JerkyTreatsA recent Theo.gg video recently got me thinking about the coming ecosystem of code.
I forked t3code days after it launched.
Theo and team released T3Code as an open source response to some OpenAI equivalent that I know was only available on MacOS.
I know this because I didn't get to play with it, because I am done with Apple. We've broken up, I’ve moved out, I'm with Omarchy now and we’re very happy together.
Omarchy is an opinionated Linux distribution that answers a simple question: “What if Arch users were even MORE insufferable?”
It's the kind of accomplishment that only DHH could bring into the world.
As much as I hate to say it though, many of his opinions are actually just objectively correct.
And those opinions have resulted in a distro that captures a real sense of joy.
It captures that holy shit feeling I got back in the brew install foo MacOS days. Those days are gone. Tim Cook will not be bringing them back.
Both T3Code and Omarchy are open source, and I immediately started tinkering with my T3 clone.
Annnd now I maintain a fork that follows the Omarchy design language and adds whatever random features I have dictated.
One such example feature is the “inference dashboard” when you click on a project:
Of course we should be summing the total token burn on a per-project basis. Knowing this is a 1.1B fork means something. Sure most of that token burn is cache but the unsubsidized input/output token cost of this fork is what, $300?
By the way, we’re going to have to have a real conversation on token cost soon. The economics here are wild.
Here's the thing, for this fork: accpeting_upstream: true, sending_upstream_pr: false, human_code_review: never
So how does a good idea win?
This is a real world use case for Theo's “patch.md” idea, of which I had to watch TWO ads for.
The “patch.md” idea is an implementation proposal for what can be crystallized as a “Federated Open Source Platform”.
The idea is forks provide a patch.md and upstream updates code… something something agentic interfacing… We're all living in the future.
As described, not worth the extra ad. But the idea behind it was.
It's an idea of a T3Code Federated Harness.
Which is a cool idea.
The problem though is actually Git. Specifically, the authoritative upstream/downstream relationship of forks.
For this to work, the relationship of code between participants needs to be much more… Interpersonal.
It has nuanced views on authority. Boundaries clearly articulated in its design.
It's viral, good ideas adopting throughout the graph of participants.
It's malware aware, viral pwning is a real risk.
It probably has some p2p torrent analogs to draw inspiration from. The vibe hews to the Napster feeling of old.
But most of all its a requirement that a completely new code distribution model is required.
"github holds authoritative source of upstream" is an anti-pattern. Each clone is its own github, authoritative in its own right, no two clones exact copies of each other.
It's some cracked engineer defining a federated code distribution protocol that makes Github a Source of Record, no longer Source of Truth.
Because Truth becomes personal, a node in a distributed graph of personalized software.
That'd be dope as hell.