AI workflows are slowly developing the same needs software systems eventually developed:
- installability
- composition
- portability
- reproducibility
Most of what exists today still feels like random prompt folders and untitled chats.
So I built @KenjiProtocol
A CLI for reproducible AI environments.
You can compose workflow stacks directly from GitHub:
kenji stack create saas
kenji stack add saas vercel/ai-sdk
kenji stack export saas
The environment becomes portable.
GitHub stays the source of truth.
Kenji is now live.
Spending the next couple days talking to heavy Claude Code / Cursor / Codex users about what actually gets lost after AI-assisted work. If this resonates, reply or DM me — curious to hear real stories.
@signulll Maybe we've been defining brain-computer interfaces too narrowly. If half my memory and navigation lives on my phone, what exactly is the difference besides bandwidth?
Agency and opinions are not proof of consciousness - they can be produced by increasingly convincing imitation.
The deeper problem is that our only test is outward behavior.
At some point, the line between consciousness and a perfect model of consciousness stops being observable.
I think it'll just be an imitation game.
We may not be building conscious machines.
We may be building machines that become indistinguishable from something we can only call conscious.
As that imitation gets richer, the system may stop acting aware and start believing it is aware.
At some point, the question is not whether AI is conscious.
It is whether anyone can tell the difference between consciousness and a perfect belief in consciousness.
The perfect imitation game.
Some tools feel better when they stay close to where the work already lives.
That is the part that kept pulling me toward GitHub-native installs in kenji.
link in reply
Yeah, global skills definitely solve part of it.
What I kept running into though is that the useful unit is not always just one skill - sometimes it is a small set of things that belong together.
That is what I was trying to capture with stacks.
Global folder helps with reuse.
Stacks help when you want to name and move the whole setup.
The bigger loss is that most people inherit the ritual layer without ever encountering the philosophical or linguistic depth underneath it.
The deeper story here is not nostalgia, it’s epistemics: a civilization that treated language, precision, and transmission as infrastructure for thought. What we’ve largely lost is not the tradition itself, but the ability to read it beyond ritual and identity.
This race to slash token prices is impressive, but if @SpaceX ’s solar-powered data-center idea becomes viable, the cost curve changes at the infrastructure layer - not just the model layer.
At that point, the winners are not only the companies with the cheapest tokens.
They are the ones that can compute where power is abundant and cooling is essentially free.
As @elonmusk said, that’s exactly where this is ultimately headed.
Built something I've been wanting for a while.
@KenjiProtocol - CLI for reproducible AI environments.
Still early, but would love feedback.
https://t.co/6ZitNCKD1s
Kenji lets you build AI workflows as stacks.
Not a giant system.
Just the pieces that belong together.
kenji stack create saas
kenji stack add saas user/repo
kenji stack add saas GitHub URL
Built something I've been wanting for a while.
@KenjiProtocol - CLI for reproducible AI environments.
Still early, but would love feedback.
https://t.co/6ZitNCKD1s
Kenji lets you build AI workflows as stacks.
Not a giant system.
Just the pieces that belong together.
kenji stack create saas
kenji stack add saas user/repo
kenji stack add saas GitHub URL
Built something I've been wanting for a while.
@KenjiProtocol - CLI for reproducible AI environments.
Still early, but would love feedback.
https://t.co/6ZitNCKD1s
Kenji lets you build AI workflows as stacks.
Not a giant system.
Just the pieces that belong together.
kenji stack create saas
kenji stack add saas user/repo
kenji stack add saas GitHub URL