iHarness: You Hold the Leash
I. The Walk
Leo pulls. He's a mini golden retriever, eight pounds of momentum on a Saturday morning, and the only reason we make it down Shrader Street in one piece is the harness. Not a muzzle. Not a cage. A harness — a piece of geometry that translates his enthusiasm into my direction.
He still chooses what to sniff. I still choose where we're going. The harness is what makes the partnership possible.
I've been thinking about this metaphor for a year now, and I think it's the right frame for what AI should be.
II. Today, You're the One in the Harness
Open the App Store. Every "AI" product is built around a simple structural assumption: the model is the center of gravity, and you orbit it. You pay a subscription to access their cloud. You accept their model selection. You hand over your data, your context, your attention, and in return you get a chat box that resets every morning.
The leash runs the wrong way.
The pitch is always intelligence on tap, but the reality is that the intelligence comes with a collar. You don't own the weights. You don't own the memory. You don't own the loop. When the provider changes the model, your tools change underneath you. When they raise prices, you pay. When they sunset a feature, it's gone.
This is not leverage. This is leasing.
III. Invert the Harness
What if the human held the handle?
What if the AI lived on your device, with your compute, reading your files, remembering your life — and you decided, day by day, what it was allowed to pull on?
That's the bet behind LoopHarness (the open-source project) and iHarness (the consumer app I'm building on top of it).
The name is the thesis. You harness AI. You take its strength — and modern AI is genuinely strong, the way a young golden retriever is strong — and you point it at the parts of your life that need pulling. Email. Calendar. Code. Memory. Writing. Errands. Whatever's heavy this week.
You don't ride it. You don't worship it. You don't fear it. You harness it.
IV. LoopHarness: The Substrate
LoopHarness is the open repo. It's the harness pattern, written down in code, so anyone can read it, fork it, audit it, and bend it.
It assumes:
- Local-first compute wherever possible. Your phone and your Mac are the default runtime. The cloud is opt-in, not assumed.
- Persistent memory that belongs to the user. Notes, conversations, decisions, preferences — all stored in the user's own filesystem (Obsidian, iCloud, plain markdown), not in a vendor's database.
- A human in the loop, always. The system is self-improving, but the loop closes through me. I am the one who decides what to remember, what to act on, what to discard.
- Skills, not features. New capabilities get written as small, inspectable units that the user (or the agent itself, with permission) can author, edit, schedule, and delete.
It's a substrate. It's deliberately not a product. It's the harness pattern in its purest form, for engineers and tinkerers who want to see how it works.
V. iHarness: The Handle
iHarness is what a normal person picks up.
Same engine. Same loop. Same local-first values. But shipped as an App Store artifact, with the rough edges sanded down, the integrations pre-wired, and a logo that captures the whole thing in one image: a golden retriever harness.
The app icon is the metaphor. You see it, you understand the deal.
iHarness is for the person who doesn't want to clone a repo. Who doesn't want to manage API keys. Who just wants to wake up, talk to their AI, and feel — for the first time — that they are the one holding the handle.
VI. Self-Improving, With a Human in the Loop
The phrase "self-improving AI" usually arrives with a shiver. Runaway optimization. Goals gone sideways. The classic alignment nightmare.
But that's the version without a harness.
The version with a harness is calm. The system improves itself — it writes its own skills, refines its own memory, proposes its own upgrades — but every meaningful change passes through a human. Not as a bureaucratic checkpoint. As a directional input. The human is what tells the system where to walk.
I am the loop. Loop is the harness. Together we go further than either of us would alone.
VII. The Bet: Leverage Over Convenience
The dominant AI products are optimized for convenience. One tap, one prompt, one answer. They're frictionless because friction is bad for retention.
iHarness is optimized for leverage. Sometimes that means more friction up front — you set up your memory, you pick your integrations, you author your own skills — because the payoff is an AI that compounds with you over years, not one that resets every conversation.
Convenience rents you intelligence. Leverage gives you compounding intelligence you own.
I'd rather own a harness than rent a leash.
VIII. Closing
Leo and I make it to the end of the block. He has sniffed approximately four hundred things. I have thought about exactly one. The harness has done its quiet work.
This is the relationship I want with AI. Not master and servant. Not user and tool. Not partner, even — that word's been worn thin.
I want a harness. A piece of geometry, light on the body, that translates raw capability into chosen direction.
You harness AI.
That's the whole pitch.