I was going to write a proper long-form but fuck it, let's just do a thread

TENEX is a tool I've been using and building for the past 6 weeks (started the last day of nostr:nprofile1qythwumn8ghj7ct5d3shxtnwdaehgu3wd3skuep0qyt8wumn8ghj7etyv4hzumn0wd68ytnvv9hxgtcqyzpanxdpfp39c0fthqv67vrycrm2ztta4z8k3vkxjgsl8f6xzuw3jddlvze )

After hearing what sounded like a bunch of cope from a ton of devs ("oh, it's so good that I'm a dev because I can understand how to build with it, so my skillset is tooootally relevant!") I wanted to explore the opposite.

Like a keynesian, what if it's not only not relevant, but you gotta unlearn in order to go with the vibe?

Here's what it looks like.

The interface literally doesn't even show you the code.

All vibes.

Reply to this note

Please Login to reply.

Discussion

So, what does it do?

Every project there is a nostr event (new kind)

And every task there is another nostr event (also a new kind)

Each project gets its own nsec when you create the event.

Agents building on the project publish nostr events constantly as they work (via an mcp tool I built that I will talk about later)

Events are just kind 1 with a confidence score and a git commit hash.

Why is this cool?

A) ... if you have to ask..."you wouldn't get it"

B)

because that means that absolutely all context is kept on nostr, is addressable, it timestamped and thought processes are preserved AND addressable.

So I can fucking reply to any part of the thought process and correct its work.

Here comes (one of) the cool parts:

because the thought process is addressable I can reply to the specific point where the thinking, the assumption, the approach, was incorrect.

I can correct it.

But the correct, and the incorrect thought-process remains preserved and addressable.

This is not the only way, I will talk about some of my agentic setup later, but this is part of the magic that makes the agents make less mistakes over time.

Literally they have a lessons file that is being updated with what the human taught them.

Talk is cheap, show me the code

Very cool. Are these events meant to be public?

nostr:npub10umqehmuluwrejgwdrpd8xq6dgcnc7jp2fw4wy48td8a7arl50xstzvxxy

Yes! You sure can.

I hadn’t seen this note before 😅

Which Nostr client fits for this task?

nostr:npub1l2vyh47mk2p0qlsku7hg0vn29faehy9hy34ygaclpn66ukqp3afqutajft Which Nostr client will fit here?

Sounds cool.

As someone who is not a professional developer, this could not be less appealing.

I can't get an AI to write simple English text in a way that feels coherent, not robotic and not generic.

If that's what LLMs have to offer for text, I can't imagine that the quality of code (which requires understanding problems at a deep level) would look like.

you are the only sane person in this thread

I must admit that with each new model, my daily workflow looks more and more like it.

nostr:nevent1qqstycwc8qu8telckv9mw4fs7wc6d54v9gnpnqrx9hpllqz994mnz4ckzvrxq