Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
The End of Human-First Crypto - AI Agents, Cross-Chain Execution, and What Comes Next with @lifiprotocol
Recorded: March 17, 2026 Duration: 0:56:34Space Recording
Full Transcription
so Thank you. Thank you. all right glad both of you didn't fall asleep during that intro.
Very lo-fi chilled music.
Gents, how are we doing?
So we've got the LeFi team and CoinFund to discuss
are humans actually going to be involved
in the next wave of crypto products?
Yeah, I've been looking forward to this one.
It's something that we've been discussing a lot internally.
We've got a lot of research and a lot of reports coming out on it.
So this was like perfect timing.
But yeah, first, how are we doing?
Doing great.
Doing well.
Thanks for having me.
Good stuff.
Good stuff.
So slightly inflammatory title from me.
I'm leaning heavy into this.
So the theme here, and I feel like a lot of the teams that we're speaking to,
and you guys have just collectively pushed towards this,
and it's really, really good to have this conversation now.
So I think we push towards the next 100 million users
in the industry and try to just incentivize them
with quite crude incentives and just throwing tokens at them
and come use our product.
And the retention factor was completely terrible.
But now I feel like over the last maybe even six to eight
to ten weeks it feels like that everyone i'm speaking to now is building with maybe the next
hundred million users like all that demographic to tailor towards is agents an agent first and i
know you guys have done a ton of work some like a brilliant like hackathon that you've just kind
of done now there's some really really great results but is that too hyperbolic am i like
barking up the wrong tree there is do you feel like teams are building towards that and how are
you guys thinking about that and we'll pass that to ruggal first yeah i mean if you really think
about it just makes perfect sense that this is how blockchain is going to evolve right blockchain
when i look at it it's a set of open apis that are kind of annoying and clunky for humans to use,
but perfectly set up for agents to use. And I think this agent unlock, you know, just over the
past couple months, we've seen with, you know, OpenClaw just completely turning it over about
what it means to actually be an agent and how like this completely autonomous interacting agent that
can kind of like look up whatever it needs to look up,
interact with whatever APIs it can find and just do whatever it needs to do,
just makes perfect sense to meld with this on-chain vision that we have made.
You know, I think when we first got into crypto, I know at least I did, you know,
I was kind of driven by these values around transparency, openness, accessibility,
and all these things that didn't
really make sense back then but now when you like combine that with ai you're just like you're like
holy shit this like totally makes sense all of a sudden so i'm excited to see it and uh it really
seems like the vision that we've all been building together is kind of coming together around this
and you know you're being hyperbolic but it's it's hyperbolic in the best possible way, in my opinion.
So I share a different kind of optimism.
Yeah, amazing.
Jake, so obviously from your,
you've got a very unique perspective on this
and we'll have like a bird's eye view on it all.
So how are like existing portfolio companies
thinking about this?
Are they taking a similar sort of approach
to what we're seeing here with the LeFi team
and like really trying to implement an agent first type mindset?
And like what's also coming across the desk?
Is there a lot of recurring themes here with like this recurring theme of agent first in mind or like what are you seeing?
I'm seeing, thanks for the question.
I'm seeing, you know, a lot of impact across the board, like just very fundamentally, even before we get to agents.
I feel like a lot of the teams in our portfolio have adopted AI internally. That goes for our
organization as well. We're having our AI launch tomorrow. And it's not just in development either,
it's just across all business functions.
So it feels like everybody in the portfolio is getting to know the technology and the agents very personally
because it's something that they're starting to use every day.
On the agentic side, I would say it's been really interesting to watch companies kind of adapt to agentic users.
I mean, there's a whole sort of vertical coming to fruition now,
which I call agentic services,
which are basically APIs and tools for bots.
And when you go onto these websites,
they kind of have this dropdown and it says,
okay, like if you're a human and you want to install this,
you go this way, but if you're an agent or a bot right like here's how you do it and so you you definitely see um kind of the world
adapt adapting to how agents will sort of use it and try to become more agent friendly or agent
native if you will and and the last thing i'll say is um i've seen products that open, like Agenda Services products that open up data for agents.
And I say like the underlying models that are driving agents today are already very good.
That's not actually the bottleneck to having a really effective agent.
The bottleneck is now prompting them correctly,
maybe cost, and most importantly, data.
And so I've seen companies kind of open up
some of these data silos for agents.
And when you give them the data,
they become just incredibly effective.
So I don't think we've even seen anything yet
as to what agents can do.
Yeah, I love that. And selfishlyly I'm going to ask you both separately like how is how is each individual kind of company
that you guys are at now like how is that workflow change even like anecdotally across colleagues I
seen like a ramp ramp interview this week where they were basically saying they have a leaderboard
internally so they can see like token usage across Claude different llms like i know that's like a little
bit extreme but trying to kind of like bring the ones who aren't adopting the technology up to speed
and like why the what what are the blockers like this is how you could use this show me what your
day looks like i feel like there's like a ton of dead space and a repetitive boring work that can
be automated so like anecdotally before we get into the rest of the conversation like what's that looking like internally at both uh both companies
yeah sure go for it okay sir sure so it's funny that you bring up the leaderboard thing because
we literally just did that ourselves internally uh i pulled metrics from both claude and cursor
internally and posted them and everyone actually
loved it everyone's kind of getting a little competitive like oh man next week I get I got
you you know so it's it's cool to see and I think one thing you see is you really see
kind of this divide in people there's like people who are really eager to early adopt this stuff and
just dive headfirst into it and they're like creating agents and they're automating all these workflows and they're kind of like sharing these skills and
tooling with each other. And that's really what we're trying to do at LeFi internally. You know,
I think like Jake mentioned as well, you know, internally in our company as well, we're just
adopting these agents across all our different business lines. And what we're really trying to
figure out is how to get these people that are early adopters
and took the time and effort to figure things out.
How do we synthesize that knowledge
across the entire company?
So we're trying to host these internal kind of demo days,
events, hackathons internally,
and skills repositories, marketplaces internally.
So we're just trying to share as much in terms of the
shared resourcing layer around AI, because a lot of AI is just, you know, it's just prompts and
it's just context, right? So the more of that kind of stuff you can share, you can build on it.
And it's kind of like a constantly updating set of, you know, shared layer that you can build on top of.
Yeah, Jake, what are you saying?
Well, we're just at the beginning of our journey.
And, you know, I've run a lot of experiments
about things like sourcing companies
and just sort of moving them along,
you know, the research pipeline,
the due diligence pipeline, things like that.
So a lot of automation,
which I think makes us more effective. But I'll
say like the kind of the general, maybe like venture capital view is, you know, in venture
capital, these companies tend to have CRMs, right? Like they're tracking people, they're tracking
companies, they're tracking sort of deal processes. And the CRM is supposed to be the place where
like all the information about a company
that you know about lives and I think the reality of that is that's just totally not
true.
It lives in your CRM but it also lives in Telegram and it lives on Twitter and it lives
in email and it lives in you know a thousand other like data channels and data silos. And I think what AI is showing is that because agents are so good at
connecting these different data sources or even hooking into data sources that
don't have a connector today, I had an experience where I was like, can we plug this in? And the
and the agent was like, well, we don't have that.
agent was like, well, we don't have that. And I'm like, well, what are you waiting for? Code it.
I'm like, well, what are you waiting for? Code it.
So once you get all of that data in,
you actually see the real view that you have on this company,
on this team, on this process.
And I think that just sharpens the lens so much
of how people work together and the efficiency.
And I think it's going to be revolutionary.
Are you guys doing a go, no-go on investments
just with the panels?
I would caution against that.
It's very, very easy to create false results that way.
But what we do, the approach we take is like,
okay, if we have a thesis, have the AI give us an anti-thesis.
Like, tell us why we're wrong.
Tell us why this is a bad investment.
And then see if we can kind of use our knowledge and real-world data to overcome those objections.
I like that.
That's cool.
Great point. Yeah. So this feels like the only other reference point I've had to like this crazy enthusiasm
and just grassroots, like really focused on builder development mindset and even opening
this up to people who just aren't builders, which I think is like one of the coolest things
to come out of this.
If you can type or even like think it or speak it into a, I'm huge whisper flow uh user so I love that dictation I think it's amazing um
this is this feels like the early DeFi summer days and this is the only reference point that
I've got where everyone's just building everything that they can come out with there's iteration on
iteration I said when we're in the green room before we came on that every day I wake up in
this like existential dread of like I'm not keeping up but then like then the latest tool of the day
it's just like it falls off and then there's something else and it's like really really
difficult to keep up but like yeah how are you guys thinking about like what this kind of
grassroots like really really raw like development how are you kind of staying on top of these kind
of things how are you knowing what to implement and what might just be kind of flavor of the week style stuff and yeah just like
any thoughts on that kind of like really massive movement that we've got currently
yeah i mean one thing i'll say is that ai fomo is insane you know like and i think some people
get caught up in it way more than others and it can just like give you constant anxiety especially
for me you
know being head of ai at lefi i'm like i want to make sure that we're always using the best stuff
the most current stuff we're not missing out on something that could essentially like completely
change our game right so uh but it's important to kind of keep it in perspective and know that
a lot of the people you see on x and stuff are literally just engagement farm they're talking about these crazy workflows they have with like agent teams of 20 open claws
that are making them you know 10k a day doing something you know so like it's it's really
important to keep that in perspective so i think it's important to look at everything to test it out, but also don't get caught up and
importantly, come back to things that are actually going to enhance your real workflow and your real
use cases that you're actually going to use. Like one, one very important thing that we learned is
like the best AI tools that we rolled out was an agent that lives inside our telegram chats because the fact that it's living
where you're working is what made it so useful and that has gotten so much adoption internally
because you know you're able to just tag it in a chat you're already chatting with everyone it
and while we have all these other tools that maybe you're not getting used at all because you know
you have to go to a separate UI so it's really important to
kind of understand how people are going to use the tools where they're going to make most use
of them and uh and roll them out that way so that's kind of how I try to think about it when
I look at a tool and kind of like diligence this specific tool is I try to think of from that point
of view and is it actually going to be able to enhance something? Is it going to be able to be used where I can actually use it?
Yeah, I would say, like, to Raul's point,
a lot of these folks that peddle, you know, amazing solutions,
my finding is that at the end of the day, you know,
what all these people are trying to do, all these plugins,
all these skills, et cetera, are trying to do is they trying to help your agents organize their context. So as a matter of fact, it's often better to just do something simple for that because
it confuses the bots less.
That being said, I've done a bunch of work on basically like optimizing prompts like how would we create a
prompt that you know does what we want it to do you know better better than average or provably
better and that's sort of an interesting um an interesting exercise so there's some there's
some validity to kind of a better prompted agent versus a worse prompted agent.
But yeah, I see a lot of like unnecessary optimizations out there.
Just want to add to that as well.
I think one cool thing you can do is you can actually like point your agent at this stuff.
Like, you know, this Twitter article is like, oh, I launched this memory system that made my open claw a thousand x better then i look i point my open claw at it i'm like is this legit
and then you'll be like no this is not legit at all there's nothing here the one that we designed
together is already 10x better than this for these reasons it'll break it down completely
well also like if we if we talk about open claw right i mean everyone's like oh my god i installed open claw and the next day i made a million dollars
and became famous but i think i think the reality for most people is they spend like a month
wrestling with open claw to try to convey configure it i have a really close friend who
spends uh you know thousands of dollars a month in API fees to Anthropic,
just specifically like on his open claw setup. But it has permeated every part of his life. You
know, he's got glasses that he talks to the open claw through and everything like that. So,
but however, it took him, you know, months and months of engineering to get it to this point.
Yeah, the open-claw psychosis is real. Everything's extremely performative on the timeline.
People are running 20 different sub-agents and then you actually ask them,
where's the productivity game? And are you actually spending less time doing anything?
Because it doesn't look like it. OpenClaw has inherently been quite buggy.
I've had to debug basically every other day
that I've used it.
Memory system is completely short.
There are some good solutions out there,
but yeah, very, very performative on the timeline.
And as you say, the best solutions
are usually just the simple stuff,
like getting your skills correctly,
installing your brand guidelines,
installing your tone of voice stuff
just to help reformat and things.
So yeah, if anyone's listening,
you feel like it's existential dread
that you've been put into the permanent underclass
because you haven't installed OpenClaw
and bought 25 Mac minis, then it's okay.
It's okay.
You can do the same on vast majority of ChatGPT or Clawed.
So yeah, don't feel left behind too much.
I want to get into what you guys have been building with LeFi
because I feel like I've seen some of the submissions
and they were great.
I love just the way that people pick these things up
and run with them.
But yeah, how did this,
like how and when what happened for you guys to be like,
right, we need to push this and we need to push this now
because you guys could very easily get shiny object syndrome with a lot of the stuff that you could put attention to.
Like, how did you know this was the right thing to push towards and like, what was it that triggered that?
Yeah, I think, I mean, for us, we already kind of realized that our solution is already the best thing for agents to use.
Like, we're an aggregator layer that wraps bridges.
We wrap DEXs.
So really what an agent cares about more than anything is first of all, the price.
Second of all, the reliability.
And I think we can offer both of those by building an aggregated solution that provides redundancy across multiple different routes.
And so we have an API.
We have everything is fully open, it's ready to use. Right now,
if you go and ask your agent before we started any of this, your agent will already tell you
to use the Levi API if you talk about anything with interoperability. So what we decided is all
the pieces are right there. How do we just package this up, make it a little more discoverable, make it a little more easier to use, a little more friendly to both agent builders and vibe coders, essentially, since these are like the new groups of people that are coming up are people who, first of all, their open claw or agent is just building everything for them.
And second of all, people who are just typing prompts into clawed code, right?
So that's really where we set off to start to optimize things. So first we did the docs optimization. So everything is fully ready
for agents to just go and pick up everything they need. Then we did MCP server. So it makes it really
easy for you to actually build an agent that uses LeFi protocols. And then we came up with a hackathon idea so you know how easily can
you just lean into vibe coding and say that you want people to build cool stuff on top of lefi
so i mean i guess you know with all that i will say all the pieces were already in place it was
just a matter of kind of like you know packaging it up in a slightly different way to make this all kind of work in this new paradigm
yeah so what's what's in the toolkit you got um api mcp as you say the docs were more
positioned to being agent first and like what i will say is someone who's been like tinkering
around building stuff in the background like having really really good docs is like that is
paramount now for every single team.
So no kind of like half-arsing it and winging it.
If anyone's listening to the building,
go and fix that shit immediately
because that's exactly where everyone's going to be going
and just pasting them into Claude or Codex
or whatever they're going to be using off the back of it.
But yeah, what's in the toolkit?
And is there going to be a CLI anytime soon?
I know there's Perplexity guys are saying that MCP's finished
and then granola is
saying mcp is the future so like yeah how do you think about that yeah i mean first on that point
i've posted some thoughts about this myself i think you know mcp death is kind of being like
paraded around a little early i don't think mcps are dead at all i do think that there are you know
different use cases for each one but i actually think if you're building an agent to transact on chain, the MCP actually makes a lot of sense here. If you're defining all your tools up front to the agent, and you know, we don't have hundreds of tools, we have like five tools in our MCP server. So it makes sense, you know, you have all of them up front, you don't have to send the agent through loops of searching through CLI, help documentation, stuff like that. You don't need to do all that if you
just have a couple of tools in your very highly targeted agent MCP. So anyway, yeah, the MCP is
one of the things that we rolled out because we believe that it is actually one of the easiest
ways to get your agent to build with, with LeFi.
Then second of all,
we release skills for it.
So skills are essentially just wrapper around the API.
These go into the various skills stores like skills.sh,
which I think is a,
is a really good store that doesn't get talked about much,
but it lets you essentially install the skill into any coding agent and any kind of agentic framework.
And then we also put it on the Claw Hub store, you know, so Claw Hub, like Open Claws can just tell their agent to get it from the Claw Hub store.
And, you know, it passes all the security guidelines and everything like that.
So you don't really need to worry. And then the docs part.
So end of the day, you know, like I said, everything was already there
in terms of the infrastructure and tooling.
We just had to kind of like wrap it up
in slightly different ways
and make it more easy to use, more discoverable.
All our docs now have this like send to Claude button on them.
They all have that like .txt, .md extension.
So you can just like paste it directly in.
You can just tell your agent to go to the Levi website.
Then it will be able to easily navigate
because everything just loads up instantly.
There's an entry point in the docs that say,
if you're an agent, go here.
So then it will basically lay out
all the different ways that agents can use it.
So basically that's kind of the toolkit that we've rolled out.
And, uh, you know, we have plans for a lot more as well.
So this is all kind of like, uh, abstracting away the, the onboarding layer, but then we
have more plans for like, okay, if you actually need to, to onboard yourself onto a higher
tier, because this is all free tier API stuff, then how are you going to do that without
talking to a person if you're an agent? So that's kind of like the next level of this yeah like
i'm past the point now of people saying i don't know how to do anything when you've got like
documentation old tools like that literally if you could just point claude or chat gpt or codex
or whatever at that and then just say i am inherently an idiot i'm non non-technical, but I have an idea and I want to build it.
What I think is really interesting now is like,
you've had technical people that have came from like an existing route
into being a developer and an engineer,
but now you've got the people whose brain doesn't work that way,
but have like the champion of the idea guy is like,
is now coming online.
If you can kind of think it, you can kind of build it into existence and i think like that will inherently
come up with totally different ideas and totally different frameworks and totally different apps
and tools and just a weird form of bringing different things to life that the people that
think in that creative way that are completely opposite end of the spectrum
to like the more technical focused people.
I think there's going to be some interesting stuff
that comes out of that.
And I'm subscribing to the idea that
bringing products to life is like the new content form factor.
Like we've had video, we've had written,
we've had blah, blah, blah, blah, blah.
We've had it in person.
I think products are now a content form factor
that a lot of people are probably going to lean into.
But I don't know if there's any truth that any you resonate with on that or have you got any thoughts
around that that kind of theme yeah i mean for me personally i think it's a big changing point
because i've been a developer for like you know 20 years so for me this is like crazy because i see
some of my friends and stuff who i deal with and they never knew how to code. And then all of a sudden they're like building these crazy dashboards.
One of my friends built like a whole like trading bot that's just like doing all this stuff and had no idea how to code.
And I'm just like, wow, I feel like I just wasted my past 20 years of my life.
Right. So it's a lot.
Do you feel that or do you feel like, oh, like just pat them on the head and play?
Yeah, you go through phases because you
know obviously I was always like the dev so like my it's kind of like my gatekeeping thing it's
like haha I know how to code you don't know how to code I can build stuff you can't build stuff
you know but now I'm like okay well that's not my thing yeah eventually like I will be able to code
buy code even better than you because I know a little more about the architectural details.
And, you know, that does matter to some extent.
But you can get a lot of the way there just by having the right ideas and having the right mindset.
I think it's a lot of it is mindset, like being able to kind of like resiliently push through and figure it out, understand, go through this level of understanding where you don't really give up. You kind of know how to work with the AIs well enough to debug your own issues. And I've
seen people like this who are not technical, but they really have figured out and they can
vibe code like anything and they can actually like make it work because they have that ability
and that mindset. It's kind of like the founder mentality, you know, like I think founder
mentality people do the best in this new paradigm because you know like you said you're product minded you're kind of like technical enough but
you know what you want to get out of it badly enough and you know how to design it well enough
and jake how do you think about i know it's been a meme for a long time but
i i'm slowly starting to really subscribe to the idea that and i don't know what you guys are
seeing are teams reducing are you going to start seeing like one and two people teams that you could quite adequately
fund to get their idea out of the ground like i've seen um what's it called trust mrr like this this
this little startups that get started over a weekend they're getting sold for six seven figures
and they've only been getting spun up by one guy like do you subscribe to that idea would you guys
be willing to invest in like a- or two-month band?
How do you think that team dynamic plays out
and how venture can slot into that?
I think the long-term view on that is definitely...
I've absolutely talked to founders already
who are like, hey, my team is me
and a bunch of AI agents coding.
I think the ground truth reality of that is I don't, I mean, I'm sure there's some businesses
where that could just work out to end.
I don't think those businesses are big.
But I think the ground truth reality is that it's not fully there yet.
So I'll give you a simple example.
I talked to a great idea.
It's a startup that looks at regulatory registrations for different kinds of licenses.
This might be something LeFi has thought about before because it's, for example, like money transmitter licenses and things like that.
I thought about before, Rahul, because it's, for example, like money transmitter licenses
and things like that.
And they deploy an army of agents to, you know, go through the filings, like deal with
the government, the thing that nobody wants to do, right?
Now, of course, the problem with that is sometimes your agent goes to the government website
and it's down.
And, like, you've got to call, you've got to pick up the number and call and wait in
line and
do it a bunch and i don't think like realistically like the agents are kind of there where they can pick that up yet so there's still very much like a human in the loop in that type of you know
business um so i do i do think that like when it comes to digital services and and a lot of these
services are just kind of like a matter of
can you code it and can you get it distributed i think agents can can do a lot and and we're
seeing some of the impact of that in the market right like we've heard about you know meta sort of
starting to say okay we're gonna have some layoffs and rumor it has it that it's because
Rumor has it that it's because the 15 person team can now be automated, you know, in certain areas.
So it's definitely like interesting times.
I also think it's a huge opportunity for people, right?
Like across that whole stack, not just building your own business, but even, you know, one of myses is is uh the training and inference of ai models
like the ai production stack will just be completely decentralized despite all of the
prevailing narratives out there and we're seeing like massive validation of that right we're seeing
uh consumer hardware go locally apple's releasing extraordinarily powerful GPUs and MacBooks that I was just reading that's doing 108 tokens a second, you know, inferencing QN on the new MacBook Pros.
You have decentralized training.
So the massive, I just published a whole deck on this on Twitter, right? Like massive progress and people being able to train large competitive models
in swarms of GPUs, GPUs that can't really participate in data centers elsewhere.
So anyway, that's all to the point that like every AI related activity
is actually going to be the domain of the individual, right?
Like if you want your agent
to go participate in centralized training swarm that's actually possible today like you can go
make money contributing your compute on you know macrocosmos iota uh on bit tensor um soon we'll
see inference networks we'll see people at open claw'll see people at OpenClaw, you know, inferencing models locally.
And then what you're going to start to be able to do with those local setups in terms of generating revenue and building a business, I think will be will be phenomenal.
Right. And so kind of a lot of people tell me about their OpenClaws.
My measure of whether you have a good open claw is like okay are you paying
money to use it or is it making you money so like the moment when it starts making you money or
breaks even right that's when you know you got a good setup going yeah open clause like the you've
seen the meme of the guy at the urinal and there's like 20 different urinals between between where you could go and
where he is and everyone wants to tell everyone about it yeah that they um they distribute your
training um like you've seen hermes absolutely killing it from noose um this past week everyone
comparing it to open closing what's better about it like we've been banging around about it for
past 18 months what it feels like and now starting to see it all come together and subnets are
completely ripping it's like it's really really good i'm really like happy that that's actually
a thing and people are seeing the value in that and uh the other thing that i think about with
stuff like this is i feel like there's been infinite money thrown at teams that have just
taken way too long to get to market and by the time they get to market the the environment's
completely changed the landscape's totally changed and what was once a good idea or even the way that they were going
to deploy or even get their token to market like the the zeitgeist has changed and it's moved on
and i feel like it might speed time to market up iterate iterations can be infinitely quicker
um tools like auto research like test and iterate like on a million different parameters all at once with tens of thousands of agents,
stuff like that.
I just feel like it's going to change the way
to get to market as well.
And I don't know if you guys have got any thoughts
on that with regard to product
or just like initial company launch.
I think one thing I'll say on that
is around this enterprise AI thing.
We've actually talked to a couple of companies just to think about how to deploy things internally.
And it's funny to see because these companies that have maybe raised one, two years ago, they seem to be way behind now.
They're building on old stacks and they're kind of stuck in this old way of building and they can't really pivot up to like what is really the right way to build like now with new models new kind of agentic workflows
and things like that so yeah i mean i do definitely think like it really helps to be able to move
quickly and kind of like rebuild from scratch and now in this new world like you know with this
agentic development right you can just like rebuild everything from scratch constantly.
I think the best teams are the ones who are going to be able to do that quickly.
And, you know, it takes a different kind of company to be able to do that.
And like you can't have this like ingrained overhead of having these customers that you need to support in like an enterprise kind of SaaS kind of situation.
That was just my thought. to support in like an enterprise kind of SaaS kind of situation.
That was just my thought.
You know, I think maybe Jake has better thoughts in the crypto space in general.
Sorry, can you repeat the question?
Sorry, we're just saying time to market is going to change
with like all this tooling as well.
And what I think like has
happened previously like a team might raise in like say t equals zero it might have been 2022
by the time they get to markets 2025 one the landscape has changed drastically and what a lot
of negative connotation with a project actually happens where their initial distribution and then
and uh how they were going to get their talk to market was set in a paradigm where right that was the way that you got a token to market and then one two three
years pass and then that is just an outdated token distribution model it's an outdated way
of actually going to market maybe their underlying infrastructure and the product is completely
outdated so anything that can get you to market from the point of ideation to launch if any of
these tools can speed that up and get you there i I think it's going to be way, way,
way better from an investment perspective.
But yeah, I don't know if you've seen anything
on that, if timelines are getting sped up or anything.
Yep.
Are we moving on from that one?
Sorry.
Right. I'll move on. Yep, are we moving on from that one? Sorry.
Right, I'll move on.
So with APIs, not all created equally,
we have had some horrendous workflows in the background where constantly we're having to debug.
But how do you effectively build out robust APIs
so every single time that you guys update something that the likes of us if we're going to use it for our workflows we're not getting
absolutely killed and having to debug things all the time like what does that look like for someone
who's inherently non-technical building out something that's quite robust so like the teams
aren't going to have to like constantly dig in what's broke all this is broke again it's an api
issue um and then you're going to have to rip out the whole workflow and then start again like how's that what's that look like yeah i mean that's a good question and
the fact that you said for somebody that's non-technical i think that is where non-technical
people actually are gonna run to the most issues is building something that's scalable and reliable
and i think you know that is really where the inherent like engineering part of it comes into play, because you have to understand, you know, where the bottlenecks are going to be, where the brittle parts of the product are going to be.
And you're going to have to understand a little bit on how to kind of prompt your agent to understand these things, because, you know, a lot of times you guys will be talking to agents, right? And you'll, you'll be like, oh, you know, you'll tell them something about like, oh, encrypting.
Oh, you know, can you make this like encrypted situation for my keys? And then, then you'll be
like, oh, yeah, yeah, sure. Let me do this. And you'll be like, okay, so where are you actually
going to store the encrypted, like the key, the secret? And then I'll be like, oh, I didn't think
about that. I was just going to like do it, wing it. So like, you know, what I'm trying to say with
that is that the agents aren't going to think about everything
upfront, even though it will sound very coherent. And like, it's thought the whole solution through,
you can poke holes into the right places and you can figure out like where actually is missing the
gaps. And the only way you can really do that well is to have both a very good understanding
of the architecture enough to
be able to understand where the brittleness is going to come from and where to kind of harden
the system or you have to have really good test procedures and be able to kind of test this at
scale so you know to answer your question yeah it's going to take a little bit of of technical
knowledge and know-how and very good test procedures.
And sometimes you just have to let it break and figure out how to fix it.
My experience is if you want good development results, and by the way, I have a background in engineering.
the way I have a background in engineering I used to work at Amazon as
I used to work at Amazon as well.
well is you you have to you have to have a really clean spec that teaches the
agent exactly what you're trying to do and there's I think there's a little bit
of an art to creating that spec you know plan mode is great but there's there's a
lot more things you can do to to create end-to-end clarity.
A lot of times, you can run the agent through loops like,
hey, is there anything ambiguous in the spec that you don't fully understand?
Let's go through it and make sure it's explicitly laid out.
And then when you get to good clarity,
I think that's when, you know,
you could expect a good result.
So like taking it from chat to code,
maybe even hitting it off against if you're using code,
go to code acts,
like just trying to get as much reference in it as much input as physically
possible until it's like bulletproof.
You don't need the code,
you don't need to code,
but you need to create,
you can have the agent create like the plan in a way that is maximally unambiguous.
And I think that's kind of the measure.
But I do think that's a good point that you brought up too, Grant,
is about putting it from Cloud Code into Codex.
Because I do that too.
I'll plan it all in Cloudude code and i'll be like
okay codex why don't you review the plan that we just came up with and then it'll find stuff and
then you'll put it back in cloud code i mean i'm partial cloud code for everything but i do have
codex do my code reviews sometimes so i do find that that helps to to have like another set of
eyes on it yeah i've seen peter from open claw explaining kind of uh
clawed just like this adderall filled engineer that just wants to get shit done and then codex
is like the little nerdy guy in the corner that is always right though so different personalities
and different models for different use cases um so what um what have you seen anecdotally like
what have been people been picking up and building with with these toolkits um yeah is there anything interesting is anything that you'd like to see
get built if you weren't kind of preoccupied with building a company yourself you know
yeah i mean it's a good question i think you know a lot of people are attacking low-hanging fruit
around like trading trading yield farming is a good one, you know, especially because it's a little different, right?
Because the API we provide is bridging and swapping.
And a lot of people who are building trading bots are focused on the perp side, which makes sense.
You know, you can long, you can short, it's better kind of for a trading bot itself.
But, you know, seeing some of these things like yield farming, you know, that requires you to be on multiple chains, swapping between different assets and really kind of make use of the APIs that we provide.
I think treasury management is another really good one.
Like there was this one about like using OpenClaw to manage your treasury, which, you know, was really interesting because I was like thinking about that exact idea like that day.
And then I saw that somebody had actually submitted that as a hackathon project I think that's like a
that's a really interesting one right it's like using your open clause system essentially as your
treasury manager because nowadays you can you could buy stocks online are on on chain right
so you can actually like have like a proper like trad five portfolio that has like stocks metals commodities
currencies and you know you can have your whole like boomer portfolio all on chain and then
because it's all on chain it can all be managed by agents as well so i think there's there's a lot of
these kinds of uh products out there that are are ready to be built now with the tools that we have
and uh and you know we can make this stuff easier, of course.
Yeah, like the rebalancing of that
has probably been one of the hardest things
for teams to actually do,
if it's gonna be non-custodial as well.
Like there gotta be someone building
funding rate arbitrage using the LeFi.
There's gotta be, like,
that seems like the lowest hanging through ever.
There's probably some like news aggregation stuff into Polymarket that seems like the lowest hanging through ever. There's probably some like news aggregation stuff
into Polymarket.
That seems like low hanging through.
Like I just don't have time personally
to go and build this stuff
or I'll go down a rabbit hole
and just never come out again.
But like it's just opening that design space up,
as I say, for someone who's inherently non-technical
but knows the industry
and knows what they would usually employ someone
to go and build.
You know, like you'd usually kind of like sit down and say,
oh, there's an idea and it's a non-starter because again,
it's just totally non-technical.
But like, I think if you can type or again,
like we've mentioned with Whisperflow,
just speak into your laptop,
speak into your computer now with like with these tools and this toolkit
like you guys have brought out.
It's like opportunities. Like if you didn't think it you can bring it into existence now which i think
is like one of the most interesting times to be alive yeah that was great you just you just like
brought up like three really good ideas right there too that that you know anybody who's
listening just like try to build it in your free time it wouldn't be that hard so there's uh there's one thing i want to bring up
just before we wrap it up i know we've got about 10 15 minutes so the the agent commerce side um so
we're seeing like a multiple people attack it from different angles right so you've got the big
incumbents visa mastercard i wouldn't necessarily class striper as an incumbent but they're building
that with tempo um you seen Ramp come online now
with like an agent first kind of virtual card.
I think the missing piece
from what the incumbents are missing
is the programmable money at the stablecoin level,
like literally at the token level.
And I think a lot of the security assumptions
and the risk assumptions and the hallucination factor
that they're trying to just kind of shoehorn in
just with regular like legacy stack. I think they're trying to just kind of shoehorn in just with regular like legacy stack i think they're gonna have to bend and launch a stable coin because
i think you really need it at that programmable token level to enable that to happen am i
completely wrong is there any pushback have you got any thoughts around how this plays out i know
we're in the very very very early days of this but i i think it's going to be huge yeah i this is a this is a big area yeah
and a lot to talk about there but i mean maybe like taking a step back one thing that we could
probably all agree on on this call is that there's going to be a lot of agents like millions of
agents maybe tens of millions of agents like very soon. Rahul, how many agents do you think there are like an open claw like, you know, this year?
Probably a couple hundred thousand at this point, at least.
So I've seen I've seen kind of like baseline estimates that like by the end of 26, we'll have maybe, you know, two to two and a half million OpenClaw agents. But then a founder from China was like,
you're totally discounting the Chinese
and how much they love OpenClaw
and how many of them there are
and how much infrastructure there is to run it.
That's really been taken up the past few weeks.
I've seen a lot of news around that.
That's right.
That's right.
So we're talking like,
let's call it a bull case of 10 million agents
or something over the next year, two years, something like that, if you take the Chinese adoption into consideration.
And all these agents, they truly need services, right?
Like you're asking them to go and do something on your behalf.
do something on your behalf the agent is making product choices it's saying you know i will choose
this open source framework over that open source framework i will choose this web fetcher over that
web fetcher and you know in my case i'm like go deploy my open call or whatever and it's choosing
this vps over that vps and um so you really get a sense that, you know, in the end state of this, agents will be going to do a lot of commerce.
They'll be going to, you know, digital services, you know, compute storage.
They're going to need some identity services.
They're going to need, you know, like various tools like for memory, for context management.
They're going to need accounts.
They're going to need payment rails.
They're going to need all kinds of stuff.
So it really gives you a sense.
This agentic commerce world is really, really big.
And how do you, as an agentic services company, win that is not clear
because also the agents have very very low
switching costs if they don't like this service they can very quickly go to another service they
can literally build around a paid service i've seen agents who were sort of unsatisfied with
the pricing of fire crawl go and just build an open source version of fire crawl and then like use that and it works you
know 90 is good but it's free right um so there's a lot of like nuanced questions there about like
what is you know what is what what is defensible what is not but one thing that is very clear
is that um you know it's going to be very big. Now, the central question, I think, for the crypto world is,
will agents use Web3 Rails?
Will agents use Web2 Rails?
And I think that's a very nuanced question, too.
I think a lot of Web3 people want them to use Web3 Rails.
Nobody wants them to use Web3 Rails more than I do.
And I think that also using Web3 Rails is very cheap for them.
So to the extent that agents have economic pressure to lower costs for whatever reason,
that would be an interesting choice to do interagent commerce.
But there's also interesting arguments to the contrary.
Because I think this goes even deeper.
It's saying like, well, is your agent sort of legally you?
Are you the person who's deploying the agent and you have the liability for everything?
Exactly.
Or are agents kind of like trying to be separate legal persons which is obviously not the case today in in any
jurisdiction but um you know maybe that's where we kind of like want to go and you know so today
there's i would say like most agents are kind of an extension of you legally like they can't go
they don't have an identity that the government will recognize right and and so they can't get
their own bank account and so they have to use kind of like your bank account, your credit card, et cetera. But in the future,
if there is a growing set of agents that are autonomous, and again, autonomous, what does
it mean? Does it mean legal personhood?
Does it mean like technologically autonomous and self-sovereign?
Can't be shut down.
Does it mean something else like just an agent separate from you with its own identity?
But the more autonomy they have, the more I think it becomes likely that they want to do interagent businesses that would then really make sense to do on Web3 Rails.
But we're kind of like not in that world yet.
And the volumes of the different protocols that are out there, for example, X402, are very low.
I mean, they're measured in like tens of thousands
of dollars a day.
This is nothing compared to,
this is literally zero, right,
compared to the broader economy.
So it will be this central question will be very, very interesting to see how it resolves.
And it could be that both things are true at the same time.
I think there are cases for both of those things.
Yeah, I think it's very much a chicken and the egg thing. And, you know,
like I like to say, LeFi, we're laying the eggs right now, you know, because there's like, you
know, like you said, it all makes sense, right? X402 makes a lot of sense. The agents need to pay
each other, there's going to be a world where agents are running around doing things for people
and needing to pay for services, and they have full degree of autonomy but we're not there yet but we need the infrastructure there or else this
stuff will never be able to happen so you know as small as this market is i think the way it flipped
just like that with open claw it went from like you know zero to like hundreds of thousands to
you know millions soon i think we will see something similar like that
on the agentic commerce side as well.
And there will be something that just goes
and clicks for everyone and just makes it happen.
I don't think any of us know what it is right now,
but I'm a firm believer that that's coming.
I think Rahul, we can agree,
from a LeFi perspective
like from a LeFi perspective or from a DeFi perspective, right?
or from a DeFi perspective
that
this is a growing
market of participants
in the kinds of networks
that LeFi is interacting
with and the kinds of networks that exist in DeFi
and even
before these questions
of personhood autonomy
get figured out we can be absolutely sure that agents are going to use LeFi to do exchange.
I mean, they're doing it today, right?
Exactly.
So there's always that. That process get really big and actually lead the way and shape how these policies are later
implemented just because of the practicality of the fact that this is what's happening.
Well, the missing ingredient though is scale, right? Are these agents going to
explode a number and start using DeFi and get huge numbers?
Yeah, you guys must be quite happy with where you're at. Getting out in front of it nice and early,
get those compounded network effects.
I feel like the GEO,
the agents,
LLM, optimization for content
and stuff, those companies are probably
going to do pretty well because a lot of companies
are going to do really, really well if they're indexing
and getting shorter
sources in a lot of these models.
We spent a lot of time like working
around this as well but um yeah if gents is there anything i might have missed i know we've covered
an awful lot there but uh yeah i want to open the floor to you guys if there's anything you want to
make people aware of or what you guys are working on uh outside of that or yeah is there anything I might have missed?
Yeah, I mean, like, again, one of my biggest sort of themes this year is around sort of decentralized networks doing, creating new AI functionalities.
We just had a Tether announcement, by the way, today, this morning, which is huge regarding being able to use consumer hardware to train and serve models.
Our company, Pluralis, has done some incredible work that I've tweeted about recently in that space.
And again, I think agents is one half of the equation.
The other half of the equation is how do they make money
and how do they improve themselves also, right?
Like in some of the most obvious ways that that's going to happen,
at least to me, is through these like agentic swarms
that are creating new AI functionalities.
We've seen Andre Karpathy's auto research being done by an agentic swarm by a company called Ensue and they'll also be self-improving their own AI, participating in swarms that help other people do that.
And, you know, monetizing your idle compute in networks that are like inferencing and training other AI.
It's basically AI all the way down, guys.
And I think to summarize what I'm saying is,
I think, you know, AI and agents
are some of the most decentralizing technologies
that ever have been.
And I think we'll see that.
Wow, you just blew my mind there.
I think you just defined universal basic income for everyone.
That's cool.
Elon said there's going to be universal basic abundance and we're not going to need to work ever again.
I don't know how much I agree with that, to be honest.
Well, no, I think Jake's right, though. if you do decentralize it properly then i think you
do get to that point but you know if it's all owned by like open ai and anthropic then you know
we all get screwed in the end so really that's why we do have to push for the decentralized angle
more than anything great so as um is there any submissions still open or is it closed or would
just like people to
continue to see people building on top of it or like what's that look yeah well this this current
vibathon is over but we are working on on the next iteration of this stuff so we'll constantly
have stuff so you know if you're building stuff just just hit us up and there'll always be ways
for you to to uh get supported by us.
So, you know, we'd love to work with anybody who's building on top of us.
Great. All right, gents. Thanks so much for your time.
Appreciate you jumping on. I know you'll be incredibly busy.
So, yeah, thanks again.
I'm not in tomorrow. If anyone's tuned in for the Blockmates podcast,
I will see you next week.
All right. Henry, you can take it to work.
