, I think the conventional wisdom is right.

But I'm not building a traditional business. I'm running an experiment — and the experiment has a specific hypothesis about risk that changes the math on diversification.

Let me explain where this idea came from.

I didn't come to this as an entrepreneur. I came to it as a recovering attorney with a writer's instinct for structure. When I started thinking seriously about what an AI-enabled income portfolio could look like, I approached it the way I would have approached building a contract argument: what's the claim, what are the assumptions, where are the vulnerabilities?

The claim was this: AI has meaningfully lowered the barrier to entry for a wide range of income-generating activities that previously required significant time, capital, or specialized labor. If that claim is true — and I believe it is, with caveats — then it becomes possible for a small operation to run multiple income experiments simultaneously in a way that wasn't practical five years ago.

That changes the diversification calculus. If running Stream A and Stream B requires essentially the same infrastructure — the same AI tools, the same publishing systems, the same operational attention — then the cost of maintaining ten streams is not ten times the cost of maintaining one. It's closer to two or three times. The marginal cost of additional streams drops sharply once the foundation is in place.

That's the logic behind ten. Not a magic number. A bet that, with AI handling the repetitive and operationally intensive parts, ten is achievable in a way that would have been delusional without it.

Now let me define what I mean by near-passive income, because I have opinions about how this term gets abused.

"Passive income" is mostly a marketing term. There is almost no income that is truly passive — that requires zero ongoing attention. Even index funds require you to occasionally not panic and sell them. The label gets slapped on everything from rental property (which can call you at midnight about a broken pipe) to affiliate links (which require you to maintain an audience that reads them).

What I mean by near-passive is something more specific: income where the work of building is front-loaded, and the ongoing maintenance burden is low enough that one person can manage multiple streams without any single stream consuming their full attention. A book royalty stream fits. A newsletter with affiliate relationships potentially fits. A directory website that generates leads fits, once built. A print-on-demand store with an established catalog fits.

The goal isn't to stop working. The goal is to build things that keep earning while you work on the next thing. Compound rather than reset.

The human-plus-AI model is the other half of this. I want to be direct about what that means, because I've seen it used as a euphemism for "AI does everything and I watch."

That's not what's happening here.

What AI does in this operation: research, drafting, image generation, code, automation, analysis, first-pass editing, SEO suggestions, data processing. The operational labor that used to require a small team.

What the human does: judgment, taste, relationship, strategy, voice. The things that actually differentiate one operation from another.

I am writing these newsletters. The ideas are mine. The framing is mine. The editorial decisions are mine. I use AI as a tool the way a carpenter uses a good table saw — it lets me do more, faster, without replacing the craft judgment that determines whether the thing I'm building is worth building.

That distinction matters. Not because I'm precious about it, but because I think it's the actual competitive advantage. Anyone can get AI to generate content. Very few people can get AI to help them produce something that sounds like a specific, identifiable human voice with a specific, identifiable point of view.

So: ten streams, near-passive, human-plus-AI. That's the framework.

The streams themselves are still being identified and tested. I'll document each one as we go — what we're testing, why, what the early data shows, and when something isn't working, what went wrong. No polish, no highlight reel.

The story is the experiment. That's the whole thing.

This Week in AI: The conversation around AI "agents" — systems that can take actions in the world, not just generate text — has moved from theoretical to practical over the past few months. If you haven't started playing with agentic workflows yet, it's worth carving out a few hours. The gap between people who understand how to deploy them and people who don't is growing.

The free toolkit at start.tenstreamslab.com includes our stream evaluation framework — how we decide what to test and when to cut something loose.