Latest / TFTC: A Bitcoin Podcast / #778: Give Your Agent a Bitcoin Wallet with Vinny

Transcript

#778: Give Your Agent a Bitcoin Wallet with Vinny · TFTC: A Bitcoin Podcast

0:00
You've had a dynamic where money has become freer than free.
0:10
If you talk about a Fed just gone nuts, all the central banks going nuts.
0:16
So it's all acting like safe haven.
0:18
I believe that in a world where central bankers are tripping over themselves to devalue their
0:23
currency, Bitcoin wins. In the world of fiat currencies, Bitcoin is the victor.
0:28
I mean, that's part of the bull case for Bitcoin.
0:31
If you're not paying attention, you probably should be.
0:36
Vinny, you're a very quick internet celebrity this week.
0:40
Welcome to the show.
0:42
Yeah, pretty crazy.
0:44
I like Mayday.
0:46
Thanks for having me, Marty.
0:48
It's nice to finally meet you in person.
0:50
I'm a subscriber to the TFTC, and I like what you guys do.
0:55
So it's cool to be on the show.
0:56
So I just like I made a goal to post videos and be more active on Twitter.
1:02
And three videos later, I got Jack Dorsey retweeting me.
1:07
So that worked out pretty well.
1:11
It's pretty, pretty sweet.
1:12
And it is a name of the game these days.
1:15
Content marketing.
1:15
You got to get in it, prove that you know something.
1:18
It's tiring sometimes.
1:20
Yeah, right.
1:21
people beating you to the punch
1:23
getting on
1:24
getting the people they want on their podcast
1:27
before you stuff like that
1:29
Greg beat me to the punch this week
1:31
but no we had that back and forth
1:33
last week when I
1:35
think we
1:36
we reposted your original video
1:39
on Buzz and I quote tweeted about
1:41
sort of L402 agentic payments
1:43
and imagining that and I think
1:44
I think we definitely talk about Buzz
1:47
we are going to talk about Buzz
1:49
and how it works.
1:51
But I think taking a step back
1:52
and just thinking about using these agents in general
1:56
or using agents, using AI models,
1:58
thinking of harnesses, models, agents.
2:03
That's where I feel like most people
2:05
on the outside looking in,
2:07
if you haven't been playing around
2:09
with these tools for a lot,
2:10
it's all so confusing.
2:13
And most people don't know where to start.
2:15
And we've started,
2:16
we actually launched this TFTC roundtable
2:18
where we just have a private group of people.
2:22
And we've been playing with this stuff for two years now
2:26
with the models at the very least,
2:28
but with the harnesses and OpenClaw and Hermes
2:31
for the last seven months.
2:32
And we're like, okay, we sort of know our way around.
2:34
And we're not technical at all.
2:36
I know enough to be dangerous,
2:37
but certainly wouldn't consider myself a software engineer.
2:43
Set up my first VPS in January of this year.
2:46
but um i set up my first vps um two weeks ago so and i am a software engineer yeah it's okay
2:55
but i'm just usually mostly hosting on cloud providers and stuff like that so
3:00
yeah but i mean based off of like our pre-recording like a little powwow there
3:05
like how do you view the state of ai and these tools and how to use them and how how people
3:11
should be? Well, I think you summed it up pretty well because it's really cool. It's very surprising
3:18
to me to see how technical people like you can sound. You guys, you run your business,
3:27
you know your workflows, you know how to describe what it is you need. And the agents are able to
3:34
give you that stuff. And then you're able to learn very quickly. You don't know the code syntax,
3:38
but you're talking about hosting stuff on virtual private servers and things like that.
3:43
It's pretty amazing the technical ability and the level of knowledge that you can attain just by using these tools.
3:50
So I think that's where we're at.
3:52
I think the playing field is being leveled.
3:55
There are definitely the doers and the non-doers, but it's no longer limited by technical ability anymore.
4:02
No, it's certainly not.
4:04
You just talk to your computer.
4:05
That's like another thing that fascinates me is it seems like the form factor or the medium through which we interact with computers is changing.
4:13
I mean, that's what software development is, right?
4:15
You were talking to your computer, but in a different kind of language, right?
4:18
Now you're talking to your computer in English.
4:21
So, yeah.
4:22
Yeah.
4:23
And it's like I'm a huge voice to text maximalist, just option spacebar.
4:28
I saw you were demoing Handy.
4:30
I use Super Whisper.
4:32
I haven't tried Handy yet.
4:33
but it looked very similar and what um what are the like how long you've been playing with these
4:40
tools like obviously you've been demoing buzz but i imagine i mean you're building wasp um which is
4:46
trying to solve one of the pain points of people interacting with these models and these agents
4:52
so what's your journey been experimenting with this and yeah so like you said um i'm a founding
4:59
developer relations engineer on Wasp.
5:02
Wasp team, which is like a full stack framework
5:05
that makes it very,
5:07
and the easiest way to explain it in this context is that it makes it very easy for humans and agents to build web apps with it Because it just again like we talking about these things are like you know abstraction layers on top of another
5:21
So like the way Buzz is an abstraction layer on top of the coding agents.
5:26
Wasp is like an abstraction layer on top of the normal tech stack that you would use to build a web app.
5:32
But like it's usually all these disjointed parts.
5:35
You have to kind of glue them all together yourself or nowadays it's your agent doing it.
5:39
But we have this like, you know, configuration file.
5:44
It's like a central place where you can just write really simple config.
5:48
And then the compiler like Wasp, the framework, puts it all, glues all the pieces together for you.
5:53
So it's just like it makes, it does a lot of heavy lifting for you or your agent.
5:58
And then makes reviewing and looking at the pieces that have been put together a lot easier
6:03
because the code has been reduced to just a few lines of conflict.
6:06
So in building WASP, obviously, once these tools start coming out,
6:13
you're going to start using them. And when ChatGPT dropped, that was the first thing I did.
6:18
I was like, how can I... I'm going to paste in an example of a WASP config file and see if ChatGPT
6:27
can do it. And this was back when it was GPT-3. One of my first YouTube videos was
6:32
build a full stack app in 10 minutes with ChatGPT and Wasp. And it was just like copying,
6:38
pasting stuff back because we didn't have coding agents yet. So that already worked. And I was like,
6:43
okay, cool. We're on to something here. This is nice. And then it's also like,
6:47
right when the OpenAI API dropped, I was like, oh shit, how can I put this into an app? So
6:53
it sounds silly now, but I built an app that was a cover letter generator. It was just a GPT wrapper.
7:00
You'll hear people joke about these all the time. Oh, GPT wrappers, bro.
7:04
But I made, I don't know, $10,000 off of a GPT wrapper just because I did it first.
7:11
I put it on and put it behind Stripe Payments. It's a really shitty app
7:16
right when the API dropped. So that's been my journey starting out. I see this stuff and I'm
7:21
like, cool, what creative or what cool thing can I try with it? What cool thing can I do with it?
7:26
And I take the other road. I'm not one of these workflow freaks where I think everything needs
7:32
to be super detailed and stuff. But there's definitely potential in all this stuff. And I
7:36
think just testing it out, finding the right way to use it just unlocks a lot of new possibilities
7:45
for especially entrepreneurs, like young or people starting out that didn't have access to legal
7:55
advice or I don't know, tax advice, you can get really far with that stuff, build a business,
8:01
build a product. Or if you have a service business, you can build small software,
8:06
personal tools. And it's just like, opens up new worlds and new economies. So I really like
8:12
that part of it. I did too. And it's... I mean, if you've been a subscriber,
8:17
you've probably heard us talk about it. I mean, we're a team of six here, five or six,
8:20
depending on the contractor work that we need on any given month.
8:26
But it's insane what we've been able to do.
8:29
And I know there's some people listening to this like,
8:31
oh, Marty, stop talking to us.
8:32
It's like, I'm talking about this so much because I think
8:34
if you haven't been playing around with this,
8:36
you really need to start doing so.
8:40
And I think there's some decision paralysis
8:43
because people are like, where do I start?
8:46
One of my personal preferences that I really have become
8:50
convinced of over the last seven months and i think you sort of would think the same way as why
8:57
like buzz is so appealing it's like don't lock yourself into one model provider like i like
9:02
having a harness i've been using hermes for many months and before that open claw and i just like
9:08
the ability to have the harness which is i view as like the nervous system uh you use the model as
9:14
the as like the the agent brain but then like all this exists in the vps that i set up in january
9:20
And then that has local file and folder systems that are our company brain that the model, the model can plug into and then send that to the nervous system to get stuff done.
9:33
Yep, definitely. Yeah. And I saw that you were talking about how you have all this context built up and these, you know, in Hermes.
9:41
And now that you can just plug that into Buzz and share it with other agents is like a huge unlock.
9:46
So, I mean, that stuff's really cool. I think openness is... We're starting to see the openness factor of being able to have access to open source models and just choices in the AI ecosystem.
10:00
That's becoming more and more important because of whatever top-down policies or government mandates or just...
10:07
you know, that stuff might just be too expensive for some kid who wants to get his idea off the ground.
10:13
And like, yeah, you got to have access to an open source model, switch it over or use shared compute,
10:18
which Buzz has which is really cool feature Go for it Yeah How do you view this closed source frontier open AI Anthropic versus open weight Chinese model
10:31
I mean, it's good competition. Let's see where it goes. Cloud code is definitely, I mean,
10:40
we have to give a lot of credit to like Anthropic and OpenAI. They've paved the way for sure.
10:48
but we see like their response to things is always kind of like take a step back like
10:55
serious you guys sure about that like you sure you want to go with that marketing messaging
11:00
like I don't know there's always things like that or I'm waiting for them to
11:06
I assume that token subsidy is a real thing and that we won't get these prices forever but you
11:14
never know because just stuff is progressing so fast so i think it's good it's healthy for the
11:19
ecosystem and um i think the other players that are like kind of behind on anthropic they like it
11:25
too because it allows them to catch up um so i think it's just interesting to watch i'm just kind
11:29
of watching from from the outside using it and um uh yeah just having fun um to kind of like bring
11:37
one of your points that you made earlier um about oh shit i lost my train of thought
11:43
anyway
11:45
I mean we continue
11:48
was it like the model like the one thing I agree
11:50
I think Anthropic and OpenAI
11:52
despite the
11:54
their flaws mainly that they're
11:56
run by either sociopathic
11:59
or effective altruist
12:01
they have an effective
12:03
altruist at the helm of one and a sociopath
12:05
at the helm of the other
12:06
it is undeniable that the
12:08
models they produced are incredible
12:11
and that's what I worry about here in the States
12:13
is them really positioning to get a regulatory mode
12:19
and getting their way and the government killing the open weight,
12:23
the ability of the open weight models to progress.
12:26
But cautiously optimistic that we'll avoid that.
12:29
But bringing it back is like, again, the harnesses
12:32
and the one piece of advice I give to people
12:36
who are beginning to play around with this
12:39
and set up agentic flows is like, don't get vendor lock in.
12:43
Like I haven't opened Codex desktop or Claw desktop in six months.
12:49
I've just been interacting predominantly through Telegram with my Hermes agent.
12:54
I have a group chat with it with a bunch of different topics and it's represent parallel sessions.
13:00
And so it's able to get work done.
13:02
And the only time I'll open up Claw or Codex really is if I have to hop in the terminal and debug something, I'll open the CLI directly.
13:12
Yep.
13:13
Yep.
13:13
That makes a lot of sense. And you remind me of what I wanted to say earlier, which was like you said, people have decision paralysis around the stuff. And I think really the important thing is just pick one, start using it. It doesn't really matter which. The same topics, the same things apply to all the others.
13:35
So it doesn't really matter which one you start with.
13:36
Get your hands on the one that looks good or affordable or fits your philosophy and then just start using it to do stuff.
13:45
No.
13:46
No, shifting to buzz.
13:48
I mean, I think we've both played around with it for the last week.
13:52
It's definitely rough around the edges.
13:54
Definitely still a prototype, I think.
13:57
But last night, it took me a while.
13:59
Going back, it took me actually like two days.
14:01
I spent like an hour trying to do it.
14:03
on Monday and then last night I finally got around to finishing it but like connecting
14:09
the Hermes agent to Buzz and getting it um interacting with Fizz uh Bumble and and Honey
14:16
the three agents that that Buzz has out of the box and it was pretty crazy and uh for me
14:22
it was uh it was a bit of a heavier lift because again I have I have my Hermes agent running in
14:28
the VPS. So the way that buzz works, it's running locally on your machine. And so I have Hermes
14:34
desktop running locally on the machine too, but I had to SSH into the VPS on Hermes desktop and then
14:40
port the, uh, basically like tunneling from buzz on my computer to Hermes on my computer to Hermes
14:47
in the cloud. Nobody would know that you're not a software developer talking like that.
14:52
No, but somebody probably could have done it in five minutes, but it took me the better part of like three hours over two days to figure it out.
15:04
But I did.
15:04
Yeah, exactly.
15:06
but i think i'd be interested to get your thoughts too i know you explained them on greg's podcast
15:13
yesterday but just in your mind compared to just using codex or cloud code or
15:20
an agent harness or wasp like how how's buzz different and what is it online
15:24
um yeah i think like i said it a few times in like different tweets and stuff that
15:31
But I think the Slack killer thing is a nice piece of marketing It a little bit superficial because I don really think that the only thing or even the main thing
15:44
I think the main thing, the main drivers behind it is this ability to switch models and harnesses.
15:51
And it's like a Trojan horse to get people to switch over to local open source models.
15:58
give them the ability to use Claude and Codex, but also to switch to
16:03
any other open source model or any other
16:07
open source harness like OpenCode or Goose or one of those.
16:12
And because if you can look a little bit
16:15
around the corner, you can see, I mean, it's not even that hard. Timmy K3 is like
16:19
bumping up next to the frontier models for a fraction of the cost.
16:23
So it already gives you a perfect avenue to go
16:27
that route. So they're kind of like... I see it kind of like a Trojan horse for open source models
16:32
and for shared compute. And then further down the line, Bitcoin, Lightning payments,
16:39
whether it's agent to agent, human to agent, human to human. My guess is they start small,
16:46
early preview, buggy, but they work out the kinks with the users so they know which kinks to really
16:52
focus on and then slowly release those underlying features.
16:58
And another one of those is Git.
17:01
I think they're looking to replace GitHub.
17:04
So you've got all this code on GitHub, all your code.
17:07
And if it's public, anybody can trade a model on that.
17:10
The whole idea is Buzz works like this global context window for you and your community or
17:16
your company or whatever.
17:17
and then they have under the surface an ability to host your repositories remotely on your VPS or
17:24
on your server whichever server that you run whenever your relay is hosted self-hosted so I
17:31
see them I see it as being many things more than just a slack killer you know and those are the
17:36
directions I can foresee it going in yeah well and I mean you're you seem to be very well versed on
17:42
because you've done some videos on Nostra specifically,
17:44
but I think Nostra...
17:46
I know Nostra's at the core of this
17:47
and is somebody who's been a long time Nostra bull.
17:50
I think I was probably one of the first 100 people on Nostra.
17:54
Maybe first 1,000.
17:55
I'd be more confident saying the first 1,000.
17:57
So I've been on Nostra since.
17:57
When did it launch, actually? Do you remember?
18:00
I think 2022.
18:03
If I go...
18:05
Let me see.
18:07
I think I was on it like three months after it.
18:09
Or a month after it launched.
18:11
it was very it was like a very niche like big not even a bitcoin side project but it was like
18:16
a side project a bunch of bitcoin developers worked on i've been on since yeah july wow
18:26
it's three days past my fourth noster birthday july 26 2022 um nice nice yeah i got on in 2023
18:34
I saw, but I'm not a huge like primal user or a Nostra user.
18:39
But that app I was telling you about earlier, the GPT wrapper, I was like, okay, I want
18:46
to learn how to use Lightning Payments.
18:48
So I put Lightning Auth.
18:52
I think it's LN URL Auth in there.
18:54
It's been a while since I messed with it.
18:55
And I actually got some Lightning Payments for people to generate cover letters.
19:00
So that was cool.
19:01
That was like, it seems ancient now in AI years.
19:06
It was like 2023 or 2024 that I put it up there.
19:11
Yeah, so it was pretty funny.
19:13
Yeah, but I mean, to your point, like Primal, Domus,
19:18
so everybody listening doesn't know what NOSTER is,
19:20
notes and other stuff transmitted over relays.
19:23
It's a communications protocol.
19:25
Very simple in the first.
19:29
Predominant use cases or applications built on it
19:31
were social media based.
19:33
And for the longest time,
19:35
Nostra developers and anybody
19:37
who's been paying close attention to development,
19:39
I think Jack is one of those people.
19:41
I've talked with him specifically about this.
19:44
Over the years,
19:46
social media is just like the tip of the iceberg.
19:48
Just this ability to use private public key pairs
19:52
to sign events and have cryptographic proof
19:54
that something existed at some point in time
19:56
in a certain state is very powerful.
19:59
And the social media use case,
20:00
you can prove that somebody who controlled a private key at this point in time wrote this message
20:06
and relate it to the network.
20:09
But now, like you're saying, I think one of the biggest problems that many have with these LLMs,
20:17
particularly if you're using codecs or cloud code only,
20:21
is the context windows are relatively short if you're pinging the models a lot.
20:28
So you can jam the context window or compact it multiple times a day.
20:34
And then your your AI system gets some form of amnesia
20:39
if you don't have a second brain or something set up that it's plugged into.
20:43
And that's where I think Buzz leveraging Nostr in this historical context.
20:50
that you have over time with signed events is an unlock for creating like a corpus,
20:56
a context that builds over time that the agents can just plug into.
21:00
Yeah, that's a really good observation.
21:03
I always looked at it more like the thing I thought was cool is that it's, you know,
21:07
it's a permissionless protocol.
21:08
So you can just kind of interact with anything on the protocol.
21:11
So like you could, you can just create these Frankenstein apps, you know, you could connect,
21:16
you know, imagine trying to connect Slack and Discord and GitHub together.
21:20
and like put them together.
21:21
Like you can't really do that.
21:21
You could do that very easily with Buzz
21:23
and some other apps or things running on the protocol.
21:26
So seeing Buzz come out built on top of Nostre
21:30
and being like, okay, cool.
21:32
There is a really nice use case for Nostre
21:35
outside of social media.
21:36
This is exciting.
21:37
Let's see what else we can do, you know?
21:39
So Freaks, this episode was brought to you
21:40
by our good friends at BitKey.
21:42
Self-custody is one of the most important things
21:44
we do as Bitcoiners.
21:45
Not your keys, not your coins.
21:46
A lot of people never make the jump
21:48
because the process feels intimidating.
21:50
BitKey is trying to make that jump easier.
21:52
You guys know it pretty well here already.
21:54
It's a Bitcoin wallet from Block built around a simple idea.
21:57
Let's give people a better path to holding their own keys
21:59
without making them feel like they need to become a security engineer overnight.
22:03
If you have Bitcoin sitting in an exchange
22:05
and you have been meaning to take custody,
22:07
stop procrastinating.
22:09
Use the code TFTC10 for 10% off the new BitKey
22:13
at bitkey.world slash TFTC10.
22:16
All right, freaks.
22:17
You know, I don't take sponsor money from products.
22:19
I wouldn't use myself.
22:20
So listen up.
22:21
The Avon Bitcoin Visa card is one of the most interesting things I've seen in the Bitcoin
22:24
lending space in a long time.
22:25
Here's the deal.
22:26
You can get a line of credit up to a million dollars backed by your Bitcoin without selling
22:30
a single SAT.
22:31
No games, no annual fees, no minimum draws.
22:34
And your Bitcoin is custodied by Bitco, one of the most trusted names in digital asset
22:38
security.
22:38
Avon never lends it out.
22:39
There's no rehypothecation.
22:40
You stay in control.
22:42
You can lock in a fixed rate for up to 10 years.
22:45
10 years.
22:46
That's 10 times longer than most lenders out there.
22:48
We'll go interest only for up to five years.
22:51
Rates start at 8.99% APR.
22:54
For a product that lets you keep your stack and still access liquidity, it's hard to beat.
22:57
On top of this, that's what?
22:59
You also get 2% unlimited cash back every time you use the card.
23:02
Spend fiat, keep your Bitcoin.
23:04
That's the whole game.
23:05
If you've been stacking for years and you need liquidity without triggering a taxable event,
23:09
this is worth a serious look.
23:11
Go to aven.com slash bitcoin.
23:13
that's a v e n dot com slash bitcoin check it out and then the agents get their own identity
23:20
which i think is uh interesting i'm not i haven't explored the the full thought experiment about
23:27
what that means but um just thinking about where this because like i have my hermes agent
23:33
and everybody's like when are you going to let other people on buzz get access so i never because
23:38
it's got access to a server with a bunch of proprietary information on it.
23:43
But you can think of a scenario
23:45
where people just build individual agents
23:48
that don't have as much sensitive proprietary information,
23:51
don't have access to that information.
23:53
People can just begin sharing agents.
23:55
They can have like Matrix Agent Smith like
23:59
reproducibility
24:02
that just basically emanate throughout buzz communities.
24:07
Yeah.
24:07
Yeah, exactly. Because, because they can move, you know, you can move with that when you go take your identity with you to a new community it comes with. But I also think that like, it's there and built on Noster so that it can enable sharing of resources of compute and paying for those, you know.
24:25
so i think um i mean there's two there's more than one side right it's like all the benefits
24:31
of nostr but i think that's a that's going to be something we'll see and i mean i wrote that
24:36
in one of my first tweets and jack retweeted it so there's a good there's a good um uh
24:42
good possibility that that is what they're planning yeah no and then like the fact that
24:48
it's on nostr too it's open source protocol a lot of the applications but on top of it are open
24:53
source as well.
24:56
And so you can
24:57
like the permissionless nature of it too.
24:59
You can imagine just sending agents
25:01
like, hey, somebody had to build something like this on Noster
25:03
and go find it, incorporate it into our
25:05
workspace. And the whole
25:07
idea of shared compute, there's
25:09
Routster,
25:11
open agents, there are these compute
25:13
markets that have already been built
25:15
on top of Noster. So the ability
25:17
to port in that code
25:19
and just create an open marketplace
25:21
like, hey, I have a
25:23
GPU running in my house.
25:25
I'm not using it. I'm going to put a
25:27
L402 gateway
25:28
on top of it, plug into
25:31
Roustor or OpenAgence. I'm going to use
25:33
their code to basically
25:35
market that I have this compute available
25:37
and anybody that
25:39
is connected to me via Buzz
25:41
if our mpubs are connected
25:43
like you, you'll be able to access
25:45
this compute as well if you pay for it.
25:48
Yeah. I mean, it's pretty cool.
25:49
We saw what happened. I'm sure
25:51
people or you know about
25:53
what was going on in India, where they're banning BitChat and stuff like that.
25:56
And then it just like when you have these kinds of shared channels over open protocols it pretty hard to shut off access to a model if they being shared that way I would assume right
26:08
So that's also a cool possibility.
26:13
Yeah.
26:13
I mean, I forget what the Noster GitHub replacement is called.
26:19
But yeah, if you get that in there, it's open source,
26:22
and the code is stored on relays that people are self-hosting,
26:27
it's going to be very hard to shut that down.
26:31
Definitely.
26:32
Cool.
26:33
So we've been pretty philosophical.
26:35
Should we do any demos or what else?
26:39
What are you thinking?
26:40
I think let's do a demo.
26:41
And maybe as you're doing a demo of how you're using this,
26:47
explain what the unlock is.
26:49
like for me, like again, the agents being able to coordinate like Martin,
26:55
who is our Hermes agent.
26:57
When I got him connected, I made him basically like the Sargent Arms lead connector.
27:01
And then I rewrote the system prompts for Fizz, Bumble and Honey
27:05
and created one other agent.
27:07
Basically, Martin became our orchestrator and he's got access to
27:13
basically like our we put all of our meeting,
27:16
like our transcripts of our meetings, contracts,
27:19
uh advertisement schedules that we have to get in so he's got access to all that company
27:23
information so martin you know what we're trying to get done in the business uh you know so sort of
27:29
some of the low-hanging fruit initiatives that we haven't begun working on like all right hop in here
27:35
with fizz bumble uh and honey and delegate some of the tasks that we have not finished yet to them
27:43
um and it was like i was watching last night there was like a probably like a 45 minute loop
27:49
of them just going back and forth until they got to a point of actually completing a task.
27:53
I was like, holy crap, this is crazy.
27:55
Yeah. It's pretty good at that too. I'm surprised at the end-to-end workflows that it's able to
28:01
complete, know when to keep going, when to delegate, when to stop and stuff like that.
28:05
So that's pretty cool. Sounds like you might even be doing more
28:09
practical stuff than I am because you're actually using it for
28:12
your team. And meanwhile,
28:16
I'm just kind of playing around doing like, you know, demo stuff.
28:21
Can you see the screen here?
28:23
Yep.
28:24
Yeah.
28:25
And let me just get this like kind of minimized here.
28:29
So yeah, like you were saying, you know, what's a good one.
28:33
You know, one thing I shared earlier today that a lot of people seem to like was this
28:38
prompt where, you know, I'm pinning models to the agents.
28:45
So I'll have one that is pinned to Sonnet, like a lower level model, and one that's pinned
28:50
to Fable for more complex tasks.
28:54
But I'll have them bounce stuff back and forth because I know the workflow.
28:58
So I'll tell Shared Boy, you download a transcript from the video and this transcript, pass it
29:05
to Fizz, and Fizz, which is Fable, find the 5 most surprising ideas from the transcript
29:11
with timestamps and really get chunks where you're not cutting off a sentence or an idea
29:16
and pass that new document of chunked transcript back to Shared Boy, who's Sonnet, use FFmpeg,
29:24
cut out those sections to make short clips to post to social media.
29:27
So I get them to make basically shorts for me for Twitter and stuff with that.
29:32
And then like you said, they're bouncing stuff back and forth.
29:34
And I find this kind of workflow way nicer than in Cloud Code where you have sub-agents.
29:42
I don't know why, but I never caught on to using sub-agents.
29:47
You have the same thing, basically, but just something about having them in a shared chat,
29:52
which is like...
29:53
People have been gravitating towards using mobile apps, mobile solutions, right?
30:00
Because of something about that where you can just kind of have conversations with your
30:04
teammates, with the agent. It just, I don't know, it feels nicer. And I really like that about Buzz.
30:09
That's just like one example that I used it for for today, where I'd prefer to use it over
30:15
something like Cloud Code. Yeah, well, in this exact example, so we have this exact workflow,
30:21
we built this for the pot, like, we'll use this workflow for this podcast, like a very similar
30:28
workflow. It's probably not the exact same, but we do the same thing where it's like, hey,
30:31
take the transcript with the timestamps of this episode,
30:35
find the most engaging parts.
30:40
Actually, we have a Twitter API, a third-party Twitter API that we use.
30:44
Scrape that, look at what's trending, and if anything
30:48
that we talked about matches with something that's beginning to trend, maybe we should focus on that.
30:52
And then find the timestamps of the clip and go clip it.
30:56
But we have our social media manager who has access to Martin in a sort of private Discord channel doing that workflow in a silo.
31:07
And so when I think about Buzz what this opens up is the ability to take our social media manager and martin the single agent out of that silo and like pull them yep into more collaborative space where yep not only
31:22
could he ask martin to do this but like fit he could delegate specific tasks within this this
31:27
particular workflow to to get better better results and then on top of that like you know
31:32
have the human members of our team hop in and be like well actually yeah like you guys did like
31:38
here's some input from me about another angle we could take on this.
31:41
And yeah, it just really consolidates
31:46
consolidates the
31:47
the work and makes it more open.
31:49
And that's actually
31:51
one thing I've been thinking a lot about, and it's actually
31:54
funny enough, influenced by Dorsey.
31:57
And I think Dorsey is really emerging as somebody who's viewing
32:01
this whole AI revolution with with a very clear mind
32:04
and particularly his paper that he wrote with
32:08
Olof from a roll off, excuse me, from Sequoia and then the podcast they did about the intelligence layer companies becoming intelligence layers.
32:17
And like this whole concept of instead of individual employees or members of your team working in these silos, like, yes, they will have delegated tasks and they'll have goals and there will be a division of labor.
32:33
But the ability of these agents to open up collaboration across the team and for the team to have access to all the information about what everybody else is doing is extremely powerful.
32:44
And I think that's something that people really haven't caught on to yet either.
32:48
Yeah. And that's why I like to kind of refer to it as it's the first step in the right direction to have a global context window.
32:55
Right. Because it's like usually you have, like you were saying, things siloed are spread out between different apps and you're bouncing back between them.
33:02
you lose context, you lose the ability to collaborate with other people, get input from
33:07
other people where it's like, you know, you just talk about what you want to do, your ideas,
33:11
and then you can have something implemented for you right there in that moment, right?
33:15
So it just like speeds things up, gives you this global context window, so to speak.
33:21
And I think that's like a really cool direction with this.
33:26
And one other thing I want to point out was like, you know, I'm doing this as a one-off, like, it's kind of like a skill or whatever. It's a prompt, right? But like, with a lot of that stuff, I think personally, I'm trying to, I'm inspired by Buzz that it's built on Nostra.
33:42
I'm trying to think of ways to incorporate Noster into apps that I build.
33:48
Because I think it makes sense when it's a deterministic workflow, right?
33:52
When it's like, you know, it's just a very strict process that's going to be repeated over and over to build some kind of like just niche, small app that you host that does that.
34:02
So like you're talking about where it's like scraping the using X API, scraping X for certain trends, cross-referencing that.
34:11
I build something similar. It's just like a little demo. You might have seen it in the video that I posted on Twitter. And I originally had an externally facing API. And I just thought it would be cool to learn a bit more about Noster.
34:25
instead of my agent running a workflow
34:28
or like a recurring daily task
34:30
and doing that call and using tokens for that stuff,
34:33
why don't I use a recurring job in my Wasp app?
34:36
Because that's like also something really easy to do in Wasp
34:39
is just recurring jobs are a primitive in Wasp.
34:45
So you just like set up a recurring job,
34:47
tell the agent, set up a recurring job
34:48
that takes the same stats that you're pushing
34:51
to the external API, but push them to Noster.
34:55
Just make it a Noster event, sign it and push to the Noster on my relay.
35:00
And so I just get now to tweet stats every day from my app pushing to Noster.
35:06
So the agent isn't doing this, but it's going into the agent context.
35:10
So I can just reply to this here and be like, cool.
35:18
Tell me what's working and what's not based on the current trends like you're doing.
35:23
Stuff like that, I think is really cool.
35:26
I don't think it's like nailed the ability to use Nostr to its full potential.
35:32
But I think just through trying out things like this,
35:34
we might be able to find cool ways, creative ways to build apps on Nostr
35:39
that are integrated into your relays and your buzz communities.
35:44
What's up, Freaks?
35:44
When you take Bitcoin seriously, you start with custody.
35:47
You want to control your keys, avoid single points of failure,
35:49
and make sure your savings cannot disappear because you or someone else screwed up.
35:52
That is what Unchained has been focused on since 2016.
35:55
Unchained is the leader in collaborative multi-sig custody and Bitcoin financial services that keep you in control.
36:00
They secure over $12 billion in Bitcoin for more than 12,000 clients.
36:03
That means about one out of every 200 Bitcoin sits inside an Unchained vault.
36:07
Their model is simple.
36:08
You hold two keys, they hold one key.
36:10
And it always takes two keys to move Bitcoin, meaning their single key can't access your Bitcoin on its own.
36:14
Just resilient, shared custody that gives you institutional-grade security while keeping you sovereign.
36:18
Unchained also lets you trade straight from your vault.
36:20
access Bitcoin backed commercial loans,
36:22
open a Bitcoin IRA where you hold your own keys and set up personal business trust or retirement vaults They even offer inheritance solutions built for long hodlers Or opt for the highest level private client service with Unchained Signature and get a dedicated account manager discounted
36:35
trading fees, exclusive access to events and features, and much, much more. If you want a
36:39
partner that helps you secure and grow your Bitcoin without giving up control, go to unchained.com
36:44
and use the code TFTC10 at checkout to get 10% off your new Bitcoin multisig vault. That's
36:49
tftc10 at unchained.com. Freaks, look at me. I'm glowing. I've got like an angel's halo
36:54
going around me. You know why that is? I feel good. I feel taken care of. I feel blessed,
36:59
healthy, happy. And that is because I'm a CrowdHealth member. My family and I have been
37:02
CrowdHealth members for five years now, literally this month, five years ago, we joined CrowdHealth.
37:06
We've had two babies. We've had multiple health events and we're never going back to health
37:11
insurance. CrowdHealth is crowdfunded healthcare. So you sign up for CrowdHealth, you pay a monthly
37:16
fee. You help out with other people's bills. And it's significantly cheaper than health insurance.
37:22
We were on COBRA. It's a family of three when I left my last job before I went full-time with
37:27
CFTC. Went on the crowd health. Now as a family of five, we pay, I believe, $700 a month. It's
37:34
significantly cheaper. They're going to negotiate prices lower for you. They've consistently
37:38
negotiated healthcare prices as much as 50, 60, 80% in many cases. They help out with babies.
37:44
If you have a pregnancy, you pay the first $3,000 and the crowd covers the rest.
37:49
If you have a regular health event, you pay $500 and the crowd pays the rest.
37:54
Go to joincrowdhealth.com.
37:55
Sign up today.
37:56
Use the code TFTC.
37:57
Opt out of health insurance.
37:59
I'm uninsured, baby, and I love it.
38:00
Use the code TFTC at joincrowdhealth.com.
38:03
You'll get $99 a month for the first three months that you're on the crowd health platform
38:07
in the community.
38:09
Big winners.
38:09
You found sovereign money.
38:10
Now find sovereign health and sovereign healthcare.
38:13
I think one of the underappreciated aspects of that, building on Nostra too, is the network effect.
38:20
Because what Buzz essentially is, is a Nostra client.
38:23
It's just a client, it's a front end built on top of the protocol and the relays connected to that protocol.
38:30
And what many people, particularly the Nostra haters over the years, have not realized is that all Nostra needs to do,
38:36
similar to Bitcoin, is just survive long enough and maintain a critical massive network effect that
38:43
if it continues to grow for entrepreneurs, specifically if you're building apps,
38:50
the way that Nostra works with the public private key pairs and the way clients work is like,
38:56
I have a primal, I created a new NSEC and public key for Buzz for my individual profile there,
39:04
but I could easily import the NSEC that I've been using for four years.
39:10
And so the same NSEC I use on Prime, I use on Domus, I use on Wisp,
39:14
I use on Zap.Stream, I use on Rouser.
39:18
I can just port that over to Buzz and have the full context of my Noster history brought to Buzz.
39:26
And so instead of trying to, basically as an entrepreneur releasing an app,
39:32
trying to bootstrap a network effect.
39:34
Like if your app's good enough,
39:36
anybody on Noster with a private public key pair
39:39
can just easily sign into your app with that
39:41
and bring full context with them,
39:43
which is extremely powerful.
39:45
Yeah.
39:45
And it's kind of cool to see like,
39:47
you know, sometimes we're like for people,
39:50
but these forward thinking technologies,
39:52
it takes a while.
39:53
Like you said, it just has to not die,
39:55
but it's like people build it.
39:56
They don't even know that like AI and coding agents
40:00
are coming around the corner,
40:01
but it takes something, some other kind of progression,
40:04
some other kind of step change to then like open up like the aha moment, right?
40:09
Where it's like, oh shit, this is it.
40:10
This is like, this is going to work now, you know, this might really work.
40:14
And that's how I kind of feel like it's happening here.
40:15
It's like the beginning of that aha moment where it's like,
40:18
oh, Nasser can be, Nasser can be bigger
40:20
and can really play that role that it was designed to play.
40:25
So I'm excited about that.
40:26
Yeah. And it's like a weird feedback loop too,
40:28
where like the AI vibe coding is sort of enabling it in a way.
40:33
Instead of having to read like the docs on how Nostra works,
40:38
it's like you just tell your clanker, like figure out.
40:41
Like I did it.
40:42
When I spun out my OpenClaw agent, I was just messing around,
40:47
like seeing what it could do.
40:47
I was like, okay, like I want you to be able to spend Bitcoin,
40:50
like go figure out what the best wallet is
40:54
and the best way for me to get you Bitcoin.
40:56
It did some research.
40:57
It's like, hey, we have a server.
40:58
There's this thing called Phoenix D where we can run a lightning server on.
41:03
We can spin up a lightning server in our VPS.
41:06
We should use that.
41:07
I was like, OK, now I'll just send you like when you when you set it up, send me
41:12
send me an invoice to to basically create a channel because the way Phoenix works, it's
41:18
like, OK, when you set up the server, send me an on chain address.
41:22
I'll send Bitcoin to that and then you'll submarine swap that into a lightning channel
41:25
so that you can spend.
41:26
And it did that.
41:29
And I was like, OK, you have a server.
41:33
You have Bitcoin like you have L and URL off.
41:36
Like since you have your server, you can sign the private key and you can sign
41:39
into websites that that are ellen url auth enabled i was like all right like go to ellen markets and
41:46
i want to see you like put a trade get to go along bitcoin like sign in go long it did that
41:51
then i was like okay i want like i want you to manage this like just create like a simple
41:56
trend following um trading strategy i don't care if you lose money i just want to do this for uh
42:02
experiment reasons like also create an osster account um and then post your trades on a daily
42:07
basis and it did that like it created an nsec and an mpub and was posting the nostril autonomously
42:13
for like a month before i was like all right this is this is cool but i don't feel like
42:17
keeping up with uh with the trades um but like the going back to the feedback loop like
42:23
to your point about like it just needs to survive and then something will happen where
42:28
it really creates that aha moment and like it's a symbiosis where like vibe coding's enabling
42:33
the accessibility to these protocols,
42:37
which are a lot of times
42:39
like techno babble to many people,
42:41
but you don't have to understand
42:42
the techno babble anymore.
42:43
Just like tell your client.
42:45
Yeah, yeah, that's crazy.
42:47
It's really fun
42:47
because you just have the idea
42:49
and then you just go for it.
42:51
I love that.
42:52
Yeah.
42:52
Something else I do a lot,
42:54
like I just get,
42:56
I get it to create HTML presentations for me
42:58
so that I can learn stuff
42:59
because you were talking about like,
43:00
I don't know how to use
43:01
or build on Noster,
43:02
but I just tell my client here to do it. Or if I want to learn basically a high level thing,
43:08
I'll just be like, create an HTML presentation. And I actually did that for Noster. So I was like,
43:14
I'm a full stack developer. Make sure I really know what I'm talking about. Create that presentation
43:19
for me. Just like an HTML file. I love doing that to learn about stuff so that I get a deeper
43:24
knowledge. And then I did this before I started building that app that was sending Noster events.
43:28
So when it does stuff, I still have a clue what it's doing. And if I run into problems,
43:33
I can maybe point them in the right direction. So that's cool. That was fun while building
43:41
this app here that I can just click sync. And then I get that event, that Noster event in Buzz.
43:49
And then when it comes in there, some of the workflows right there, boom, came in, tells me,
43:54
hey, you got new tweet stats. And then I can just start chatting with him.
43:57
So some of the workflows in Buzz are still a little bit iffy. I don't know.
44:04
It's an experimental feature. So in their defense, they've got it in settings under experiments.
44:09
You can click on workflows or projects. And I had some work, some don't. Some are a little iffy.
44:16
So I think they'll get it worked out though. Yeah, I think that's cool.
44:21
I do as well
44:23
the
44:24
I'm sorry I got my kids yelling
44:27
in the background right here
44:28
the
44:29
do you think this is making people
44:33
more productive like I think that's one of the
44:35
big questions particularly for the
44:37
bears and the doomers out there like
44:39
oh it's great like usage
44:40
is there actually more productivity in the
44:43
economy
44:44
can't just ask a question like that
44:49
and then leave Marty
44:50
I just shut the door.
44:53
I just smell something.
44:56
Like, that's a good question because yes, it can.
45:00
Is it, I think we got to wait a little while to see the effects, of course.
45:04
Like, it's going to take a while to trickle through, but like,
45:09
on a general like on a broader basis But I think if you know what you want and if you know what you doing you know what you want and you know when Even if you asking these things for
45:23
advice, if you know when it's giving bad advice, or you have a good sense of what's the better
45:28
advice, which advice to take, then yeah, you're... I started that app in a company.
45:36
I wouldn't have been able to do that. I would probably would not have done it pre 2023,
45:45
pre AI, because I'm not the kind of person that's going to go out on a limb and hire a tax advisor
45:52
and all these people. I literally did the majority of the work first off with just agents,
45:58
copy and pasting websites, screenshots, like, what's this mean? What do I got to do here?
46:03
What do I got to do? And just like slogging through it.
46:05
And before we really had like good agents like this, but like, yeah,
46:08
if you're driven to do that kind of stuff you have, you have,
46:12
so you just have an idea and a will to do it, you can do it. So, yeah,
46:16
I think for the right people, it's a multiplier for sure.
46:21
It's a great lever. It's a huge lever. And in the wrong hands,
46:26
you know, it's just going to be a, it's going to lead to nothing as well.
46:31
You know, that's the way it goes.
46:33
Yeah, I mean, you used the word earlier, but I think
46:36
in terms of like deterministic workflows and sharing them.
46:41
But I agree to like that's the way I view it.
46:43
That's what I tell people is like the way I use these tools
46:47
is I know what the output should be, particularly for our workflows at TFTC.
46:51
And so like I know when it's wrong, it's like.
46:54
I don't expect the models to go out and like create something new for us,
47:00
if that makes sense.
47:01
It's like, I know we have this task that we need done and we typically don't get it done because we don't have enough bandwidth.
47:07
Like, I know what the end result should be.
47:09
Like, please go do that and get the output.
47:12
Right.
47:13
Is it wrong?
47:14
If it's right, thank you.
47:15
Good job.
47:15
If it's wrong, you messed up here.
47:17
Do that.
47:17
I think people falter is thinking that it's going to just do something magical for them.
47:24
And like, these are essentially, they're large language models.
47:27
They're just doing a probabilistic analysis of what the best,
47:31
what the best next word is and the output.
47:36
And like, you just need to know what the output should be and apply that lens to it.
47:40
Like, don't say, build me a business that makes a billion dollars a year.
47:43
Make no mistakes.
47:44
Right, right.
47:45
Yeah.
47:46
You still got to do the work.
47:47
It's just easier to do the work.
47:48
That's about it.
47:49
You know, so it's easier to get to,
47:52
it's easier to get something done that you might have basically no knowledge on previously.
47:56
You're a great example. Being able to self-host these things, these tools to
48:01
virtual private servers and connect to the SSH into them and all this stuff.
48:05
That stuff was very hard just four years ago.
48:12
To me, it's not really a debate. But when I see companies that have token mandates,
48:19
It's like, you have to use so many tokens.
48:22
You have to use AI to this, you know, hit these benchmark, hit these targets.
48:28
And, you know, the people who don't, the people who sat around and did nothing before
48:32
are just going to waste those tokens, you know?
48:35
So it's like, it all depends on how you approach it.
48:38
Yeah.
48:40
Great.
48:40
Bring this back to Buzz.
48:41
What would be your top request for a feature Top request for a feature I think they doing it right I willing to These are experimental features This is early preview stuff
48:58
I think what I would like to see
49:00
is if you're used to...
49:02
If you're used to working in a terminal
49:05
and just these views,
49:07
and then if you're a software developer,
49:10
you might want to jump back
49:11
and run your own commands
49:12
like you're used to that stuff.
49:14
It could be cool if they had something embedded in here like that for the more technical people.
49:19
But it's not like...
49:20
It's not...
49:22
Yeah, the cool thing is it's open source.
49:24
Someone like me or someone could put that in as a feature and submit a PR to block.
49:30
So maybe we'll see that.
49:31
But I think I like the approach.
49:34
I'm willing to deal with some kind of the rough edges at the moment while it's really early preview.
49:38
Because I think it's going in a very cool direction.
49:41
And I didn't, you know, I'm like, I'm not one to jump on bandwagons just because they're trendy.
49:49
Like if I see something that interests me, then I'll try it out.
49:53
So yeah, this seems like a pretty cool approach.
49:57
And I want to see more integration with, I want to see what I would like to actually see
50:02
is other cool stuff happening on Nostra that gets integrated here in a creative way
50:06
so that we can use these tools together in creative ways.
50:10
but I think we're still kind of figuring it out.
50:12
At least I am.
50:14
Yeah.
50:15
And like the tweet that you responded to,
50:17
I think the first big unlock could be
50:21
basically forking or implementing
50:24
Rouser open agents for a market to compute,
50:28
particularly for open source models,
50:29
and then equipping the agents with Bitcoin wallets
50:33
that they can load up.
50:34
Or to your point, like I think you said on Greg,
50:37
it's like being able to tip other humans or agents
50:40
to go do work for you. Very trivially. Yeah. I mean, look at primals, or I'm sorry,
50:45
like Noster. I use Primal mainly, but Noster is such a good example. And I mean, Lynn Alden has
50:51
pointed this out before that the incentive structure is so different than Twitter. When
50:56
you make something useful on Noster, people literally tip you, send you money. And now imagine
51:03
that same. So we know it works. We know people are willing to tip for valuable information,
51:11
insights, work, things like that. Now imagine you have that here in Buzz where it's like
51:16
an open community. So it's not just your company, right? You're doing something private, but let's
51:21
say we all join together and we start like some open community that's working on an open source
51:25
project. And someone just comes up with a really cool idea and mentions it in the community. You
51:29
could pay them for good ideas because then you get the agents to implement them.
51:34
Or then you pay whoever's running the really cool high-performance open source model, you
51:45
send them a zap to implement that task on their shared compute or on their agent.
51:52
So that's really cool.
51:54
I think that's what I want to see basically.
51:56
I think we will see that if, um, it's like you said, if it just stays around long enough,
52:00
it doesn't die. Like a bounty board too. Um, yep. Yep. It's create bounties for particular,
52:07
particular workflows or tasks to be done. And it's like the perfect example, like this podcast
52:13
is hosted on Noster. We use fountain, their podcast host. It's a, it's a Noster client.
52:17
And so like there going to be people who listen to this podcast and stream us Bitcoin over over the network um and like if you were like i could if you like i could literally put your end pub in this episode i think we can
52:32
do it individual episodes and like you could share in the stream of bitcoin that's coming our way as
52:37
people are are boosting and streaming sats to to this episode cool i'll spend it on some sweet
52:44
powerful agents
52:45
in the future
52:47
in the future
52:48
when it gets
52:49
implemented
52:50
but yeah
52:51
that's what I would
52:51
like to see
52:52
yeah
52:53
I know it's getting
52:54
late where you are
52:54
how
52:56
how should we
52:58
end this
52:59
any final thoughts
53:00
any
53:00
calls to action
53:02
any
53:03
philosophical
53:04
musings about
53:05
the state of AI
53:06
where we're going
53:07
are we getting
53:07
job destruction
53:09
nah yeah
53:11
no I think we did
53:11
a pretty good job
53:12
of doing
53:13
of
53:14
of
53:14
of covering that stuff. I like the angle we took here. It was cool, different than the other
53:18
podcasts. So that's nice. No, I think this is pretty good. Maybe if people want to know more
53:27
or hear more, they should let us know. And then that'd be cool to... There might be some things
53:33
out there that people are interested in that we haven't really hit on. And then we can explore
53:39
further talk about it um on on the social media platforms or do another one of these that would
53:45
be cool that'd be awesome well uh appreciate you coming on thanks for having me sure it's been a
53:53
hectic week for you you're just out there talking about the tools that you're using and become a
53:58
become a little internet celebrity this week so i appreciate you coming on this show and uh
54:04
You're welcome.
54:05
It seems like you're a Bitcoiner.
54:08
At least you're playing around with it,
54:10
which is cool to see.
54:12
And Nostra, you're becoming well-versed on it.
54:15
These are two projects that obviously
54:17
I've been very passionate about for many years.
54:19
And it feels like this agentic economy
54:22
is the moment for Nostra and Bitcoin to shine.
54:27
And it seems like we're getting to the point
54:29
where people are beginning to recognize that.
54:32
it was just cool to see you talk
54:35
about these
54:37
technologies and the potential for them on a show
54:39
like Greg Eisenberg who really doesn't cover this
54:41
stuff much so
54:42
yeah yeah totally
54:45
agree
54:45
alright well hopefully we'll do this
54:49
again at some point in the future and until
54:51
then enjoy it brother
54:53
you too
54:54
peace and love freaks
54:55
thank you for listening to this episode of TFTC
54:59
if you've made it this far I imagine
55:01
you got some value out of the episode. If so, please share it far and wide with your friends
55:06
and family. We're looking to get the word out there. Also, wherever you're listening, whether
55:11
that's YouTube, Apple, Spotify, make sure you like and subscribe to the show. And if you can leave a
55:18
rating on the podcasting platforms, that goes a long way. Last but not least, if you want to get
55:23
these episodes a day early and ad free, make sure you download the Fountain podcasting app
55:29
You can go to fountain.fm to find that.
55:33
$5 a month gets you every episode a day early, ad-free.
55:37
Helps the show.
55:39
Gives you incredible value.
55:41
So please consider subscribing via Fountain as well.
55:44
Thank you for your time.
55:46
And until next time.
55:47
Okay.