Enjoying this issue?
Get tomorrow's AI & engineering digest in your inbox — hand-picked, summarized, and always spam-free.
TLDR
Pi agent's success comes from its minimal, bash-centric approach that lets models pipeline commands efficiently rather than pulling everything into context, a pattern that has become standard as models improve at using computers directly. The creator argues that the current agent interface is analogous to DOS—functional for power users but not the final form—and that the real unsolved problems are classic system architecture challenges like state management, durable agent sessions, and giving agents the ability to present custom UI, not new AI breakthroughs.
Key points
Pi agent relies primarily on bash commands rather than many specialized tools.
The creator believes current coding agents are like DOS—functional but not the final user interface.
Durable agent sessions that can be suspended and resumed remain an unsolved problem.
Agents are very good at Linux because it is well-represented in training data.
The creator argues that open source projects are judged by longevity, not initial quality.
Tools mentioned
Techniques
- bash-based agent tool calling
- deferred tool loading via interleaved system messages
- server-side session compaction
Stop scrolling. Start reading smarter.
Receive the day's most important AI & engineering updates in one concise email. No spam.
Transcript (captions)
This is Armen, the creator of Flask and the founder of Arendelle, the company currently in charge of PI agent. In this podcast, we talk about why Pi is beating cloth code and Codex, his own custom
agentic engineering setup, and how to build great software in the era of AI coding. This is the David Andre podcast. Enjoy. All right, Armen. I want to start by talking about PI agent and how it's
winning a lot of the different, you know, benchmarks and comparisons between harnesses despite being one of the most minimal harnesses out there. Why do you think that is?
>> The models themselves are getting very good at using computers and pi just gives you basically bash. for
the most part harness are getting really good at at just doing the basics and in a way like over time that became more and more obvious. Uh so like at one point pi was sort of
uh maybe at the forefront of doing this but I think at this point this has become pretty standard. Um a good example of this is I think that if you are using codeex even though it shows
you at this point that it discovers a bunch of files and it found a bunch of files it doesn't actually have many tools left anymore. Codex as a harness basically also relies on bash for a lot
of stuff. So it it directly if it discover if it says like it's looking for files it basically just calls RG um to find files. It has been a very good approach in general to just rely on
bash. And a good reason for this is that rather than pulling stuff into the context it can it can pipeline stuff together. So it can do find me some files but in the same pro in in the same
command also add some separation markers um so that it knows um this is program one to run then it executes program two to run executes like if if you look into like how it actually executes the bash
commands you can see uh it's getting quite um creative in in being relatively context um efficient I would say. So where where does that going to leave us like in the future maybe say like 6 12
months you think it'll just be like models going more and more low level? >> Look I think like I felt reasonably confident last year to sort of say like okay this is probably where it's going
to go because um what what I felt like probably tells us a little bit on the future is like where is the training data what is easy to train into the models in a reinforcement learning
setup. Um so I I felt some level of confidence that Pi's approach probably is a pretty good approach. Um now that it has become like the center of what everybody is
focusing on like coding agents turned from oh this is one approach for AI to basically this is the one approach of AI. Um and so now it it I think like at least temporarily it has become a little
bit less clear where it's going because the model labs themselves um probably are going to compete on also getting other training data into this is sort of the part I think where it becomes maybe
a little bit less clear. At the same time I think all of these labs are now also competing on just being really good programming agents and they're not diverging that much from each other in
terms of what the basic ingenuity is in in tool calling. Um but we we are definitely seeing some changes in in the capabilities of the model on the inference layer that are maybe
indicative of of of some other approaches. I think like the biggest change that I noticed is that um as an as an example you can now have interled system messages in many of the
state-of-the-art models which allow you to do deferred tool uh loading as an example. Um, it's not a huge change to how you would write an ancient harness, but it is a meaningful enough change
that certain patterns now become possible that previously were not possible, but I don't know if it's a huge difference in in what we would all be doing.
>> The most common mistake I see new founders make is not enough feedback from real users. Sure, with AI, you can build software much faster, but the faster you build, the more important
real customer data becomes. After all, you need to know whether you're building the right thing. That's why I want to tell you about today's sponsor, Post Hog. Post Hog shows you how people are
actually using your product, like where they get stuck, which features they are using, and whether they are coming back. And with their session replay, you can even see what the users actually did. Y
Combinator, Superbase, Fireworks AI, 11 Labs, Railway, all of these companies use Post Hog. In fact, Superbase used Post Hog to spot a wave of new users coming from AI app builders and based on
that data pursued partnerships with those companies. As one marketing expert at Superbase put it, Pose Hog has literally helped us get 10x more weekly new users than we did a year ago. When
you're shipping new features every day, your feedback loop needs to keep up. And that's exactly the problem Post Hook solves. So visit go.postthawk.com/david posth.com/david
and learn how Posthawk can help you understand your users and build better products. That's go.post.com/david. The first link below the video. What do you think is like the reason why Pi took
off? I mean, obviously people like it for its customizability, minimalism, but it feels like the whole world is trying to build a harness, right? Like we see literally every week multiple new
harnesses being released and it's speeding up. So like why do you guys think like you know relatively small team especially outside of Silicon Valley like why are you able to stay on
the cutting edge is it just like thinking about it deeper is it you know taste how would you how would you describe it? >> Uh look I I don't think we have done a
ton of like uh going into ourselves trying to figure out like why is pi popular. I'm not particularly preoccupied with like keeping people to be pi users of what we have today. I'm
I'm more interested in like where should we be going and and what it will look like then. I think like if you ask me like why did Pi become popular around Christmas? I think what the first thing
that Pi demonstrated people hooked on to was the extensibility part. It was a very minimal agent but it was one that you could make your own. And that was also taking place at the time when some
of the other agents were turning the other way where it basically turned into massive amounts of tools. every single update from claude code I remember being a pretty meaningful impact to one's
workflow. Um and and you can see now that that claude is actually slowly moving the other way because like clot has fewer tools now that at peak of cloth tool usage.
>> Yeah. >> And if you now look into open code 2 for instance it's all based on plugins. Um so I think like this this idea of like self extensible software. Um maybe pi
was was rather early at this. Um, but uh, yeah, I don't know like if honestly like at the end of the day, I don't necessarily know why things are popular. I think like Pi struck a pretty good
balance. Um, and and that's as far as the software goes. I think like beyond that, um, one thing that probably set Pi apart and I think maybe to an extension sets Arendelle
apart is like this idea is like, okay, we have a pretty powerful machine here in the form of an LLM. How do we actually make it work for you? um which is a positioning question, right? It's
like he's like, "Okay, we're we're sort of sitting in the middle of like we we think AI is really really great. Simultaneously, we have a lot of skepticism about it and we're trying to
navigate the the balance between this. Um and I think that that is probably a more European approach in some ways. >> Yeah. So, you started you switched I think in December, right? Like fully to
PI and then uh you like four months later you acquired it. So like what was the like deciding factor that like hey I want to have it under Aaron drill was it like Mario was it the project taking off
like how did you make that decision? >> Uh we were talking with Mario even like in in the summer of 2025 if he wants to join the team. So like in many ways I think like the the challenge with hiring
people in an early in a startup is often like how do you how do you meet like how how do you how do you cross the paths? Um, so in many ways I think that the question was less of pi as it was like
what do we have to do to get Mario on board? Um, so that to me at least is like my version of the story. I think Mario wrote quite a bunch about on his blog um how he how he joined Arendelle.
Um, Arendelle right now looks like it's the Pi company, but that's really not our goal. Um it it might continue to be the PI company for a little while longer just because like we we have like a
bucket list of stuff we need to do. Um but but our future is unlikely going to be uh we're going to be the harness company. I think we we just want to be a company that makes uh AI work for
everybody. >> Um and we just have to start at the at the harness. >> So what what do you think that will look like you know going forward? Like how
are you think about it? Is it you're talking about the people on the lab? Is it like you're you know deep testing yourself? How do you how do you like try to see where the world is headed?
>> Um it's almost it's a pretty loaded question. Um I think that >> it is >> um look I think it it the very simple
version of this is like someone puts money on uh like in a into a data center data center appears some other company puts money into models keeps training models and disappears and then um both
of those are really great ingredients to hopefully drive a whole bunch of really interesting interactions that an LLM can do together with a human driver. Um, in many ways we're incredibly early
in this and um, I don't think we have a particularly strong like vision of this is exactly how it should unfold. But I don't think we would be satisfied if the experience that you get from an agent
doesn't also deliver value to a non-programmer. like the fact that this seems to be like we're we're we're moving this way with like cloud co-work and things like that,
but in many ways what you can do with an LLM is like an a programmer can do a lot more than a regular CHBD user can. Um and and we're we're trying to figure out like
how do you how do you bridge that? Like how do you >> is that even possible? I need to push back on that because like people who are more technical, they can use more
agents. they can figure out how to manage dozens, hundreds, thousands of agents, you know, what models to use, what harnesses to use, which ones to run locally, which ones to run on the cloud.
So, like it would be ideal to make everybody, you know, a power user of AI, but like what we're seeing is that the people who know how to use it are getting more and more out of it, you
know, like the the saying, the smart. >> I mean, I think like at the I'm not saying like everybody is going to be a programmer. I don't think that that's either the the way we're looking at
this, nor is that really the goal. I I think that look I I happened to grow up with DOSS. I don't know if you've ever used it. It basically the shitty version.
>> I know what it is. >> It it was basically like nowadays you would basically like funny enough like you look at a coding agent and they're very good at writing bash, right? And
and Unix was really a a great operating system from the start for a power user. DOSs was just a really crappy textbased computer, personal computer, very optimized for resource utilization and
so forth. Um, nobody would would use DOSS if they have a choice to use something else. And it took a bunch of iterations until a desktop computer turned into something
that became really something everybody could use and wanted to use. Um, and I think like we're a little bit there with the agents right now. The agents are basically like not the user interface
that really anyone should be using. Like I I don't think that the I have a hard time imagining that we are all going to use uh coding agents in the current form four years from now. That
to me doesn't seem particularly logical because on the one hand all the people that are token maxing are going to deploy this thing into like software factories where they're not going to
spend a lot of time working with any agent individually and the people that are going to want to use this for a bunch of um like domain specific work. they're probably not going to hang out
in like an enchanting interface of today either, right? And there are a bunch of technical reasons why like what we have today is not going to be a sufficient solution for like really
good software of the future. Um, but we're like someone is going to build it. Um, and that to me seems very obvious and we just want to take a shot at that as well and at the very least build a
really good infrastructure that someone else can do that too. We would like to end up in a world where from Aaron's perspective, we would like to end up in a world where people can use the stuff
we built to build really good agents themselves. Simultaneously, we want to push forward the foundations of um all the bits and pieces you need for really good agents so that we can also
build our own solutions on top of it that we can make available to non-engineers. >> I see. So apart from the harness, what do you see as some of the biggest pieces
that still need to be figured out? >> I mean, I think that a lot of them like one of one of which is obviously that um it's not so much that a piece needs to be figured out, but like the entire
economics in the space have to move towards real competition on the models. The good thing is like that I think that seems to be somewhat happening at the moment. So you have a choice of a whole
bunch of models. But um one thing that is maybe suboptimal right now is that a lot of the new features on some of the um closed weight models in the US are they kind of lock you into the ecosystem
more and more. >> My my classic examples are the the serverside compaction right now creates a non-portable session. Um that's suboptimal.
Um, one of the things that we found needs solving is just durability that you can take an agent, suspend it, and continue uh where it left off. Um, we're we're we're doing rather well in many
ways with coding harnesses today, but it's just not quite at the level where you can build durable systems that don't require a human in the loop. Um and then um like we I think nobody has quite
managed so far to take an agent um from a terminal user interface into the web and make it the most powerful experience that it can be. Um most most agents in the web look a little bit like just a
web user interface to a terminal user agent. Um but there's there's a lot to be explored in that space. Um, and then obviously the whole topic of like we're not really strong believers
in memory, but I think we are believers in like giving agents ways to manipulate data. And so the question is like how do you make uh how how do you give agents really good database access um so that
they can store their own data um and also make that available to a human. So that's there's just a lot in that space and there will be more that we will find. Um but it like I
there's an almost like never ending list of technological challenges that everybody's working on right now just to make these agents actually deliver value to everybody and not just be um yeah a
coding agent I guess. Yeah, I think the interface problem is an interesting one because you know right now basically it's like terminal or some GUI but like I imagine it could
be even something like a more like video game where you have you know the teams of agents you have your resources like maybe your limits you know >> I I don't know I don't even care about
this is that much I think like on a very basic level I think like it is it is ridiculous that I mean like it's great that openclaw and and Hermes and all the systems are are sort of like
demonstrating the value of agents but many of the problems that I have are not to be solved with text only and these agents are not really able to bring up custom UI um in a durable way right it's
like um if if I want to use my agent I don't know to manage my home assistant solution then surely my agent should not just be able to like follow my commands but also visualize me in my household
and all my smart home devices Right. And so like the fact is these agents are yeah they can sort of give you a little bit of UI but not a lot >> that in itself I think is a problem that
needs solving. >> So it's like it's the it's the fact that like an agent is kind of constrained within a chat transcript >> is is one of the things that that is not
optimal. >> So is that an issue on the architectural level then? Because like I feel like some some things like you know the lack of taste, lack of creativity is it's a
it's within the transformer, right? So like is this even solvable or do you think we need like new AI breakthroughs? No, I don't think you need new AI break like look I think like at the end of the
day a lot of the things right now that restrict what an agent can do. It's just like we moved from humans are writing code to agents are writing code and like some of the things that are cheap for
humans are expensive for agents and the other way around. And so like I you can notice for instance now that Linux has a resurgence not just because of Omari but also because like
um agents are really good at Linux because they found a lot of stuff on the internet about how Linux works. So they're they're very very good at remote controlling a Linux environment and you
can basically take an offtheshelf um YUbuntu or Arch Linux or or Nixos installation and just fully remote control it and customize it with just encoding agent. So like that is a that
is a huge change like you I would make the argument that even nowadays my mother would probably be more successful on using a coding agent for personal Linux customization on Linux than on Mac
customization on Mac. just because the Mac is not >> training data >> train train as much and let alone Windows. Windows just is like completely
outside of the distribution. Um and so um there are certain things that previously very few people were very good at and used and and even though that technologically maybe really a good
choice like I'm pretty sure we're going to see more people write Rust. I'm pretty sure we're going to see more people use Nixos. I'm pretty sure we're going to see more people
uh I don't know try like really complex databases or work with Kafka event streaming or stuff like this just because like that was always a better solution but it was just very complex
for humans to work on it but for an agent not so much and so like we're in this phase now where it's not really a question of like do we need an AI breakthrough I think we just need a
breakthrough of like making these system like figuring out which systems work really well with what we have and so like the question of like how does an agent and bring up UI is not a question
of like creativity or anything like this. It's mostly just like state management uh component libraries. Um like how do you build databases that uh an agent can work with with a destroying
data like this? The none of those really are like AI problems. They're like just classic system architecture problems. >> I see. So we almost like took the in the you know last 20 30 years like a detour
optimizing everything for humans. But now that's kind of crashing down because it doesn't matter and the agents are becoming the user. >> Well, I mean I think both of them are
becoming users. Um like this this idea that now there's only agents doing stuff to me seems a little bit question. >> But if you take it to the extreme like if the models keep improving, harnesses
keep improving, input output is easier, isn't it going to be obvious that like agents are going to be 99.9% of the users of any software? I don't think so. I
>> Well, I mean like probably it depends a little bit on how you measure it, but like the so I mean on a very basic level the number of humans in the world if
anything is stagnating, right? So it's like it's like agents have not created more humans. So the addressable market of people that give a is the same. So um if like the software doesn't
matter if only an agent is using like at the end there has to be a human like the GDP is like driven by people buying stuff and not by like agent doing things for fun like that that like an agent
like a a computer system on its own doing something is just like burning energy for for nothing. So so someone has to benefit and that has to be a human. Maybe you can extend it to
animals or something, but like so at the end there is going to be a human. And like this idea that now we're all just sitting in a a sun launcher and we're to
talking via devices to it each and every once in a while. That doesn't seem very likely because like this is always going to be inferior than the person that is doing the same thing and puts a human in
front of it. like this um because like if if if machines are driving everything then um like the person that uses the machine that everybody else has plus a human is
going to have like an advantage. So >> yeah I agree with that. >> So so there will always be software that has a human somewhere there. Um then like obviously you can make the
arguments like there will be a lot of software to support agents. That's probably true, but there has always been a lot of software to support more software to run. Like every
observability company ever, most of the code is not there for a human to look at. It's for like processing all the data points. Um and and now like okay like maybe previously 70% of queries
that were sent to an observability platform were driven by humans and 30% by automation. Maybe now it's going to be 30% automation, 60% AI, 10% human. But that then
observability platform would never give a human access to data to me doesn't make any sense at all because like anyone who then still provides that for a human is is is going to be the
more like there's also this idea that like if agents get really good at just writing assembly or a version of it. I don't think why that will ever be true because like if you if you lose the
ability as a human to understand what's going on then you also don't have a ton of trust into the system and the moment something goes wrong um someone's going to complain
>> like I I made this argument recently like I don't think banks will ever vibe code because the moment something goes wrong like their customers will not be happy
>> the accountability part like you need to blame somebody >> exactly and you're not going to be able to blame a machine that's just that's not how society works and that's that's
not how society will ever But isn't that just like a you can blame the human that used the machine you know like if there's a farmer >> but once once once the sort of the idea
sets in like okay you have to blame the human that uses the machine then the human that uses the machine is going to demand better in introspectability and if the human is left with like okay
I'm just looking at binary right now and it's just like I don't know anything that human will no longer use the machine. So I think that's mutually exclusive. Like if we have super
intelligent agents with like you know 200 300 IQ models surely they can cook up a custom user interface. >> Like there might be a future very far out there but we are so far away from
this they just don't think it's like particularly important to think about this right now. in the in the foreseeable future, humans will want to understand what's
going on. And so they're going to have the same interface, give or take, as an agent has just because they want to look at what the agent didn't understand it without having to ask another agent that
might lie about it. Um there there's the other >> that's why Markdown I guess that's why Markdown is winning. I mean it's it's ironic that like these all these sort of
like really simple things like this markdown JSON files um Unix pipelines are becoming like the underpinning of of of everything that's going on. But um in a way it's not surprising because it's
also what humans understand. >> So walk us through like your typical usage of agents. H how like when you sit down in the morning like how do you work? How does your setup change? What
does it look like right now? I'm not particularly interesting I think as a use case because I I don't run a soft factory. Um it's not that like I don't try to do to
run one but like I haven't succeeded. Arendelle actually might be the only harness company right now that doesn't have a robo pie or something like this um that just automatically works on
issues. Um I think we might want to go there but but so far a lot of our engineering is basically still the same which is um we still mostly use the coding agents on our machines. Very
little actually sits in the cloud. I have um >> interesting >> I have a desktop Mac that's not sorry not this I have a desktop Linux that I
run via SSH some things on if I want to keep them running um but most of it is still um where the code is. Um we have a little bit of automation in in the in in GitHub actions so that you can run pi
there. Um a lot of what I'm doing actually is not so much code generation is actually trying to figure out like how stuff
should work. So a lot of it at the moment in particular with the with the new harnesses like investigating with the machine together like here's a here's a report
how to get repo out of it stuff like this. It's it's very traditional engineering for us still I think. Um yeah, it's the compared to I think what a bunch of
people are doing. We're we're very conservative still. Definitely. I am very conservative still. I see. So you mentioned that you still are like heavily local but we see like a
huge push you know especially by the big companies like OpenAI, Cursor towards cloud agents. Do you think it's like valid argument to say you know soon enough your machine won't be enough you
need to move everything to cloud or is that like more of a you know ecosystem locking tactic where the moment people move all their environments and everything to the cloud it's like very
hard to migrate away from that >> I don't think it matters honestly um because like there will be a sufficient competition that you can do whatever locally or in the cloud um
I think the question is like what do I get from having it in the cloud versus what do I get from having it local? And so, for instance, one of the reasons why I don't really care that much about the
cloud right now is because I have a really kick-ass Mac and it's just way faster doing it locally. Like when I I I do have a pretty decent like I don't know Blink Linux machine and it's it's
it's fine, but it runs the tests like at half the speed of what my Mac is doing. And like there's and there's like this inconvenience still that certain things I want to have on my machine and then
like bringing that data into the cloud is just a little bit more annoying than it should be. So it's like in some ways I just have a little bit too much friction to do more of that and it's
just a little bit too slow. One thing that I noticed for instance like there I I I had this conversation like this is already like a month or two months ago where um I was I was helping someone
debug a problem and they were using uh I think it was Codeex um the cloud codex version or maybe it was cloud code I forgot but it was one of the desktop apps and he was programming with this
and I was like how how do you run your database tests here? And I was like, "Oh, he doesn't actually know how he run the database tests." And then we looked into it and actually none of the
database tests ever worked in his cloud environment. >> So missing some environment variables or what? >> No, because like there was actually no
Postgress running like he he did like the the cloud environment didn't actually manage to set up Postgress ever. it just mocked out everything and and even nowadays that continues to be a
pretty significant issue like um actually getting good like I found pretty good success in having a machine set up with a dev environment that's just SSH into or like they're better
solutions than SSH but like to SSH into and just have my stuff running than trying to get any of those cloud solutions work um for day-to-day project engineering um they will get there for
sure but it's just um I just have too much stuff that needs running and I found it easier to just here's a machine that I control. Everything is on there. Um,
we're obviously going to >> I was just asking. >> Yeah. I I mean like I think we're obviously going to move to this. Historically the biggest reason why this
didn't work was just because like it took people did not spend a lot of time building ephemeral deaf environments. I remember like when we onboarded centry engineers like in the first week it took
them like two days, three days to have the deaf environment going. Um like we had an entire team at one point of three engineers that did nothing for a little while to get the dev setup working
better. But even then you ended up in a situation where you had one deaf environment. So if you want to run seven agents on your machine, then you need seven databases, seven kafka clusters,
seven like it was just none like there was complex enough software that it was just very very hard to actually get all of that set up. And now even I I don't know how the centry engineers are doing
this nowadays, but like I'm pretty sure it still would take you like seven minutes to bootstrap a def environment from nothing even with caches, which is pretty expensive for an agent to do
still like like that means like every single change now has this extra leadup time or maybe have a pool of them. But it's also getting really complex. Um I'm pretty sure people working this I'm
pretty sure that some people have this perfectly figured out for themselves. It just happens to not be yet there for me. um or at least I found something simpler um that that continues to work. But I do
think that moving stuff into the cloud makes a lot of sense for a lot of reasons and in particular like even even just me having access to my Mac I just use tail scale for this um because like
pretty straightforward to do but um that still requires me to set this up. Um, and if you look at for instance what like MMP does with orbs or like the cordex incloud stuff that is much easier
than figuring out how to do tail scale and SSH and all of this. So like from a from a UX perspective my setup obviously sucks. Um, but yeah it will get better.
>> Yeah, I mean right now I'm the same like most of my stuff is local. I'm just, you know, maybe I'm asking because I'm getting this FOMO of seeing, you know, all this talk about cloud agents and
then like thinking from first principles like if we have 100 times more agents, you know, thousand times more agents like there's going to be a point like, you know, I have also like a really
expensive Mac and it's sometimes overheating, you know, if there's like multiple agents running multiple work trees and like they're all decided to run tests at the same time like you can
see the fans start to spin up which is unusual on like a topsp spec MacBook. So I'm wondering like where is the limit you know like is the next generation of models or the next after that is it
going to be that good where like I actually physically cannot run all the agents on my computer. >> I mean I guess the question is like what's what what like why are we doing
all of this I think is a pretty important question. Um I think that like I think the simplest way to look is like did like for how much money
enterprises now spending on AI coding did their yeah but particularly like did their revenue actually go up like I'm not questioning that there are a lot of AI
startups and they're making a ton of money right it's like I'm not questioning this but like if you take a traditional enterprise and they're rolling out coding Asia on mess like did
there like does it get reflected in the numbers or did it just get more expensive because like we now have a line item that we all pay which is basically these models are getting so
good at security research that we also have to pay money now to permanently check our code paces against security issues because someone else is doing it against us otherwise right so we have
primarily made it more expensive right now but on the other hand I think a lot of engineers are finding on their free time a lot of motivation to build side project of a side project of a side
project and that also is in many ways positive for GDP even if it's not necessarily being used that much but like all of these sort of side experiments are are are not useless and
so it's like who is benefiting of this right now actually I think like is people with the side projects are having way more success than companies actually rolling this out and I think this the
fact that this happens like this is sort of my biased view of this like some people might actually say like hey We wrote this on the company and actually here are some numbers that really
demonstrate that this works. But for how much more expensive the workforce has become, I don't yet have these great examples of like like we our token spend went up 500%.
But maybe like the number that we have moved in the company is like maybe there's some number that we can demonstrate like x amount of support tickets closed or something but like
it's very very hard for for for a lot of people to see anything other than number of commits. >> Um and I think that that is like >> like do we even need that much like
concurrency and and all of it like um we are producing a lot more code. That's true. That's very noticeable like GitHub is like struggling with staying up because like so many commits are going
up. >> But I'm still like I'm I'm in the morning. I'm taking a train. I'm taking a bus. I'm cycling somewhere. All the stores
around me behave more or less the same as they did three years ago. Like I haven't yet seen this huge change in society. On the other hand is like the last shift before
that was mobile phones. And like whenever I go to concert now 100% of people standing in front of a stage hold up the mobile phone right so it's like that has permeated society aggressively
um I think chd has also like you you see chd everywhere um but um yeah I don't know I I haven't yet like seen the sort of that that impact it will come for sure but I think so far we're all trying
to figure out >> yeah I mean this >> it is exciting times it is absolutely exciting it's it's fun to explore all of this. But at the same time, it's not
very hard to be at least a little bit skeptical about the impact of it because like so far it it it is seemingly takes a lot of time to show up in results. But it's it it doesn't take nearly as much
time of showing up in in terms of cost, infrastructure buildout, all of it. >> Yeah. I mean, this is a classic like response, you know, when somebody was on Twitter flexing that they are burning
more tokens than I think somebody from open code is like, why aren't you more successful then? you know, it's like there's so many people running their 10 software factories and like, you know,
they're not producing anything. And actually, your point was a great leadup to my next question, which is like why is so many of these, you know, successful projects start as like either
a joke, either a side project, either something that the people don't think could be successful and they just kind of put it out there and then it turns out to be like capturing interest. Why
is this a pattern? I mean your story with flask is a great example but like I want to understand what's the magic there where like the creator thinks like ah other people won't like this you know
whatever and then it turns out to be the thing that grows. >> Yeah I mean I think to some degree it is maybe not so much that is the choke as it is genuine curiosity
like if you build if you force yourself to build something you don't believe in it's it's not going to succeed. Um, if if you have an itch to scratch, even if it's just for fun, um, like most most
systems basically are just because people keep doing stuff like like you you show up in the morning, you do something, you keep doing that.
That's the real indicator. It's not like Flask was maybe a joke initially, but like at that like after that it turned into just work. like you just had to do a lot of commit had to work with the
issues that came up and like eventually make it work and keep doing that for a long enough period of time until like eventually like it's not an overnight success as it is like you have to keep
doing that and even I don't know like like everything goes so much faster but like if you look how many commits Mario put into pi early on like that was like not a tremendous surprise that pi um it
wasn't just that it was unique it was also just like it worked really well um for for proportionate to to everything else out there on the market at the time. Um, so I think it's it's I don't
think that's like a big surprise and it's like you put time into something and and obviously you can't put time into something that nobody cares about. I think that is that is
true. Like a lot of my open source projects are like this is my thing because nobody else cares. Um, and then you can sort of try to do something with marketing to see like is are there
enough people that care about it? But like there have to be enough people that stick around and and give you all the problems that you can see like what what the real issues are and and then you go
from there. >> So I guess it's being like very specific, very passionate about a single thing like specific thing being on the cutting edge of that and that giving you
like the unique insights like this could be better that could be different. I don't know if you have to be at the cutting edge like you can you can yeah maybe but like like I think a classic
example of something recently is like um like Mitchfield built uh ghosty or ghosty y I I never know how to pronounce it or ghosty really ghosty but um >> like ghosty now might be cutting edge in
many ways because it also does so much but like for a really long time ghosty was look I don't want to say anything wrong here but because like it was always fast But
like it had something to catch up to, right? It's like I remember when I used it very early on, there was a lot of stuff missing, but it was it good enough compared to
some other things that was really interesting and it was very noticeable. It puts a lot of energy into it. So it felt like if I keep using that eventually it's going to be great,
right? Um but it was not cutting edge in the sense that there was so much stuff missing that the cutting edge was kitty or the cutting edge was a bunch of other stuff right so um like I don't think the
cutting edge part matters it's just like there has to be >> one unique insight yeah one unique insight one unique take something new >> doesn't necessarily even have to be new
>> I'm trying to like >> you just have to be But >> I'm trying to like explain it for people because you know a lot of people they want to build something and you know
they start off by like building something that nobody wants. So like >> I think that there has to be something that there has to be something that gets people excited about it. Like the idea
of like why am I building this like ghost TV was built for it? I want it to be fast. I want it to be like I remember I watched them very early on when when he started doing the dev blogs. It was I
remember like okay he wants to make it native to the platform and fast. It's like, okay, there are two things that there has been native or fast, but not both. It's like, okay, cool. If he's
going to achieve that, there's going to be enough of a reason for me to use it, right? But it was in that sense, it was like it was a good story. There was a lot of energy behind it. Um but for
instance like Django as a web framework I still adore to this day because like primarily what Django really did was yeah it was a little bit new in that it has the admin of everything but it it
was primarily it was people kept building it and maintaining it and like putting energy into it and it felt like you can actually fully rely on this thing because it's going to be there
decade later and it is still there like two decades later. Um and so in some way or curl is a perfect example like everybody uses curl. Curl is not unique or like um like cutting edge. It
is just very reliable and it makes it a very good choice and maybe like that was not obvious in year one or year two of that project but it's very obvious nowadays. Um so so some things are just
literally because people keep putting energy into it. um like century is the same thing. It's like was sentry the best choice when we started. I would argue it was but there was a lot of
competition on the market um for this but many of them just didn't have the longevity and like continue putting energy into it. So like eventually only century remained um so like some
projects are just literally this is like keep doing it. Um but then it's like obviously not an overnight success and it's it's it's going to take you years. So if if you're like after what can I
build that I can sell to like another backholder after 6 months then you you you work differently but that has never been my my motivation. >> Yeah that's very helpful. Let's let's
talk about the I guess open source because I think there's a trend of more and more of the projects that are being released and are like taking off are open source. Do you think like it is
because people you know trust it more if it's open and they can build on top of it and they can look inside and they can you know fork it easily. Do like do you see the percentage of software in the
world being more open or do you see it like constant? How do you think about this? Are you noticing any interesting trends in in this direction? Um I don't know but I here's some bunch of
arguments that I think uh why um so first of all I think the general move is not necessarily there's more open source a very classic move very annoying move has been you start out
open source and you close it I mean this this has been the move forever >> um in the AI space incredibly common like Daytoner started out open source and then found an argument why they
shouldn't be >> um even Cal which like was a company that um or kel.com or whatever whatever was a that like say like oh we are great and open source like even them are no
longer open source. So I think like I think like the the the careful thing is like who builds an open source project and continues being an open source project is a very different thing from
like who started an open source project. Um one reason I think you're going to see a lot of AI code being open source is actually a completely unrelated reason to if they wanted open source or
not is because if you're an open source project you're going to get free infrastructure. So the cost difference of having GitHub actions on open source versus closed source is a lot of
dollars, right? So it's like um and and presumably the cost of someone cloning someone else's code is very little. So if you want to have free infrastructure, you build it open source. That that is
the status quo right now. How long Microsoft wants to continue doing that is a different question. But um obviously there's a huge monetary advantage right now um to start at open
source. like it's marketing and you get free infrastructure. Um I'm just afraid that longer term it destroys open source um a little bit or at least it's going to make some it will have some negative
impact on how open source works because a lot of the people that are now open sourcing software have no idea about licenses, what it means to be an open source project, how to actually make
finances work for an open source project. For them it's mostly just a marketing channel. Um, and as someone who has been working in open source for a long time, that kind of upsets me a
little bit. Um, but maybe that's just a temporary thing that will play itself out in one form or another. >> I was asking from like the, you know, end point. Again, I'm not saying it's
going to be like a year or two. I'm saying like you know like you said if cost of replicating software is cheaper and cheaper doesn't that like track as you know agents get better models get
better people will be like able to replicate more and I'm not making the argument that like everybody will you know replicate their own slack or their own CRM like I'm not trying to make that
argument I'm trying to make the argument of like as agents write more of the code and it's easier to build basically anything do you think more of that will be released open source even if it's for
the marketing reasons like it doesn't matter what the reason is I'm just thinking like is the trend >> I really I don't know I I I I I I wouldn't even know how to know it
because like I don't know what the percentage of closed service open source code was in the past let alone like how it's now one thing is pretty clear there's a there's an inherent advantage
if you build something that you want agents to build when it's open source because it's maybe unless they're changing something going to end up in the training data
>> and that is that is that is actually pretty significant issue. Like I I I had a conversation even with with with people within Microsoft where um they were evaluating the quality of what the
models were doing for um outside code versus what it means to write an Xbox game. Uh because like very little console code has ever been publicly shared. In fact, it was pretty much
disallowed. Um and there was also push back within Microsoft as far as I know to um to make the Xbox training the Xbox code available to models um because they were afraid of it. So I don't know where
that landed but um this is sort of the implicitly there's a benefit of uh being if if you want others to build on your stuff with agents. It's great if your stuff is at least
in parts open source because then it will end up in the training data. >> Um >> but again like this space is moving so quickly. Who knows what's going to
happen in two to three years. like it's and then also it doesn't really matter like nobody's choice is going to tremendously change. They're going to do whatever they want to do anyway. So I
think like people are doing what comes naturally. Um I hope that people are going to continue maintaining really great open source software. Um that's what I hope. Um and I don't think that a
lot of the projects that are starting out now have the right reasons to doing stuff open source. Um that probably was always the case. It's just very noticeable now I
would say. >> So what is the like is it like the raw technical capability? Is it like the thought like how would you put like a mediocre open source project or a great
one? You know like what are the things that actually >> is that project going to be there in 10 15 years? That's like the only question really that matters. Is that project
going to be there in 10 or 15 years and still open source? If yes it's a good open source project. If not, then it's not a good open source project. Like a bad open source project is one that
goes away, has no future, turns closed source, um gets abandoned, that's a bad open source project. Or it has no users. But if it turns out to be useful for society, continues to be maintained, um
people build stuff on top, then it's a great open source project. So it's purely just something we can judge in retrospective. It's not something we can judge now. Like as an example, I don't
know if you've ever downloaded PHP 1.0 0 or PHP 2.0. You can go to museum.php.net. You can download the old tarbles. It's horrible.
Like the old P like it's objectively horrifying what the old PHP looked like. And I remember when I started doing PHP early and like um and then there was PHP three times I think and then that was
already the good version of PHP. There were so many people who were saying like oh like PHP is trash like look at look they don't like the guy doesn't know how to make a programming language. Um, but
look at PHP nowadays. Like it is an actually really good programming language and all the people that worked on it learned a ton and you can build very productive code with it. Like
Laravel is a great ecosystem. And so like if you were to sort of judge PHP as an open source project on like year one based on what Lars did when he started out, you would have a horrible opinion
of that language. But nowadays you can look at this and say like this is in fact a great open source project because for a start it's still around. It's still eval evolving. It still has people
supporting it. Right. So we we can really only judge it from from from after the fact. >> Yeah that's a that's a really great point. Like basically half the success
is just sticking with it and you know keep improving over time and >> that's how you gain the trust of people to build on top Pi will be a great open source project if it has meaning a
couple of years from now. If it turns out to be not needed anymore, then it was an interesting blip in moment of time, but it was not a great open source project.
>> Let's talk talk deeper about the being in a training data because like that's a very interesting point and we're starting to see it with like you know when you start a new project what the
models are pushing you to choose as the text tag as the deployment platforms. Do you think like a lot of it has been already written where like you know version control it just goes for git
right or like like you mentioned the operating system they're great at Linux because there's a lot a lot more of that in the training there compared to Windows or Mac OS. So like how do you
think this impacts the future? Um under present day model architectures being in the training data is a great benefit.
Um, and at least that is the general um I think agreed upon interpretation. Um, and I don't have a counter argument to this. Um, and and
that goes as far as I have observed that people that are in fact in the training data are banking on this. Like I I know that there are teams um in in companies that are under represented in the
training data are trying to figure out like how do they get into this because like if you are if you're the top choice of an agent to do something like I I don't know who which company it was but
I think like one of the companies for their space figured out that they're really good in all the models and I believe they if I remember correctly they had a billboard on a it wasn't a
billboard it was an ad on a bus in in San Francisco where they said like hey do you need I was like observability ask ask your AI what what it would use right so there was a power move that you could
do because you were the default choice um and so so there's obviously a huge benefit to this particularly because people are increasingly using with these models as
a replacement for Google search um obviously to some degree if you have web search also in it then it's also an SEO question right because like the agent might do some investigation rather than
just going to the weights um but but there's obviously benefits. That said, that alone doesn't really help you fully um one because um these agents will always
have to solve problems which are uh in your codebase which is hopefully never in the weights unless you work on open source code. Um so that enough work and training will have to go into in context
learning and and making that really good. Um so maybe it's not so necessary because you can point people at documentation and so forth. Um
but yeah, I mean in general I think people will do everything they can to get into the weights in one form or another. >> I don't think it's changed.
I want to talk about like more future things like recently I had the realization that like inference is going to become more and more of the spend because I mean first of all we you
know I can look at myself and like I don't really spend more money on random vibe coded AI SAS but I spend more money every month on AI inference right and I don't see this stopping I mean sure
everybody wants to like make it more efficient and you know spend optimize their tokens get subscriptions not like pay API prices but Like do you see this again assuming model progress will
continue harnesses everything will keep getting better. Do you see this becoming like eating more and more of the software market whereas you know people in the past were paying some consultant
or some researcher now can they can just do two prompts as a deep research and get their favorite agent to like answer that question. So that spend goes from that consultant to the inference like do
you think it will just eat the soft market? I think it will depend on what the cost of tokens will be presumably if the cost of tokens will go down which doesn't see so cost of tokens is going
down today but simultaneously the actual cost of solving tasks still seems to be going up um at least from my personal observation my sessions are not getting any cheaper like I'm my my my
general spent on what I'm doing um is is not having a downwards trend let's put it this way Um, and and if I were to have to pay API prices rather than subscription prices,
I don't think I would actually be using AI much. Like I I I would feel weird burning that much of my investor capital on AI. Maybe I shouldn't because like that's what they
want, but I would personally feel weird having such an expensive like expenditure to inference um within my company. So
I think it will purely be a question of like how expensive is it because if if it's very like if the if the human in comparison can be both trustworthy
and be um more reliable and they themselves then use AI to like augment their capabilities, there might actually be a better deal than me trying to get rid of all the humans in a company and
then just purely using AI for especially at the prices where we already are. But I mean I I can't foresee the future. Um it's also going to greatly depend on like what society wants. It's it's very
clear that outside of programming AI is a mixed bag. people are using it for sure, >> but they're using it in the same way that they're using smartphones and Tik
Tok and and YouTube where like plenty of those are saying, "Yeah, I'm using it, but I still hate the companies that make it or I hate the fact that it uses so much energy or I hate the fact that my
electricity price are going up." So, like they can both find it somewhat useful but not really like the technology. And I think depending on how much they like the technology that will
really greatly impact what companies will do, what society will do and and that I don't think I can predict in any form. I think this can go either way. Um okay, I have two thoughts like I think
first you could see your token spend going down if you were optimizing for it, right? like you know if you really had a small budget and like you were like stacking o open code subscriptions
like you you could get more and more every month with you know better open source models for tasks completed that cost less. So I I think like part of that is just like maybe you don't care
that much to optimize for it to go down and you know you're comfortable spending with your spending like if you didn't have the money you would probably find a way to run some AI maybe not the same
quality of models but like you know cheaper subscriptions and I guess the second thing I I want to say is like that uh yes you still want to talk to a human in certain cases and if that human
is using AI then maybe like if we say like accountant or whatever tax adviser that accountant that's AI enabled can serve more clients Right? And as a function of that, he can charge less
price per client but like accept 10x more clients and undercut his competitors who are not using AI. So that is still like moving the cost to inference because he will make more
money but his margins will be lower because he's using AI to fulfill his work. >> Yeah. I mean I think I got a general thing I'm worried about is like
percentage wise a greater part of the um like in in proportion to European GDP more money will be funneled to American companies. I think that's just like that's a general worry that I
have because like Europe >> yeah so like I I think that that is that is most likely going to happen in one form or another. Where it lands I don't know. um as a as an example for instance
um I feel like I don't know like what the trend here is but at one point um you would find no designer using AI
but you could go to chat GPT and make yourself a picture. Then there was a percentage of designers that would use AI with you. So you you say like hey I want
this and this but you also like 30% was slop right because like all of the sudden like there was AI generate stuff in your image and at one point that became at least in
certain circles to be such a negative sign that you now almost pay a premium for that person not to use AI in the first place. Like I know that is a topic for instance with um with ads like
>> there are the ads are representing your product and there are people paying premium now for their ad not to show any AI in it or at least obvious AI in it because like their customers having
negative response to that. Um, so I think that will greatly depend on what it is that you're doing. And I think there's a there's a chance that if more and more lawyers are going to use AI and
then maybe like some of their court cases are actually they're losing their court cases because maybe they were badly researched or they made us some stuff. um then like there will be a new
balance to it and like on the on the upper end of it then maybe they will still use AI but you're not really paying because they're not doing more more like court cases but in fact
because like they're winning them better because they are they found a good way of using it. Um but I I don't think it's so simple to say like okay if we give every doctor in the world now access to
AI they're going to handle I don't know 10 times as many patients and um and as such the quality of like healthcare will go up. Um that's it could be >> well I mean I didn't mention quality I
just mentioned like the spend of inference right like the quality as a whole another topic I meant like the percentage of software spending just going to plain inference
>> but presumably presumably there's a counter move to that because like if if >> yeah maybe that maybe there is maybe it doesn't follow maybe maybe it is in fact yeah
most likely what's going to happen is that how much we're spending right now on tokens is a fraction of what we're going to spend a year two years three years four years from now.
>> Yeah. >> Um like as a society at the very least. Um I think that is probably going to be true. However,
are the tokens still going to be that expensive if they don't deliver quite as much value as we assuming they they will. It's a different question, right? I think that there's it's not completely
impossible that we're building too many data centers proportionate to how much it costs us to do that and how much the companies can actually charge. Um there's there's fierce competition. Um
so maybe the prices will just have to go down in the same way as the prices for like electricity go down or should go down if there's if there's sufficient supply. Um at the very least solar is
getting cheaper. So maybe tokens will eventually also get cheaper. I I don't know. I I really don't know. I what I'm what I'm hoping where this lands is that society rebalance itself around this
somehow where um it doesn't end up like social media did where like most I my my suspicion is that at least in Europe the majority of people hates it now and still uses it,
right? It's like >> social media or AI? social media like social media like short form video content like everybody's using it. >> Yeah.
>> But if you if you make a opinion poll on it, >> people are incredibly frustrated with it, but they also have they feel like it's almost like an addiction like like
smoking is. >> Um and so like I I would hate that we built AI in a similar way where like everybody's using it but simultaneously everybody's hating it. Like that would
be an unfortunate outcome. >> Let's talk about Europe. You mentioned it a few times. you know, you're from Austria, I'm from Czech Republic. Are we just falling behind? Like h how do we
save this continent? >> Um I mean, we are falling behind. I think that is uh that is probably true. Um how do we save the continent? Um
a good question. Um well, I think like I I know a bunch of ways in which we're not saving it. I think like right now what's definitely happening is that all the motivated
people or not all the motivated people but a sufficient number of motivated people are just going elsewhere. I think that that probably is not great right. So if it like the starting the starting
situation is not overly amazing. Um and I think a a fundamental issue that Europe has is that Europe in in sort of in the stage in which it is is all about preservation of the past and not so much
about enablement of the future. Um we have a lot of examples of this but Europe also has some things figured out that maybe some other countries have not figured out and if they get into that
position they will have to figure this out. um like a a growing middle class in China will have different demands to to to their lifestyle than than than China did up to this point, right? And so
Europe is a society where people want a lot of rights where where where things are balanced out more and that makes it also harder to start new things because you have to run up against all
the regulation that's in place and it's not a question of politicians, it's a question of like what does society want and society wants it that way, right? Society doesn't want data centers to the
degree that maybe some other countries want, society doesn't have a great optimism towards large tech companies and if anything is happy that they're not here. Um, so
given how the how the European societies largely work, I think it's sort of a tricky situation. Europe's biggest issue is that Europe is not one country. I mean that like the US is a massive
single market. China is a massive single market. India is a massive single market. Russia to some degree is a pretty big single market. Um, at least Russia plus the other six countries
together are um Europe is 27 countries that are constantly fighting against each other and then the UK fully left, right? So like it doesn't even a question of language. It's just
literally a question of like does every European country give the full worker rights to every other European citizen? And the answer still is like there are a lot of like all the country are trying
to put as much friction into it as possible to benefit its local population over everybody else. Like we have 27 armies, we have 27 um legal systems, we have 27 um um
worker regulation pieces. Um like >> starting a new company, you have to like VAT in every country. It's terrible. >> Yeah. It's like there's there's just a
lot of it, right? and and it's it is because we haven't quite managed to come over this idea yet that um uh we're just Europeans and and I think like right now the the mindset is not quite yet to
enable that. Um and so maybe that has to be the first step or at least it has to be pretty significant step. I think if compared to where I grew up, the times where I grew up to now, I feel like it
did go in the right direction because I remember like Austrians were deeply critical of every neighbor country that was not German speaking. Uh I think that has changed. Um like I I I I if anything
now it's all about the countries that are not in the EU yet. It's about Montenegrians, about Albanians, it's about um maybe to some degree Ukrainians. Um so like if if there is a
suspicion towards other Europeans it is sort of at the boundaries of the European Union and maybe like plus minus Norway, UK and Switzerland. Um but like there's there's less suspicion
I would say. Um but there's still a lot of suspicion and I think that holds back Europe just intrinsically and I don't think that I don't think European politicians can fix that. I
think this is just a question of like how how does society get to that and I think then society will have to start pushing back against some of the that happens on the European Union level
and that happens there not because European bureaucrats are idiots it's because like the countries themselves are like these compromises are coming because 27 different countries have to
agree and they and they're doing what what what sort of comes out of it. Um that's that's the unfortunate reality. >> All right so I think we can end it here. I want to be respectful of your time.
Appreciate you coming on, Armen. Where should we send the people? >> Where should we send the people? Send them to pi.dev arendel.com and um yeah, probably there.
>> All right, I'm going to link all of that below as well as your socials. Again, appreciate your time and have a great day. Right. Thanks for having