Projects
![]()
No projects found
Investors
New
DAO Treasuries
DATs
Lend / Borrow APRs
Chains
Categories
Spaces
DEV🕶️TIME: AI-Powered Development
Recorded: Jan. 29, 2026 Duration: 0:48:39Space Recording
Short Summary
In a recent Twitter Space, developers discussed the transformative impact of AI tools on coding practices, highlighting trends in efficiency and innovation. As the crypto community embraces these advancements, optimism for growth in 2026 is palpable, with AI playing a pivotal role in shaping the future of development.
Full Transcription
Thank you. Thank you. Hello, can you hear me guys?
I'm just going to test the mic and welcome all of you here.
Dima, can you hear me?
Yeah, yeah. I hear you perfectly. usual hi hi nice to hear your voice i see vadium is joining as well so let's invite him
to the stage as well um he requested the mic and i'm'm doing great, Dima. How are you?
Yeah, I'm doing good.
I'm a bit tired due to the workload recently, but it's okay.
Yeah.
And you?
Yeah, I'm okay.
Hello, Vadimka.
How are you?
Can you hear us?
Yeah, I'm good all right guys so uh hello everyone and welcome
actually to the first twitter space we have this year if i'm not mistaken uh regarding the dev time. Today's topic will be AI-powered development.
And yeah, the session is recorded,
so anyone who couldn't join or will be coming later,
they can listen to the Twitter space afterwards.
Also, all of you who are listening to us,
it would be amazing if you can drop a comment
under the Twitter space so we can share the
twitter space more more people can see that we are live and more people due to that would join
um so yeah guys how how was your uh like a way or like your first steps in new years so how you are enjoying 2026
it's great year already
uh okay there's so many good things done in this one month this first month
so yeah it feels like it will be a good year for me personally.
And I think for everyone.
Yeah, I hope so.
But to be honest, it's going good so far for me as well.
I just came from my vacation, which is like the first
vacation in one and a half year.
So it's, it's feeling great.
But so that's the reason why I will be tired
for a couple of weeks from now on.
Good.
So you're well rested right now.
All right. So since everything is going
well. What about you, Claire?
It seems like this year
will be nice as well.
I don't know. I feel like
this year started a little bit better than the last year, so I'm positive as well. I don't know, like I feel like this year started a little bit better
than the last year, so I'm positive as well. This year will be much better, but I'm always positive
with new year, so we will see. All right, guys, so I think we can start, we can jump to the first
question, and since I mentioned this space is about AI, so let's start with some of your personal experience
and how you are guys actually using AI tools, for example, like JGPT, Cloud, or etc.
in your daily development work on Vara.
And if you could maybe compare what's changed in the last year if you can see some changes if
not you don't have to comment I can start answering this because I mainly use several
tools for AI like AI agents and I personally usemini, like it's one of the best models and
tools on the market. For example, it's very good in coding and I use just Gmini DLI utility and you can make, install like extensions for it and it can be run like an agent in the
background process and you just need to follow up sometimes.
But I heard that right now it's very trendy to use cloud code or something like this.
And it's like all-in-one utility like Cursor or Winsurf that is used to just full scale
development like IDE.
And you can use agents there and you can spawn some new tasks and it will do it in the ground.
But I haven't used it. The Google company released recently their own type of IDE,
the fork of Visual Studio Code, it's called Antigravity.
And it's a new thing for Google.
It's like full-scale IDE, like the Courser, Windsor, but it has built-in
Antigravity API that it's like you can use a lot of different models inside Google ecosystem because I'm using it.
inside Google ecosystem because I'm using it.
It's very flexible.
It's very flexible and also a new trendy thing is called skills, agent skills.
Like you can teach your agent some kind of skills or roles or predefined configurations.
Like you are the cyber security engineer and you need to look for something,
or you are the project planner, you need to plan something and call other agents to do small tasks.
And comparing to, for example, like a year ago, it was not so good.
Like a year ago, it was not so good.
Like right now, AI agent coding is very advanced since last year.
Like you can build your own application just in one prompt.
Like you can use one initial prompt and AI will leverage that to like the final project.
And maybe it will work or not.
You just need to check this, but comparing to the last year, you can finalize
finalize your initial idea into something really useful.
So that's how we went from the last year status.
Right now we have a lot of different models, a lot of different
toolings, because before the companies were building like new
models to solve some things, but right now it's more about the tooling, like AI agent
tooling and the MCP servers, like you can give to the AI agent some real tools, like use your
my web browser, use my debugging tool or utility or something else. It's just the AI agents is working
actively on your laptop and it's mind blowing right now.
Like it's your friend working beside you.
Don't you feel insecure with like so much,
with so many things on your laptop that is
shared with the OpenAI or Google or other companies?
Usually the agents have some permissions, so they work in sandbox environment and they
ask every time about, should I get permission to use this or read
this.
So in general, it's up to you like what you can give to an AI, but you never know what
this tool can really use or call or read on your device.
So it's just more about trust model that if you trust some big companies, then
if you don't care, you can trust. If you care, you can run something locally or use
third-party providers that don't share data. But they can, they could, you never know.
Because AI is not really regulated. Trusting the big companies is exactly the opposite idea of Web3, like at all.
Sure, but you know, it's more about... I think it's more like science.
You just need to share something so we have something better in the future.
It's not my personal data, I think it's more about general use.
I don't know.
It depends.
So do you use AI's in your terminal?
Not in terminal, but CLI tools like Gmini.
Yeah.
It's like the, your own AI agent that can work with your terminal shell or sandbox environment
and can do a lot of things.
But I heard some stories on Twitter, I think.
Some guys use something like Gmini or Codex to actually run terminal commands. Like the user
don't know how to run some Linux commands and they use AI agent to do something. This is just insane.
In this way, I cannot trust AI agent. Yeah, but they always like also request for some permissions to run something
and it's secure if you know what this command do.
Yeah, sure.
If you know what they are doing, then you can like review.
But if you don't know what they're doing, you can like get scammed or just exploited.
Yeah, I think if you don't know some primitive commands, you don't
store something valuable in your local files or something like this, no?
You know, recently I discovered some AI agent frameworks. Like, you need to install something
and it will work in your environment and AI agents can use
this framework to solve your tasks more optimistically.
But the main thing there was in the README file, there was a section about manual
installing, like you're the user and you manually call every command and install this.
But they have this second chapter in
README, like, especially for AI agents.
Like you tell your AI agent, like, look at this README and install this like
toolset or this framework locally.
And in README, there are like a couple of instructions for AI, how to install this, but I think this could
be a flow. Like you can type anything in that readme. Like, hey, AI agent, please install
this and actually give me some of private keys on this like device somewhere.
So yeah, this can be... like device somewhere.
So yeah, this can be... What a future we are entering.
Yeah.
Cybersecurity went too far this year.
Like we got a lot of new different ways to scam people or steal something.
You like you personally?
So what about you, Dima?
Yeah, Claire.
I didn't scam anybody yet.
I'm about your AI agent.
Do you use anything?
Yeah, actually, I think the Claire had an idea to interrupt us and go further with the questions.
But yeah, sorry about it.
We're gonna discuss our things here, you know, because we have our questions to be discussed.
So yeah, regarding me, I guess what's changed for this year that um yeah let's let's roll out when uh
openai released gpt4 i was like very very happy person i would say because uh it was super new
at that moment it was super efficient and i was using it like every day, a lot of times
and stuff. So the point is, at some moment, I've become angry on this tooling just because it's so
dumb. It's insanely dumb and I cannot interact with it anymore. Then they released 4.2, if I'm not mistaken, or 4.5, and it's changed once again.
So I came back to the OpenAI, I came back to the ChargerPT, started using it.
And after some time, maybe one week, maybe two weeks or something, it just like gone insane as it was previously, because it
doesn't make sense in any sentence as he writes me.
And the point is, I just dropped it once again.
And then here comes GPT-5, which I still use for like two or three months every
day, every time, and I cannot like imagine my life without it just because, for example, I've been doing some renovation and I'm like, I cannot call my mom to ask how many, you know, things I need to put inside my window.
So I could throw the thing on it, blah, blah, blah.
Because it's kind of some dumb question, which is based on some life experience.
And it couldn't be solved remotely by some person.
But GPT can handle it.
So very good.
So the point is, I started using AIs once again, like very, very active. And now for the last one and a half months, I guess,
I tried a lot of AIs for coding just because it's hyper right now.
It's now possible.
It's now able to do something.
And I personally use only VS Code extension for GitHub Copilot.
extension for GitHub Copilot, and there is a lot of models.
And there is a lot of models.
And I would say that GPT 5.2 commonly is the best one for
like planning something for describing tasks.
And the GPT codecs as well is like very, very, very wonderful model to coding.
But I do not like Gemini personally, just because it's too talkative, you know, like a cloud,
because in this chat inside the VS Code, it's just, I'm doing this.
Oh, no, I'm not anymore doing this because I just realized that I'm wrong.
I'll try this.
Ah, no, it doesn't make sense.
Sometimes it's just going insane.
No, no, no, it's just every time like this because it likes to explain me every step
of his job, wasting my resource, wasting my internet, you know.
So, yeah, I think that's pretty much it.
Wow, guys, you went really deep.
So I don't want to jump to the next question
because your discussion gave me actually more questions.
So do you feel like people are exchanging privacy
or maybe they are not as careful
while using all these AI tools
in exchange for fast and, you know, pretty, like,
easy solutions for something.
Yeah, like, a lot of, like, you can upload a lot of work, right?
Like, people are getting super lazy,
and it's just easier for them to put it all in AI.
But do you think maybe we are, like, forgetting
that we are, like, exchanging our privacy
and we are basically giving that we are like exchanging our privacy and we are basically
giving all these free data which we were always talking about in web free that we need to be
secured about like what yeah that's exactly it's exactly what's happening so that's the reason why
i why i asked vadim right now about the privacy and about his files, just because it's, you know, it's insane being able to build like, like, internet computer, blah, blah, blah, decentralized computing systems, verifiable computation.
And at this very moment, we're just sharing our everyday lives photos with our GPTs, asking, hey, sir, how do you think? What about the color of our something like very, very intimate and private.
And, uh, yeah, personally, personally, I do not like share anything, um, about me
exactly, but it's still, uh, sometimes happens, but it's not like something
that, uh, nobody can find like on the net
i think in my case like if there would be one thing i wouldn't want anyone else see
is my history with ai because sometimes my type of questions it's like crazy you know like when
you are three years old and you don't know anything and you just go and write and ask AI like, hey, what do you think does this means? Do you think my dog is dying because this is happening? You know, it's just this is like one thing I feel like this shouldn't be released like our AI like history.
our AI, like, history.
Yeah, the dumb questions, like, helps a lot to our society,
to our humanity, to, like, improve the basic knowledge of things,
actually, because my whole life I've been thinking about the day
when I will have, like, a kid, and he would ask me,
Dad, do you know why the sky is blue?
And I'm like, yes, son son i can explain you why the sky blue
and now it's totally it's totally useless skill to be able to explain some primitives some physics
just because everybody could like not google ai it for for just one second get the most uh precise precise explanation of the things.
By the way, guys,
oh, sorry, have you ever watched the recently released
series called Bluribus on Netflix?
No idea what is it.
I haven't heard about it.
Really? It's a new series. It's about the new type of virus for humanity.
And this virus connects every person to some kind of general mind.
Like every person, they share their thoughts and minds with everyone.
their thoughts and minds with everyone.
Like it's completely
the whole
like the superhuman experience.
Like the planet in the
Eureka Morty, right?
Something like this?
Maybe, yeah.
So these people got connected to each other
and they share thoughts.
So it's not only one person,
but every person is just
one superhuman. So, and some people, uh, uh, they're not infected and they remain like
real people that they can use their own thoughts and their own mind. And, uh, they started to live with all the other people who are infected.
Like, if you talk to someone, you're not infected, and you're talking to someone who is infected,
they know everything about you because your close friend was infected,
and all his, like, mind, all his memories, and all his knowledge about you is shared between everyone.
So if you speak to anyone, he will know everything about you, everything what you like, what you not like, what's your age and like every private thing about you. And I think this is like the same idea that goes for current AI usage.
Like if I am like a dad, cannot explain something to my child and he will use
he will use ChatGPT to get some basic knowledge.
ChatGPT to get some basic knowledge.
He will get the same idea, the same general mind, like everyone else who is
using ChatGPT for his personal needs, like for his personal knowledge.
So in like 20 years, the future generation will grow up in the same cultural space, I think.
Multicultural space of ChatGPT and LLM models.
The shared memory will be dictated by big tech companies.
But it's not necessarily wrong.
Like, you know, a lot of others don't know actually why the sky blue.
So the point is...
But you know?
So I guess the point that we need some verification mechanism for the AIs, but it seems like impossible for the moment, just because we have too many, like, content on the net, which is maybe 51% already that was
generated by the AI and it's coming, coming further.
So yeah.
Speaking of the coding agents, have you ever like used ChatGPT to like, hey man, could
you explain my copilot this task just because I don't want to write words precisely?
I use it all the time, so it's like a digital way, you know?
But I think what's scary are people in the future or even kids which are growing right now they might never go like read a
book if they want to find out something right they will use AI like all of us are starting to use AI
like I'm even using perplexity instead of like a google search you know because it's it's just
gonna find everything in a better way but like then in the future I think what Vadyan wanted to say or how I understand this
was that all these AI tools will have some general like general like information which we will be
learning but who is gonna check or double check if it's actually true like so so all the kids will
be growing with the information provided by AI because they will be like, you know, they will know to do this,
just the only thing, but they will not go and double-check the information.
So maybe slowly and slowly the AI, you know, apps,
or I don't know, the people who created them,
or even bigger, bigger people in bigger positions
could like slowly be changing the narratives,
the like words, explaining, I don't know.
It's just like, it's kind of scary if you think about this.
Yeah, but the point is we as well haven't like, you know,
took books to learn something except in school just because it was a requirement.
books to learn something except in school, just because it was a requirement.
We always like our whole lives, almost Googling something or like searching
net forms and stuff.
That's the first point.
And the second point that actually we don't know who have checked the books
that we had in school.
So for example, do we know exact person who
was like, who has written your book of history, for example? Does it make sense? Does it like
somehow true? Can we verify? We cannot verify anything actually, like speaking of that kind
of knowledge. So we're going to have a fragmentation spread through the world.
But it's not the problem of the AIs, it's not the problem of currently running something,
it's the problem of our humanity, it's our basics, you know?
Yeah, I think you're right too. And Dima, I have a question for you.
Do you believe in Earth that is flat?
Like, do you think our Earth is flat?
How we can check this information? Yeah. So the point is, for example, I do not suspect my school book of mathematics to be wrong
just because I have verified it for my entire life a lot of times.
But there is a couple of things that I cannot verify myself.
So for example, some historical facts I cannot verify. I cannot without some, you know, instruments verify some biological stuff or some chemical stuff.
I'm more about like, you know, the knowledge that is spread through the speaking, through the books
only or something like this. And regarding the flat earth, the point is if the Earth was flat, I think I wouldn't spend so many times in flights, you know, just because it's so boring to fly from Asia to Europe and backwards.
So it's like, you know, a disaster.
Oh my god, we are getting into really deep topics, but maybe we should go back from the
if Earth is fun.
Yeah, let's go back to AI coding.
So we kind of spent a lot of time on different topics, but I actually really like it, so
it's okay so but let me ask
maybe like to go back to the topic so I have researched that maybe like the AI
tools can boost your engineer engineering to throughput maybe up to
67% and it's gonna actually help you but like what are some trade-offs like where the AI actually
slow you down so I know Dima you for example mentioned that some of the tools are like very
talkative right like I'm doing this now I'm gonna do this oh my mistake etc but is there something something else specifically in like this development technical side?
I would say that...
For me?
Yep.
Sir.
I'll continue.
No, you continue.
No, no, you.
Okay, okay, okay.
Okay, okay.
For me, for example, I spent so much time to just review what the AI agent did.
Like he has some kind of summary output, like what I did and you need to check
everything, everything what he say.
And he can say in his own reports, like, I don't know what is, what he did, like, what, what is this summary?
And I need to check every file, every, every statement, every result.
And sometimes it, it goes, the AI goes crazy.
Like if, if you use your AI too much, there is some context window, like, uh, 200, thousands of tokens.
And if you get calls to the, to the limit, the AI will get dumb, like really dumb or
forget something.
And if it did some kind of complex task and you're trying to get something useful from the work he done,
it can last for 30 minutes, like just to understand the results of AI agent. And I got it like, uh, you need to explain in details what you want to achieve,
what you want to get done, what tasks.
And after that, AI will get more, uh, context about tasks and it won't spend so much context tokens for
unuseful info, data or files.
So that's the main issue I have is spending so much time just to explain
what I want to get done for AI.
Well, actually that's a use case to use a GPT to explain it instead of you,
you explain it in short words.
Yeah, that's why we have a lot of tools right now that a lot of frameworks where
where you have some kind of C-level agent,
and he explains tasks to his employees or his sub-agents.
And it looks like a good system,
but you need to review everyone everywhere.
And if someone in this chain fucks up his tasks, his tasks, you need to start
like from beginning, you need to explain again everything, what you want to achieve.
And like AI agent can, can do his tasks for five minutes, 10 minutes.
can do his tasks for five minutes, 10 minutes.
And if he, uh, get into, into troubles in this chain, you need to just reset
everything and it's better to start from the beginning and you just spend like
30 minutes of your life just to do something.
Yeah.
So I think that's one of the problems we have uh i'm not sure from where
uh these benchmarks were taking like like why is it like 60 plus percent or something but i think
the trade-offs here uh the quality of code uh it's uh maybe simplicity, too much simplicity that like comes as a boilerplate
and it comes to be a legacy in your project.
It's like not that quality of code that you would like to bring for your life as a project
for your life as a project and stuff. So yeah, for example, I have one thing that's totally
and stuff.
like a game changer right now about the AI coding and especially exactly about the wipe coding and
stuff. So if you are like a developer and you have some expertise in things and somehow you know understand how is it like
all doing but maybe you do not know exactly how to do something you can like adopt some coding
models for it and it would like lead you to success so for example right now in a sales team, we are preparing the new proposal for ideal version two.
And the thing is, we do not have right now some free guys, guys with free time, who like know how to write something in TypeScript.
TypeScript and our Rust developers just sitting and wipe coding on a TypeScript using like
GPT or other models to achieve our goals just because you do not need some exact skills
because it helps you to cover the majority of the things you need to do.
Like what you actually need, you need to have some quality control, you need to do, like what you actually need. You need to have some quality control,
you need to have some understanding how should it be done. But all the mechanical stuff like right
now could be solved by the AI. So it's like also the point why AI wouldn't be able to like, you
know, fire a lot, like fire the whole planet of juniors, of junior developers, just because somebody need to orchestrate, orchestrate this like power.
Yeah.
So it's just like, it brings me to another question, like just additionally, what you just said, guys.
Like just additionally, what do you just said guys?
So would you say that your role has changed and would you call yourself like more as a code reviewer or a code writer now with like work of AI?
I would say that I feel like I'm more, I can spend more time researching the idea and like more engineering my task and give it to AI to prototype it or to code it.
And yeah, I feel like I have more resources to just think and research.
research. Again, I use AI for research for doing my initial prototyping, but I can spend
more time to go deep into question and into the task and then ask AI to code this idea.
If I have some kind of good documentation, good research, then I can give it to AI
agent and it will get things done.
And again, you need to spend a lot of time reviewing it, but the time from
initial idea to the final, like working code is much faster than I would
done it alone, for example, solo.
Dima?
Uh, yep.
I would say that at some point of your developer career, there is a choice to become the sole
code writer or becoming the code reviewer.
So I guess it happens with me, with my metamorphosis for code reviewer far before the AI epoch.
But anyway, when we speak about some white coding, of course, it's like you're just a spectator, you're just having fun. So it's like, like, very, very awesome to build a front end when you have never tried like TypeScript, JavaScript, HTML, blah, blah, blah. And you but still you can like run some application in maybe one hour or something like this and it will look like exactly how you want it.
Or for example, the opposite way, if you are a front end developer, you could like run your smart contract without any like pain, without any understanding of things.
and you're just doing what AI says to do.
And you're just doing what AI says to do.
But I don't know.
It depends on what exactly you're trying to code.
You could be both, you could be a code writer, you could be a code reviewer.
But actually, I think in the production or almost production-ready solutions,
there is always need in reviewing the code even if you
are trying to write something i'm not about the ai so basically you have some functionality and you
need to implement something new you implement it and then you start like reviewing yourself you
start reviewing everything around it and so making sure that it's all
works. So I think there's like not a lot of things that changed but it will
change in the future much bigger than right now I guess.
Claire?
Sorry I was trying to respond to some comments.
Sorry, I just fell asleep.
No, no, no.
So I just actually wanted to ask, we have around 90 listeners today.
So is there anyone in the audience who would want to ask you guys some questions related to AI?
ask you guys some question related to AI.
Maybe someone has experience with some tool
or maybe has some feedback about something we discussed.
If yes, don't be afraid to raise your hand
and we will add you to the stage.
But if not, I have probably last question for you guys.
For the listeners who are builders,
but maybe are not as experienced developers yet,
what can AI...
Oh my God.
I'm not going to be able to.
How can AI help them build on Vara today?
And where would they still need a human expertise, like in reality?
Yeah, I think the basic knowledge of things,
that's like there is like a VS code, there is a Rust,
and it compiles, blah, blah, blah, here is a variable.
It compiles, blah, blah, blah.
Here is a variable.
It will be enough for now.
It will be enough for now.
And to like, to power up your coding system, you better to like share a link to our wiki or some examples.
And I think, if I'm not mistaken, at this point, Codex like really know about the VARA and about sales and coding on it.
Vadim, isn't it?
Have you tried it recently, if I'm not mistaken?
Yeah, it is very good, but my approach is more like, for example, I was prototyping
my project for VARA.ath and I just git cloned the official example for varath there is a typescript scripts foundry scripts solidity
programs varas sales programs and i say said to uh codex like explore this currently working
currently working on mainnet, for example, project, look for its code, look for its patterns,
and build based on this my own idea. Like you give it some context about the
recent, recently used real working example, and it will update his own knowledge to something new and
adapt so that that's how i solve my ideas right now yeah also if i'm not mistaken maybe
louis could help us because i see louis in our listeners we had like some frontend, some application about the AI for builders that the page that allows
you to generate some code for VARA, isn't it?
Do you, Claire, maybe know?
Maybe we could share a link about it.
But I may keep going right now just because I didn't finish one thing
about, like, the junior developers, not only on Vara, but, like, in general, in
the world, I guess, for you, if you're willing to improve your skills, I think
using the talkative models like Dream and I, like, load is, like, much better for
you just because it helps you to start thinking, not as a machine, but as, like load is like much better for you just because it helps you to start thinking,
not as a machine, but as like real developer, as like engineer, and you could adapt your vision,
you could adapt and improve like your mind flow, how to solve like tasks just because us because due to the fact that like some machine learning could could teach some code
to solve some some like arbitrary tasks and it's like totally possible for for anyone
on the earth who's like a human like you know there could be there could be be a joke about Zuckerberg, but I just... no.
So I invited Luis. Hello, Luis. And I think, Dima, you were talking about the AI cogenerator.
Yeah, exactly.
Hola mi amigo.
Yeah.
Yes?
Hello, Luis. yeah yeah hello louise Thank you. Thank you. Thank you Lois for stepping in. I have also shared the medium link and also the link for
the AI generator here under the Twitter space so anyone who would be interested to check in it,
you can just click on the link and check it out um so i think we are
like finishing the twitter space i think we covered quite a lot we even got a special guest
so thanks for joining us louis um and yeah i think we can wrap it up uh thanks to all the
listeners who join us today and listen to the Twitter space and also thanks
to Dima and Vadim for
joining, answering all the questions.
I think we kind of
slipped away from the topic, but
I think it was even more exciting
and interesting and especially
thanks to Luis for joining
as a special guest. Guys, feel
free to also share some of your
closing points
for the session. Yeah, feel free to also share some of your closing points for the session.
Yeah, you just forgot to thank yourself, Claire. You just forgot it. Thank you for hosting us.
Please thank you for hopping on the stage. Yeah, thank you, Vadim. Thanks to all the listeners.
See you later. Yeah, thank you everyone. Spend a good time discussing this nice topic.
Very interesting and hope we will discuss it in the future.
Thanks guys. Bye.
Host
Vara Network
from
Vara Network
