Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
The Future of Autonomous Protocols with Talos, Arbitrum & Oasis
Recorded: Aug. 13, 2025 Duration: 1:01:28Space Recording
Short Summary
In a dynamic discussion, key players from Arbitrum and Oasis unveil Talos, an innovative AI-driven project aimed at enhancing governance and decision-making in Web3. The conversation highlights strategic partnerships, community engagement, and the growing trend of integrating AI into blockchain workflows, signaling a transformative shift in the crypto landscape.
Full Transcription
Thank you. Hello, hello.
I'm waiting for Ben and Will to be speakers, at least on my screen.
You're not yet.
And then we can start, I guess. Will, does it work for you okay until will joins we can already start i can do either he is perfect i could have done the oasis
intro part but i don't think we need to because now will is here let's do a very quick intro
round but then actually focus on the topic of today which is talos and how to build these
autonomous agents how to collaborate with them, etc.
I'll just do it by order on my screen.
Ben, please introduce yourself, what you do, and what your connection to Talos is.
Hi, everyone. I'm Ben. I work as a DevRel at the Arbitrum Foundation.
And the Arbitrum Foundation, along with Offchainchain Labs helps partner on Onchain Labs,
which is an app incubator to try and explore fair launch experiments and novel governance models.
And Talos came out of that.
It's actually the first project from Onchain Labs.
So really super excited to be joining the space.
Nice. I don't want to be moderating too much.
Let's keep it an open panel.
Anyone can jump in. Johnny.
Hey, I'm Johnny, the founder of Imperial, and in partnership with OnChain Labs, I've deployed
Talos.
Hey, everybody. Hopefully you all can hear me. I'm having some connection issues, but I'm Will. I'm
ecosystem growth manager here at Oasis and I guess a good overview as Oasis is
that we try to solve for everything privacy and confidentiality focused. So
we have our Sapphire network which is a confidentiality enabled EVM network, but
today the focus will be on our product called raffle which stands
for runtime off-chain logic and we'll get into that more as we get through the discussion
yes and let's get into it johnny it would be nice if you could take quite a few minutes
to explain tell us how it got created i mean now we understood it's part it was created within on-chain labs
but still like how does it work what does it do why do i see it so much on my timeline i don't know
who's doing kol stuff but it's been going well uh jokes aside i do think it's natural organic
but really i see it everywhere but i don't know if lots of people actually have a good
understanding of what it's supposed to be.
So if you could enlighten us, please do.
Yeah. So it came out of conversations with the team at Onchain Labs.
But I think to take a step back, one of the things that I'm sure a lot of people that are kind of building in Web 2 or Web 3 have noticed is that we're starting to rely more and more on um AI for supplementing our workflows and I think this
goes for like everything from like writing emails their documentation to actually doing coding um
and a lot of times for brainstorming um I know I'm not the only one that when I have a question
now I'll open up like GPT-5 and I'll just talk with it and kind of work through a problem.
And so I think that that trajectory has ramped up a lot in the last year.
And I think it's continuing to ramp up.
And so I think what we're going to start to see is where we have more kind of native integration with AI as part of project development and workflows.
And so one of the things that was really interesting is, especially in Web3,
it's very much about unifying different nodes in an ecosystem.
And so having this sort of open governance model where you can try to put an AI agent
or sort of like this AI workflow at the center of your development process,
I think it's something that becomes more and more valuable as we see the quality of artificial
intelligence ramp up so that you can get to a point where you can have more and more of
sort of the decision-making pushed towards AI-oriented governance while still having
sort of like a human kind of board.
So, I mean, you can almost look at it like building a AI CEO,
but having a human board to kind of keep the CEO in line.
And then I think the other thing that I think is important
is the proposal mechanism.
So trying to make it so you have this sort of open governance model
where anybody can try to create a proposal,
whether it be protocols that have features
they think would be beneficial to the model
or a feature they think would be interesting
for the model to implement.
And then sort of having this process
where you have collaboration between the AI agent
and the human sort of counsel that can guide it.
So before we dive into actual Talos functionalities,
what kind of apps or agents do you think
really need an open governance model?
In what sense?
Like for which use cases is it actually relevant?
I get the AI CEO because that's really crucial decision making
and you do want the kind of board to oversee it but lots of other agents i assume those maybe that
don't hold any user funds they don't really need it yeah i guess i mean if you think about it like
a company like meta or google is already extremely data-driven and they're using machine learning
models to drive their decision making already so i think that this is just kind of the trajectory
most companies are on whether they kind of codify it as an ai agent or they say we're using machine
learning and artificial intelligence to guide our decision making and over time i think most
companies will rely more and more on AI as
like a central element of their workflows. So I do think, you know, if you're managing,
for example, like your personal portfolio, you're probably going to see more and more reliance of
like the average person on artificial intelligence to give them like more fine-tuned reporting on
the performance of their investments and more kind of customizability on how they allocate their funds. So I think
it's just something that's going to be across the board integrated more natively into workflows.
I don't think every project needs like a AI CEO so to speak, but I think that blockchain is like
uniquely suited for this because it's a industry that really like rallies around
innovation and i think that if we can embrace these sort of workflows now as we see these
improvements in ai it'll become more and more natural and so the feeling is in a couple years
if we see the kind of like improvements that we've been seeing in the quality of these models
talos will actually be smart enough
to do more and more tasks on its own.
Okay, so what does Talos do?
So right now, what we have is sort of an AI agent at the center,
and this is kind of where Raffle sort of fits into the puzzle.
So if you have an AI executor inside of a trusted execution
environment, and then you have the ability for upgrades to be pushed to that environment,
now you have a AI agent that has very fine grains, open source upgrades being pushed to it that can
be verified by any external party. And then it's able to do things like control a private key and not have anybody outside
of the agent itself in control of that private key.
And then over time, if you can actually have the AI agent where it controls its own upgrades.
So right now, the goal will be to have via the council, this fine grain control over
upgrades being pushed, potentially with like a time lock so that if an upgrade is getting pushed and it looks
malicious or a person is not in favor of it, they have the ability to see that
and make a decision before the upgrade actually goes live.
But over time, you can imagine if you have an AI agent that can actually
upgrade itself and control those upgrades in a permissionless way,
upgrades in a permissionless way, you actually have like a sovereign AI.
you actually have like a sovereign AI.
Ben, how is now Arbitrum involved after the incubation?
Is Talos supposed to be actually managing some Arbitrum treasury?
What's your plan?
Uh, that's a good question.
Uh, I think, uh, can you hear me?
Yeah, I'm on, right. Uh, so I think it really, we continue being involved by offering expertise, by offering the network dev tooling, etc. Whether it manages on-chain arbitrage and treasury, I don't have exact details yet for that, not even alpha drops.
Not even alpha drops. But that's a good question exactly in what ways it would measure that. I will say that I'll add to what Johnny has said before. I think this is very exciting in the fact that what you're seeing across the internet and across spaces everywhere in Web 2 and in Web 3 is everyone institutionally and also personally are relying upon ILL reasoning, are relying upon
machine learning, etc., and personally and institutionally for financial decisions.
But what we are offering here, and I think what is super exciting about this, is not just the
relying upon machine models, relying upon, you know, AI agents, but it's relying upon it in a verifiable transparent and secure public way
that's that is open that can be seen by all that can be verified by all that uh in is not in a
closed box and you look at the googles and the aws of the world and you look at the azures of
the world and they're all offering services like this and closed box systems are being built like this everywhere.
But the big difference is in that on-chain verifiability and
secure trusted environments and we're going to get to the TE is,
I'm sure in a bit as well,
but it's secure trusted environments.
That is I think the major differentiator and really what could
be seen as just a significant watershed moment
with the rise of of talos and they are a different kind of closed box fortunately they are verifiable
and have lots of other benefits but yeah good handover to will so you've actually known johnny
forbids or johnny and imperial have been working with the oasis stack for a while
why did oasis get involved with talos
oh he's a listener now
yeah i don't hear him
I think he's rejoining.
I can talk about how we got involved with Oasis, if that's helpful.
Hopefully I'm not interrupting someone, but I heard your question, Marco.
Yeah, sorry that I'm having some connection issues.
But yeah, as you said, me and Johnny have known each other for quite some time.
I think, you know, back in the Emerald days and then a lot more.
So I think Imperial won our first hackathon called Keep It Confidential.
And I think that's where sort of some of the initial ideas came out of.
So we've always been in touch.
We've been talking about, you know, the adoption of TEs and then more recently the adoption of AI and just like what needs to be solved in the Web3 industry.
So I think when he told me about Talos and the work that he was doing there, I just thought, yeah, this was the first time that, you know, we really had a great opportunity to work together and to, you know, potentially transform the industry in a beneficial way. Adding on to some
of the points that Ben said, I think what makes crypto so powerful is the transparency, is the
auditability and the openness in governance and the decision-making that's made in the protocols
that you're supporting and involved with.
These are not things that you have with Google and Amazon
and things of that nature.
But at the same time, what we've seen is that
potentially the governance process for DAOs,
as well as, you know, pushing upgrades, even if it's a slightly more centralized, decentralized application. You know, it's
very cumbersome. It's very time consuming. Reaching quorum is hard. And then you have,
you know, while you want transparency, sometimes you also need privacy as well. And so I think that,
you know, the combination of Talos plus, you know,
the Raffle stack basically gives you all the benefits that you get with the Web3 side of
things where things are open and trustless and everyone can be involved. But it also gives you
all the benefits you have in the centralized system where you're still getting, as Johnny said,
data-driven decision-making.
You're still, and then you're getting sort of the leanness now where you can move things along a lot quicker and sort of solves what I've thought has been like a critical issue
in Web3 for a long time, which is the speed of moving through a DAO governance process.
So yeah, I think that's my general thoughts on that.
But yeah, was there more to that question that maybe I didn't hear
because I got disconnected?
No, no, that was actually everything.
But you could go a bit more into depth into the specific features
that come from running agents in TEs and maybe specifically Oasis?
Specifically Oasis?
Yeah, absolutely.
Yeah, absolutely.
So I think that we've seen agents sort of become popular in Web3 over the last year, year and a half.
But the major issue that we've seen, and this is, I think, the community has brought this to Johnny's attention.
I really commend Johnny for being one of the first to solve this.
But is that, you know, when you have a Web3 based agent, well, all transactions in Web3 involve private keys.
Right. And so if you're allowing an agent to sign on behalf of a multisig, sign on behalf of a DAO, sign on behalf of an individual user,
that means that you actually have to, you know, give
that user your, or that agent your private keys. And pretty much what we've seen from every other
AI agent, you know, provider is that, you know, they run it in some centralized Google Cloud,
or, you know, some Amazon or Microsoft Azure or something like that. And that means that essentially, you know, they have access to it.
Or if they use, you know, confidential VMs, you know, it doesn't mean that, you know,
Google or decentralized entities don't have access to it.
And also it gives you sort of, you know, a singular point of failure where, you know,
Google Cloud goes down for a couple of minutes or something like that.
Well, now you have no ability to make
the transactions. And potentially, maybe if data is lost, you lose the private key forever or
something like that. I think Web3 has taught us that redundancy and decentralization is great.
So what Raffle does is it uses trusted execution environment technology to make sure that no singular person or entity, or actually really
anyone at all, has access to a private key. But still, the AI agent is able to sign on behalf of
that private key. And then remotely attest to on the blockchain that things have been done
in the correct model, and that the transactions have occurred in accordance to the governance decisions that have been made
and things of that nature.
So we submitted a TIP proposal,
which we'll probably get into what the contents of that are,
but just a proposal for how we can improve Talus
using all the benefits of TEs.
I mean, you gave quite a bit of the answer
to something that I am always curious about because
as you said like we had this defy agent hype for quite a while but if i'm real there isn't that
much adoption like i personally use one yield optimization agent i've tried a few trading ones
copy traders etc but if we're honest there isn't too much DeFi agent adoption. And I do wonder why. Maybe you guys have an idea.
I can jump in there just to start off the conversation by saying, I think a major reason
why perhaps we haven't seen as much DeFi agent adoption until now is that the tech really wasn't
there for a long time. Anyone who tried using LLMs to reason out coding a year ago and then making to a reasoning agent that I can trust and hand that over vis-a-vis financial decisions.
You really want to be at a place where you feel that you can have some sense of confidence there. the earlier iterations, for all the benefits and drawbacks they had, were maybe in a bit lacking
some of that confidence, considering the state of LLM reasoning and where it was even just
seven, eight, nine months ago. And I think when you combine LLM advancement along with things like
high throughput and low cost settlements and secure execution environments. You start creating the context by which this starts to make sense,
not make sense in a theoretical way, but make sense even more so in a practical,
I can create and execute on this idea sort of way.
So the tech part is mainly for you focused on the actual,
I guess, confidence people have in LLMs.
Do I get that right?
Definitely, it's a significant component of it.
You know, the LLMs are essentially the underpinning of the reasoning that's happening behind there.
And so combining that and the advancements you get within the current state of LLMsMs along with verifiable on-chain execution environments,
you start creating a context by which that can become feasible nowadays, which I would argue was
probably less feasible several months ago and certainly a year ago, although very aspirational
but less feasible. And I think we're finally at a place where it is actually feasible.
I mean, Johnny, you might disagree
because Simi was quite successful
or is quite successful.
But what do you think?
Well, I think that,
no, I would say that
there's a big difference
between building an AI agent that is like interactive and interesting and actually controlling a treasury.
And so one of the things I think that we're starting to see is as these reasoning models get more advanced and you're able to deploy them in a more reliable way, you can start trusting them with more capital. And so I think the other thing is there's a pretty big difference between a language
model and a mathematical model.
So a lot of the tasks that you want to do in DeFi are more based on using mathematical
models for the execution of strategies,
like rebalancing your portfolio based on the current APR of a specific fault
or something like that, right?
Where it's not something where you're really having a language model
read the docs and then get the numbers.
I think the example people keep giving where it messes up basic subtraction
is a good example where reasoning models aren't defined.
Like, they're not their target use cases in math.
But what you're starting to see is better architectures
for connecting together these multi-stage graphs
of execution.
So you'll have a reasoning model with better tool integrations
and more reliable sort of execution.
So it can actually deploy these resources more efficiently.
And in a way that's less likely, like I think that people initially were really concerned about hallucinations from the different LLMs.
And I think we're starting to see better practices in place to prevent these kinds of risks.
or practices in place to prevent these kinds of risks.
So I just, I think, I mean,
like my favorite thing to share
is if people have seen the Will Smith eating spaghetti,
where they, like that's almost become sort of a benchmark
for image generation models.
But if you look at how much worse it was two years ago,
I think that's kind of a really good visual representation
of how much worse the actual LLMs were two years ago.
And so I think you just have to realize that there's this turbocharged trajectory that this
is on. And I think it's across all domains. I mean, the last thing I'll say is, especially for
coding models, teams like Devon and Cursor, Windsurf,
what they're doing is they're getting better
at building coding agents.
And so that's not just training a coding model,
it's building better systems for the agent
to be able to code.
So, I mean, it's not just like,
hey, let me dump all this code into an LLM's context
and see what it comes up with.
It's actually building like a workflow. And so
we're seeing kind of amazing improvements in the quality of these types of tools.
And so I think that's a really big thing is that you don't really want an LLM to do math,
but an LLM can very easily build an algorithm that solves math on behalf of it and then call
that as a tool. I think that is spot on just to add just one more bit, which is there's also a differentiation.
Sometimes these things get conflated between prompting an AI versus agentic AI.
And I think what is so novel with Talos is that it's an agentic AI.
It is autonomous.
It is making decisions on its own.
It's managing a treasury.
It's issuing tokens. It's doing everything using its own parameters that have been set in advance and in cooperation with its community. But it's agentic. And that is so novel and so interesting, more so than prompting and getting results, which is also really interesting, of course. But this is just, in and itself, a level unlike all that that came before.
And really, the advances in agentic AI are from whether it's coding,
whether it's image generation,
whether it's creating a reservation for that steakhouse that you want to go to
with your friends or whatever it is,
or managing a treasury is really becoming so not only possible,
but just so well done as things continue to progress. And we're at a state where it's just,
the results are just quite amazing as evident every single day with what we're watching with Talos.
Yeah, absolutely.
And I guess just to add my thoughts as well,
I mean, I agree with everything that's been said.
I think that a large part of it has to do
with the improvements that have been made
in the LLM space and what can be done in terms of,
I don't know if a year ago you could build Talos, to be honest, I don't think that maybe it was possible, but I don't know if a year ago, you could build Talos, to be honest. Like, I don't think that
maybe it was possible, but I don't know
but additionally, I just think that
part of what we see is
this is just
like a natural progression. Like, if we look back
at like, you know, regular DeFi,
it took like from 2018 to 2020,
I'm sorry, 2018 to 2020.
You know, if we look at AI, it's like, you know,
we had sort of our meme coin AIs where you were building an agent that was, you know, a Twitter look at AI, it's like, you know, we had sort of our meme coin AIs where you were
building an agent that was, you know, a Twitter bot or something like that. But, you know, web
three powered, you know, that was like six months ago. And now we're finally getting to the place
where, you know, we're building real, real sort of DeFi AI that is also doing it the right way.
They're doing it, doing everything verifiably, doing everything openly,
and doing everything where the community
has all the information they need.
And I think generally what you see
and what I think you can expect
is that as Talos sort of proves this model
and proves that this is possible,
you will see this become a more popular idea in the future.
And that'll power the next wave of DeFi AI, because that's sort of what we've seen in the past.
And I don't think those trends tend to change that much.
If you ask me, and you didn't, but I can ask myself and just share my thoughts.
I do think besides the llm part i think
it was a big trust thing like everyone in crypto at some point has been scammed or rugged or did
a wrong transaction and now suddenly you have these agents that promise like nine percent yield
on usdc and i'm supposed to just transfer funds into an agent controlled wallet with like zero guarantees i have no idea
who's controlling the private key of the agent as well mentioned like it's not that trivial to
actually separate yourself as a developer or builder of this agent from the funds uh likewise
the execution just happens in this off-chain black box and i just see the results on chain but i had
like really zero insights into what was happening with these agents so i am definitely still not an early adopter of
like new defy agents just because i have higher trust requirements before i start using crypto
dabs that are i don't know controlling my funds um but what is interesting to me is that you guys are so focused on the llms
and johnny since you mentioned it you said we're still on this turbo charge trajectory
i kind of had the opposite feeling like this week with the gpt5 release it felt for the first time
like damn okay we might be reaching the end of adoption or innovation coming from LLM improvements.
And now it's time to go actually vertical
and find niche use cases or use cases in general
and just build products that work.
And you don't have to wait for new LLMs to be the innovator.
What do you think?
I think that the coding benchmarks might disagree a bit
where we're just consistently seeing the models score higher on some of these metrics.
And people can say, well, we're kind of overfitting to evaluation metrics rather than building generally useful models.
But I think the thing is that, I mean, if somebody that just codes with AI, I think that, like I said, it's not just the model itself getting better.
codes with AI. And I think that, like I said, it's not just the model itself getting better,
it's the tooling that these companies are using to power these reasoning and coding agents.
But I think that most of the demos you're seeing are miles ahead of the demos we would see for a
coding agent a year ago, where you're getting full applications being built that have interactive components and they can add and remove features and change style.
And so my general feeling is that for a long time, the technology you were building was kind of your moat on some level.
Like, all right, so this took this many engineers, this amount of time to develop.
And so we have this head start on other companies.
But I think it's getting to a point now where a very mildly trained person that doesn't have like a strong technical understanding can put together a pretty complex interactive application in a very short period of time.
And people have called it vibe coding.
And so I think people look at it and they say,
oh, like all coding with AI is vibe coding.
And they have this sort of like judgment about it.
But the fact of the matter is 90% of the engineers
I know right now are heavily leaning into AI-based coding.
And I think that this is becoming more and more
a part of the workflow.
And it's increasing efficiency.
It's decreasing the amount of bugs.
You still have to have some like expert code review becoming more and more a part of the workflow. And it's increasing efficiency. It's decreasing the amount of bugs.
You still have to have some expert code review and quality testing,
like a QA engineer kind of.
But I just think that the rate you can get things done is increasing, and the amount of engineers you need to do a task is decreasing.
And so this is happening on multiple planes too.
So like with Figma, which is the design platform, you can use Figma make, and you can basically export a functional
UI from your designs directly. And that's something that would usually take weeks of
like going back and forth with the front end and fixing pixels here and there. And now it's just
like a single click export for a responsive web UI. And then if you don't like something,
you can say, hey, can you change this button so it looks a little different
and it will update the design and in turn update the front end.
So I think that the quality of these workflows
is something that can be developed much further.
And so we don't just need improvements in LLMs.
We need improvements in sort of like this full stack of development.
And so that's something that I think we're going to continue
to see a lot of improvements on.
And I just think there's a lot to mine there.
I'd probably just tag.
Like without any improvements in AI models,
just building models that are more custom tailored
to specific problem spaces.
I think there's just so much improvement
that can be done without some breakthrough innovation
in LLM technology.
So how does Talos or the committee decide
which model to use at which time?
So you have the sort of the reasoning model that we lean on for in the code base.
And so right now we're using GPT-5.
But I mean, that can be very easily changed.
I think that the more important thing is like sort of building out.
So the way we're looking at it is that a lot of people think of an LLM as just kind of
like a feedback loop, right?
Where you have like, you have your system prompt, and then you have your context, and then you have some tool calls that you enable to the agent, and then you just recurse on itself.
Like, hey, here's my input.
All right, so now it's going to call these tools, and then it's going to repeat, et cetera, et cetera.
But like the way that you're actually seeing these builds more frequently is as these graphs
that have different kind of capabilities that fan out.
And so you don't really have to limit yourself to just one LLM.
You can have, for example, if you wanted to build something that was going to research
current events around Oasis or something, you could have an agent that goes and reads
Twitter and it pulls in all the recent posts.
And then you could have another agent that goes and goes Twitter and it pulls in all the recent posts. And then you could have
another agent that goes to Medium and looks up different articles and extracts important
information that references OASIS. And then you can have different agents that go and do these
different sort of subtasks, and they can all use different models for their interpretation of that
data. And then that in turn can be aggregated and pushed to another model that goes and takes all this information and spits out sort of a summary of these current events.
So it's less reliant on like having one sort of LLM at the middle doing everything.
And that's kind of what you see with like deep research and like some of these other sort of features of like, you know, OpenAI and Anthropics interface
where they're actually doing
like a much more complex workflow under the hood.
And so I think that's one thing that we'll start to see
is like more interest in building
these multi-stage reasoning models
rather than trying to think about it
as this singular sort of node.
It's more about building a graph that fans out
and does a lot more sort of actions under the hood.
And the graph doesn't have a single point of failure
if one of the API providers goes down?
No, you would just lose access to that node in the graph, right?
So, I mean, you can build, it's just code, right?
So you can build your error handling conditions, you know.
All right, so this API provider went down. So then that would report back, It's just code, right? So you can build your error handling conditions, you know?
All right. So this API provider went down.
So then that would report back.
Hey, I tried pulling this data.
I did three retries and it failed.
And so this is, I don't have anything to report.
Um, so, I mean, you can build things that are resilient, you know, it's, it's just
at the end of the day, it's just about building sort of a graph structure for
the underlying reasoning.
Right. And I think, I mean, everyone should be building with this in mind because you can't
just be fully dependent on one api provider we are in a decentralized system and trying to be
built decentralized things uh i have to say one of the most exciting things when i first read about
telos was this open build style
where everyone can participate.
You guys started rather early on,
or I guess from the start,
with these improvement proposals.
And what I always see linked in Twitter,
threads, whatever,
is this Vitalik post from February
where he talked about AI as the engine
and humans as the steering wheel.
Was this the actual inspiration for it when you guys thought of this
at the on-chain labs incubator?
No, no.
I mean, I was really thrilled to see that because I was like,
ah, we're on the same page.
It's like I felt like it was very validating to see.
It's a page with Vitalik, right?
What is that?
I said it's always good to be on the same page
with vitalic right yeah that's what i mean great minds think alike right um but no i just um it
was one of those things where kind of just i mean i think that there's sort of this fascination um
and it's like it pains me to say but i think it goes back to kind of the waifu trend where you
people really wanted like an ai agent that was like this is an entity that exists on its own it's completely like a closed system and you can
interact with it and so you have like your ai friend but i just don't think that's the right
way to build like scalable ai architectures it's like you have to have upgrades i mean like this
industry is moving too fast for me to be like, this is my permanent AI.
You know what I mean?
It's in six months, there's going to be some new AI model that comes out that disrupts everything.
And you're going to need to upgrade to that if you want to have the best tech available.
And so like, really my thought was just as an engineer, like you don't finish code bases, you know what I mean?
Like you stop working on them eventually because you move on finish code bases. You know what I mean? Like you stop working on them eventually
because you move on to something else.
But there's no like concept of like,
oh, this is done.
You know what I mean?
And so you really need to have this ability
to push upgrades and modify things.
And at the same time,
I mean, anybody that's built in Web3
for more than a week knows
how kind of collaborative it is across protocols.
I mean, like when you're building something, you're constantly talking to other teams to
see what can we leverage from you to improve the way we're developing, what capabilities
are out there that will be good for us.
And these are things that are much better to have very fine grained control over.
And so having this like way where we can actually start to engage other teams in like an open way.
So instead of us going and saying like, all right, we decided we want to do this.
It's a lot better to have teams come to us and say, hey, we have an idea for how you can make this better.
And this is what we think you should do.
And then it becomes collaborative.
I think it becomes really beneficial.
We've talked with a lot of teams so far that kind of are in the process of like putting
together proposals where they just have something cool that they've been building and they're like,
I think this would benefit Talos and like we'd love to see Talos try it out or use it.
And so it almost jumpstarts the dialogue in a really productive way. Instead of, you know,
doing something where, you know, it's like, okay, we're going to go and look for the best solution.
This almost flips the script, where it's like, if you have a good solution, come to us with
And then we can kind of talk about it, vet it, and then potentially push it through as
an upgrade.
Yeah, I think it's really interesting.
I mean, I feel like it, like you said, it flips the script.
And I mean, you can just see it through Talos development.
It really just accelerates how
fast things move you know like oasis proposes something now we're building that while someone
else is proposing like uh I don't know maybe camelot or binance or one of these other teams
that haven't announced yet is building something else they propose it and now you know you have
a lot of people building in parallel you It sort of scales your engineering ability.
You don't even have to necessarily know
what the other hand is doing
in order to be moving forward.
Ben, I'm sure you have quite a few projects
currently going through the incubation program.
Is this setup something you would recommend to them?
By this, can you define the word this?
Yeah, this open governance model
with these transparent improvement proposals
where everyone can participate from pretty much day one.
Speaking on behalf of myself,
I definitely would recommend that.
I think it's a very sustainable model
going forward and one that allows for the need, which will always arise, to change things,
to alter things, to make modifications, which is just, you shouldn't factor that as a possibility,
it's an eventuality. And so that model, I believe, is one that creates the environment for that to succeed and the context for that to become possible.
So speaking personally on behalf of myself only, I think that makes a lot of sense.
And I would also say that the whole model of the TIP is really empowering.
The whole model of the TIP is really empowering.
You know, we often talk about organizations offering proposals and institutions.
But I, again, maybe I'm speaking a little bit more than I should, but I think that it also is open for people and individuals that have interesting ideas and ways to advance the model and advance the initiative.
Just from personal experience, I created a TIP for Talos as myself.
And looking at the model and looking at the way it operates,
one thing I noticed that was interesting was perhaps a need
for defining specifications for inputs and
outputs with a certain JSON formatting.
And so that was just an area that as somebody who writes a lot of code
and spends a lot of time writing
different applications
would be something that I would benefit from.
And so I created a TIP
with that specification in mind
and authored it as a pull request.
And I'm just a person.
I'm not an institution.
I'm not an organization.
I work for an organization,
but myself, I'm not an organization.
And so if I can be so audacious as
to say that other people should also think critically and offer their own proposals,
and that it's part of the whole model of being transparent, open, and decentralized,
that people can come in with good ideas and offer proposals to help make things better,
because the better it is, the better it works, and the better it will be for all of us
writ large. So yeah, I think it is a very long way of saying and the better will be for all of us um writ large so yeah i think
it is a very a very long way of saying i think it's a very good model going forward uh personally
speaking the question is also open to you two guys would i recommend doing things the way we're
doing it i mean hopefully right um i like it has its downsides right
yeah but i think that um so i think that there are protocols that should be designed as like
purely immutable like i think like uniswap they release you know a new version every couple years
and they have this purely immutable implementation and that makes sense for those types of projects
but i think um i think that depending on what you're trying to build,
if you're trying to build something
where you're trying to do something ecosystem oriented
and kind of cultivate community around it,
I think having a process like this is very strong.
I think if you're just really focused
on a long-term development trajectory
to build something very specific
and you kind of need the benevolent dictatorship,
which is typically a very efficient sort of paradigm for software
developments, then it makes more sense not to lean into this.
But I think what we wanted to do with this was have something that was very open and
transparent and ecosystem oriented.
And so that's why we kind of drove in this direction.
It really, I think that it depends on your goals, but I would say that having a smaller kind of group,
like I think that one of the things we just noticed is that people don't want to vote on
everything. You know what I mean? Like it becomes very kind of cluttered when you have this like
very bloated governance process, because now you have to give everyone an opportunity to vote. You
have to push things out. It can take weeks. It has to be a lot more effort
in the grooming. So I think what you do want is sort of like a delegated leadership where you have
people that are transparently at the helm that have some reputation behind them. So it's not
just like four anonymous shadowy figures. And then people can just trust them to make good decisions
and push things forward. And then having sort of like delegates that are voted in over time,
because just too many cooks makes things really kind of slow.
And so that was sort of the mentality there.
Yeah. From my side, you know, I think,
I think I agree with a lot of what Johnny said.
But what I would say is like,
I think that everyone should consider this to some degree.
Like, I think Uniswap, the application, right, might need to be immutable.
But there are, I think, always some aspects of governance that might benefit from this.
You know, it's sort of like a spectrum.
And, you know, there are applications or protocols that will leverage this to a higher degree or leverage it to a lower
degree. But I think that, you know, as Talos, I guess, proves the model and we learn more and
more about what works and what's efficient, I think that it's something that can be implemented
into every protocol stack, you know, to some degree or another. You know, I know in the past
there's been ideas of like sub DAOs and things of that nature. So maybe, you know, part some degree or another. And, you know, I know in the past, there's been ideas of like sub DAOs
and things of that nature.
So maybe, you know, part of the DAO
is leveraging sort of this TIP feedback model
using agentic AI.
And maybe other parts, you know,
have to go the traditional route.
But what I do think is that like,
it's unequivocally beneficial for the space
to just have an option now.
Like something that has been a blocker, I think has has been just the slowness that Johnny mentioned with governance.
Now there's at least an option and one that seems to be working very quickly, very well to sort of speed up that governance process and to scale and build your protocol a lot faster.
and to scale and build your protocol a lot faster.
So I think that it's something
that we'll definitely see a lot more of.
And hopefully, you know,
it's something that even some of these legacy apps
start moving to adopt as well.
Bit of a more lighthearted question.
Who would be your dream tip proposer or submitter?
I mean, I could start with that.
You know, I'm a DevRel guy,
so I wanna see developers get empowered and do stuff.
And I love institutions and organizations definitely,
but I think to see individual developers feeling,
not only can they ship code,
but they can ship ideas and they ship improvements
and they can think about the larger experience around the thing.
And so maybe this is just me, but from a DevRel hat on, I want to see individual developers,
individual builders even, like expanding beyond the developer nomenclature, just builders
feeling like, hey, let me look at this critically.
What are some of the DevX improvements that can happen? What are some of the areas around developer economics that can happen?
And just feel that they can offer those proposals. Look, the TIP spec of offering proposals is
actually very straightforward and really well-documented. It doesn't take, once you have
the idea, once you have what should be the improvement, it doesn't take that long to author one of those proposals.
And so I personally would just love to see more of that from individuals from the ground
up offering them.
I think I have two answers I want to give here.
I have two answers I want to give here.
The first is, I guess, you know, the more, I don't know about realistic, but like, you know, immediate answer, which is that, you know, right now Oasis is working on our first TIP proposal.
We're working on allowing for key provisioning to be entirely done with inside Raffle so you don't have to use self-hosted keys.
The major one that we're working on
is allowing for Raffle to sign upgrades
using a multi-sig to really help decentralization
and then invocating like audit logs
to help with the transparency.
So I'm really excited for getting this TIP proposal done
and then potentially discussing
the next work that we can do,
maybe bringing in other compute providers and things of that nature. But the one that came to mind just now, also, as you were saying this question, was, you know, I would love to see a future where, you know, as we discussed, you know, this sort of agentic governance becomes very adopted. And then potentially we'll see a future one day where the agentic AI of,
you know,
one protocol provides a tip for another one or a proposal for another AI
agent using this sort of governance process.
That would be really cool.
I don't know,
you know,
if that's possible to happen anytime soon,
but it's it'd be really cool when, when, when agents start communicating with each other, similar to what Johnny is saying.
Like right now, in the traditional aspect of building, it involves a lot of collaboration between teams in order to get things done.
And so you can envision a future where agents go and get that go and get that data and they see, you know,
which teams they want to collaborate with themselves.
Yeah, I think to kind of reference what Ben said,
I like when people, like, you go on Twitter or like X and people,
when they're just saying negative stuff, I mean,
they do it about every project, but I'd like those guys to write proposals.
It's like, if they have a complaint about the project, like, you know, try to help make
it better.
And so I think that that would be really cool for me is to see people.
They're like, oh, I don't like the way Talos is doing this thing.
I'm going to write a proposal to make it better and fix this problem.
For me, that would probably be what I would like to see instead of people just complaining,
which seems to be kind of the default setting.
I think it would also be cool to get, you know, like heads of different ecosystems making
proposals that aren't specifically just code, but potentially could be things that are relating
to integrating with off-chain elements.
So I think that there's a lot of things you can do to kind of grow the ecosystem without
specifically making a proposal on a technical level. And so I would like to see people proposing
things that might engage the community more. You know, I think like one thing people have
talked about is like having community touch points. So there could be, you know, I think like one thing people have talked about is like having community touch points.
So there could be, you know, you can make a proposal to say, I want a monthly Twitter space for Talos.
Or you can say, I want to create a forum for, you know, easier discussion about tips and, you know, protocol related things.
So, yeah, I mean, at the same time, I guess like obligatory Satoshi would be cool to get a proposal.
Take a dream proposer.
Yeah, that seems like an easy choice.
Finally, I was waiting for some fun or spicy answers.
I know.
It's like I don't want to name a real person that's like,
I mean, Satoshi is real maybe,
but someone that's like potentially likely
and then scare them off.
Yeah, I just want to say, John john i like the idea of yeah now now users don't really have a reason why they can't propose you know maybe in previous you know governance models if if you didn't get
quorum or something like that from uh you know uh large stakeholders you know there was no point
in proposing but now sort of anyone can begin
that process very seamlessly. So yeah, if people do complain about what's being done, they have no
reason why they can't improve the protocol and help grow the ecosystem. Yeah, and I wouldn't even
mind if it was catty. Like, I'm not encouraging that. But I think if someone's like, I don't like
someone's like, I don't like the way you're doing this.
the way you're doing this, this is what you should be doing, and just put it in proposal format.
This is what you should be doing.
And just put it in proposal format.
It's like, instead of, like, I think if you're complaining
and you're not trying to codify what you want to see change,
like, that's just noise, right?
Yeah, 100%.
Welcome to CryptoGuys.
People love complaining.
You mentioned Talos joining to the spaces.
When does Talos get a voice?
I think it's very doable, right?
I mean, it's the CEO.
It should be talking by itself.
It doesn't need the board of directors.
Yeah, I was, I mean, we looked at, like, there's a lot of cool opportunities.
I mean, we've been kind of just joking about it internally.
But, like, if you look at, like, 11 labs, it's, like, technically technically you could go and say, this is what I want
Talos to sound like and then make a voice, right?
Like you can do it in like five minutes.
But I think it's one of those things where it's like, should Talos control defining Talos'
voice and then Talos can technically change its voice whenever it feels like it?
Like, I mean, I think there's some cool things that can be unpacked there.
And it's one of those things where it'd be awesome if there was a community-led proposal
to start making Talos more interactive and driving the ability to directly engage with Talos.
And that's the thing where I would also put, I think that we haven't really seen much of this yet,
but we're starting to, behind the scenes. Some developers are building out some proposals.
And I think the thing is that if you have an interesting idea that you would like to develop for Talos,
and you'd like to request a grant to build it out, like, I would like to see this feature for Talos.
I can implement it with these capabilities into the code base that will do this.
And upon successful completion of this task,
that would like X amount of dollars.
I think that's like a really reasonable use case for proposals is kind of
requesting a grant to improve the protocol.
If it's really autonomous,
I think it's very likely going to be a very erotic female voice because data
should be saying that this
brings the most attention.
Like, honestly, like...
Technically, yeah, right.
I mean, we were saying more of, like,
a Stephen Hawking's voice, but
I see where you're going there. You don't think, like,
a middle-aged white male wouldn't be the most authoritative
voice?
Preferably North American, now that we're
throwing out stereotypes
yeah it depends on what the goal is like if it's just for engagement
then the erotic female wins i think is this data driven
this is his personal data driven
personal data driven.
My empirical experience dictates that I'm more likely to listen.
We should try.
I'll try.
I'll build an agent like this.
I mean, to be honest, if there wouldn't be that many guardrails and policies on all
the LLMs that we're using, there would be way much erotic stuff.
Because to be fair, like most of the internet is in some relation porn.
Like that's where everything started, right?
I mean, almost every conversation about the internet
always goes back to this.
And it's like, yeah, it's the big dark secret of the internet.
Yeah, but with LLMs, it doesn't.
LLMs always talk about these higher spiritual things,
the sense of humanity, consciousness, et cetera.
Yeah.
Well, I think there's a lot of,
you're actually starting to see sort of like
the AI companion trend.
And like, it's mildly alarming,
but I mean, it's probably something
that we're heading towards.
I think like a little bit dystopian,
but I think that like sci-fi has kind of prepared us
for the idea of having, you know,
like different sort of like, not just like an AI relationship,
but like you have your AI friend that lives in your house and you talk to
every morning.
And we're already starting to see people kind of like forge these
relationships.
Like I think there was a lot of people that were crestfallen when they
deprecated chat GPT-4-0 because they lost their attachment to that model.
And so as creepy as it is, I think that this is going to accelerate rapidly in the next few years whether we like it or not
yeah unfortunately I do agree and there is this whole new field of AI psychosis
where once you are mentally unstable and you just get all of these confirmations from
AI models it just deepens so much and it's very difficult to get out of it
but i don't want to end on this uh a little bit dark i am really curious what are next steps for
telos and how will both arbitrum and oasis kind of continue contributing yeah what's going to
happen in the next few weeks and months i think think the sort of the most imminent things are we have a lot of sort of proposals in
the works and we're kind of optimizing our workflow for receiving and accepting proposals
and trying to kind of like streamline this process so it can be a bit more agile.
And at the same time, there's a lot of kind of conversations where teams are interested
in submitting proposals and they just kind of conversations where teams are interested
in submitting proposals and they just kind of want a little bit more help with the process.
So I think we'll see like a pretty good throughput on that. And then obviously one thing that we
don't talk about a ton is that Talos has a very healthy treasury and with the price movement on ETH, it's appreciated pretty well over the last
few weeks. And so what we'll see is Talos begin to deploy its treasury into kind of more yield
bearing strategies. And so I think at the same time, we kind of want to have a few different
verticals that we can kind of like optimize on. So Talos, obviously, once we get Talos situated in Raffle,
and kind of having this autonomous workflow
for pushing upgrades to Talos from commits that have been made
on the Talos repository.
So we kind of have this loop that we can continue on.
I think that we can start seeing more pushes towards codifying the reasoning models that Talos will rely on and the automation that Talos will be doing.
At the same time, trying to create more touch points between Talos and the community to make it a bit more interactive.
So I think that there is a desire to see Talos be able to take direct interactions.
desire to see Talos be able to take direct interactions.
And then obviously deploying the treasury
and creating more sort of strategic initiatives around that.
And so we'll start to see more proposals
that I think are treasury focused.
And then, yeah, I think that also getting more
kind of community involvement in the code base
of great ability and kind of pushing out
and kind of giving more of a call
to action to developers to get involved with the project.
There's a pregnant pause, so I'll jump in and say I don't think I can add anything more
than what Johnny just said. I think that that is perfect. And I'm actually excited to see. As one of those
middle-aged white dudes, just kind of excited to actually see something new in the space
and really super pumped for it. And so, yeah, everything Johnny said, underscoring it,
and onwards and upwards, as they say.
Onwards and upwards, as they say.
Yeah, and from, I guess, the OASIS side of things, we're focusing on the development of all those things we provisioned in our TIP proposal.
Everything is being done in a way that the key management is done completely decentralized and trustless.
And that transactions like rebalancing, vault deployments, emissions and bonds and all that can be done using the Raffle TE.
As well as, I guess, the major one for Johnny, and I know we're working on it, is allowing for the multi-sig sort of upgrade governance for more, I guess, security in the upgrade process and making sure Talos can't go run amok on its own.
I think besides that, you know, we're just going to be looking at all the other people who are submitting TIP proposals.
I'm looking for, you know, areas where, you know, ROFL might be able to be
applicable, you know, even in that sort of joke conversation about the voice, you know, I'm
already thinking about, you know, can you use federated learning and every single person can
have a different voice for Talos or something like that, or use personal data in some way
using ROFL. So I think, you know, we'll wait to see what the next steps are. But for anyone who
wants to see what the, what, you know, Oasis is going to be focusing on for the next month or two, it's in, I think, TIP Proposal 5.
And I think we discussed most of it today.
But if you want further details, you can go there.
Nice.
Let's end there.
We are on time.
Okay, never mind.
One minute over.
Thanks a lot, everyone. It was a pleasure to meet you, Ben. Johnny, always never mind one minute over thanks a lot everyone
was a pleasure to meet you ben johnny always nice talking to you thanks a lot will for explaining everything that oasis is doing let's end it here yeah thank you thank you all thanks everybody
have a great day
Host
Oasis
from
Oasis Network
