🚀 Add to Chrome – It’s Free - YouTube Summarizer
Category: Tech Trends
00:00
I was asked to talk about the future of software. So, a lot of this talk is going to be about what we're doing at Replet, where we think the future of software is headed and some kind of trying to make some predictions or try to think out loud about really what the
00:15
future holds. my mental model for for our business and really for the moment we're in today.
If you think back uh uh on the future on the history of computing, mainframes were kind of the the first mainstream computing devices
00:33
as mainstream as it gets back then. And to use a mainframe, you needed to be an expert.
And then PCs came around and initially PCs were kind of toys. you bought a Mac and you did Mac Paint and
00:48
things like that. There wasn't real business use case.
I mean, people like made fun of Apple at the time uh until the Excel sheet. The Excel sheet was the first software that was actually useful on computers.
And now
01:05
PCs run world economy. Like they actually if you go at a data center, it's also only PCs.
It's x86 computers. So you go you go from something that was used by a small group of experts that had to to have a lot of training to
01:22
something that started sort of as a toy and is used by by everyone. Same thing with software engineering like uh the modern software engineering career you can sort of trace it back to the u 70s with the rise of maybe uh uh Unix and
01:39
the C programming language. That's when people started uh kind of being trained to become software engineers.
You still needed four, five, six years of uh educ college education. You need another two or three years of uh training on the job
01:57
to be able to actually do the job very well. I think today software is going through the same transition from something that only experts do to something that anyone can do.
02:13
And this is what we're really rep uh building replet for. I've been working at Replet for like almost nine years now.
And our vision has always been to solve programming to like make programming so make it so that anyone can uh write software. So we built um an
02:32
IDE, we built uh language runtimes, we built like a online sandbox environment, we built deployments, we built cloud services around all of that. And then when AI came on the scene, we realized that the ultimate expression of our mission is to make it so that you don't
02:48
have to code. code is the sort of bottleneck to actually getting a lot more people making software.
So around, you know, late 23, early 24, we decided to put all our resources into agents. At the time, agents sort of barely worked,
03:05
but you could tell by looking at a few benchmarks that were headed there. So, SWEBench is a software engineering benchmark.
Uh it is basically a collection of um issues on on GitHub from major repositories and the unit
03:20
tests and pull requests sort of end state of those issues and uh and the way you test an agent is you put in an environment and have it solve some of those issues. You could tell like in in 22 sort of it barely worked.
23 started sort of working and you could tell early
03:36
sort of early 24 where we're on this trend where you could tell that software engineering is getting automated or like big parts of software engineering is getting automated. Uh and now we're probably I think this is like a little outdated.
We're like at 70 80% Sweetbench. Now if this benchmark gets
03:53
saturated doesn't mean that we automated all of software engineering but we're on on our way to make really useful arguably it's already here really useful software engineering agents and by the way this is true of any agent if any any of you are building sort of agents
04:09
startups uh just like really believe that it's coming really really like I I keep telling my team we need to be okay with building crappy products today because two months down the line the models will get better and your business
04:26
your product will suddenly become viable. So to today's kind of the moment for for uh for agents so rapid kind of went all in on agents but agents that can write code uh is actually the easy part.
The hard part is the
04:42
infrastructure around it. Sometimes I call it the habitat for which the agent uh lives in.
So what you need is you need a a virtual machine, ideally in the cloud, ideally not on your computer because you know agents can actually
04:57
also mess up your computer. They could do a lot of scary things.
So it needs to be sandboxed. Uh it needs to be scalable.
If you're running a product like Replet, you need to be able to, you know, scale up to like millions of users and uh you need to be able to support
05:13
every language out there, every uh package out there. Um the way uh software engineering agents are trained today is they're trained on standard Linux environment.
They need to be able to use the shell. They need to be able to write to files, read files.
Uh but
05:30
they also need to able to install packages either system level packages, Linux packages, but also language packages. And many cases agents want to actually use more programming languages.
And so a lot of environments today where people are trying to build agents are very constrained. But what you what you
05:47
want is an environment as open as possible similar to the kind of environments that software engineers work in. So what kind of other things you need to ship real software you need deployments, you need databases.
Uh really think about everything you do as
06:02
a software engineer and all those tools need to be accessible uh to software engineering agents. So actually I saw earlier today if you were at uh Karpathy's talk he he talked about how you know the the coding part is the easy part.
So sort of similar to the points
06:17
I'm making but he talked about all the different things that are really unsolved but in reality we actually solved a lot of them. So replet of the gate comes with o uh agents are actually not very good authentication.
It's better to use a service built-in service. So replet actually one line of
06:34
code we turn on co uh off. So when ask replet agent to integrate off uh it will actually just use replet off.
It will just like basically turn on setting and and then uh you have user authentication you have user management those users
06:50
information are being stored in the database. You can also obviously deploy the app you can uh link a domain to it.
We have secrets management secure ways of kind of uh using API keys. We have background jobs.
You know, a lot of applications need to be able to run
07:06
continuously in the background, especially in this era of agents. Storage again, uh, you know, agents need to be able to store things.
They need to be able to grab things from the web, images, documentation, whatever, and store them for the application to use them in the future. Few other things on
07:22
the road map, universal model access. So, it's really a pain right now to ask the model to like to ask for an application that can generate an that can do something with an images or videos.
You have to figure out which model to use. You have to go get an API key and do all of that.
Pretty soon, any
07:38
model that you ask for at Replet, it'll be just available in your app directly. We'll handle the the billing and the API integration, all of that.
Payments is very important. Payments not just for your users to pay for your application.
07:54
Say you're you're building a startup on Replet. You're you're an entrepreneur.
you obviously need to collect uh user payments, but also I think some sometime in the future you would want your agent to have some kind of wallet to be able to go uh pay for services. So let's say
08:12
you know your your agent decides that it needs a Tulio integration and replet or whatever system you're using doesn't have a tool integration, it should be able to go put in its credit card and provision that service in the in the background. A more radical idea is that
08:28
your agent needs to be able to like hire people. For example, if it hits a capture and it doesn't know how to solve a capture, it should go and task grab it and ask a human to go solve the capture for it.
Whatever it is, there's a lot of tasks that you still need humans for and you would want your your agent to be
08:44
able to uh to have money to pay for services. And similarly, agent to agent.
Um, you would want your agent to be able to go on the on the market and find other agents that can it can hire. Too many YC startups are building uh agents
09:01
sort of agents for accounting, agents for sales and so you need your software engineering agents to be able to integrate those agents as well. So it's so I I know a lot of people think of MCP as such an agent to agent tool but actually MCP is a more traditional RPC
09:16
protocol. So it's not really going to solve this.
Another model on our sort of business or technology is think about sort of the level of autonomy. So when I started working on what Replet would become like years ago, perhaps decades ago, the state-of-the-art code assist
09:32
was a language server, right? That's IntelliSense if you're using VS Code.
And you can think of it as level one autonomy. You know, if you think about the uh sort of drive assist in self-driving cars or like in cars, you know, would be kind of the lane assist,
09:48
that would be the first level. Uh AI code completion co-pilot, that would be level two.
Uh level three is what we worked on um when agent when replet agent first launched. Agent v2 I I would
10:03
call it almost 3.5. It can work up to 10 15 minutes on its own but it still needs your input every now and then to test the app and make sure the app is working.
And right now we're working on V3. I'll talk a little bit more about V3 in a second.
Uh but V3 is sort of level
10:20
four, right? Like you're almost there.
It still needs some of your attention, but it it kind of works fully uh autonomously. Bore Plus, which I assume we're going to get to in the next couple of years, you can really spin up a thousand agents,
10:35
give them a lot thousand problems and and reliably be confident that like 95% of them is going to work. Like we're going to have a really high reliability rate.
any kind of engineer or product manager really anyone can spin up
10:52
hundreds if not thousands of engineers to do work on their behalf. So they need very little supervision and therefore you can uh increase your impact exponentially as as a as a programmer.
So what we're working on right now with uh Asian v3 is uh that you know it's
11:11
based on uh basically three uh three pillars. One is uh end to-end testing.
So today computer use is um so in models what's called computer use if you've used openi operator it's the idea that
11:29
models can go into a computer you know click around and use a computer like a human does they're slow they're expensive they're not very good but this is what I talked about earlier you want to build a product at the edge of what's
11:44
possible right now the edge of what's possible is like computer use in my opin is really at the frontier of what these models could do and I think over the next 3 to 6 months they're going to get a lot better and it's going to enable an entire new market and also probably
12:01
start to automate a lot of real jobs once we have app testing uh you know this this kind of annoying thing that replication does where keeps ask asking you to do QA for it'll start doing QA on its own and that will allow it to work
12:17
you know 30 40 up to hour maybe two hours of work. Sort of the hype today is test time compute.
If you think about the sort of 03 uh or like oer models or deepseek R1 the kind of main insight there is the more tokens the model is
12:34
able to consume or produce uh the more intelligent uh it gets. Now today with something like 03 the model is generating a lot of tokens and trying to reason but a lot of it is sort of
12:50
synopsistic. It doesn't get feedback from the environment.
It's almost like it's just sitting in place and thinking. What you'd want in a real computer environment is for the model to generate hypothesis and test its hypothesis in real time.
So at Replet, we built a uh
13:10
fully transactional reversible uh file system. So when you're on on Replet, every edit you make to the file system is an atomic snapshot in time.
And that allows us to have very cheap copy and
13:26
write forks of the file system. And so our idea for this is that um anytime there's a tough problem or basically if you have a lot of budget you can have it on all the time but every time the agent is making a big change it forks itself
13:45
and the environment in number of times to solve this problem in different ways and then find the best solution and then take that solution merge it into the main branch. So think about you know the idea of
14:00
simulations like when you're thinking about the problem you're often simulating different branches of things that you could do you have different hypothesis you want to test and so we want to also uh give agents the ability to do that.
14:16
So at any given problem generating a ton of different ways of doing it and then testing all of them in parallel. This will bring up reliability of agents by I think two to three folds.
So that's sampling and simulations. And then
14:32
finally is uh for the uh model to be able to generate tests for every feature that it creates. Today replet replet agent often creates a feature and then later on breaks that feature but also
14:48
true of clot code and cursor and all the others. So we want to make it so that once the agent makes a set of changes or feature, it always has tests that it runs on every change to make sure it's not breaking the software.
This is actually harder than it sounds like it
15:05
sounds like okay write tests and let's run them. But often actually models are pretty bad at generating unit tests.
So there's still a lot of work uh to do there. It needs to be fast as well so so that it happens on every change.
So I so
15:21
that's what we're working on with V3. That's a lot of infrastructure work.
Want to create the best habitat for agents to to live in and be able to u the most be the most reliable possible. But like let's fast forward to like what
15:37
I talked about with like level five autonomy. Really the most autonomous system we can think of.
YC's next batch is now taking applications. Got a startup in you?
Apply at y combinator.com/apply. It's never too early and filling out the
15:52
app will level up your idea. Okay, back to the video.
My prediction is that all application software will go to zero. In other words, software will be dirt cheap, that no one will be making money on the traditional type of SAS software.
I'm
16:09
not saying this will happen tomorrow or even next year. I gave up on the the um trying to predict timelines.
I know it's going to happen on the order of years. If anyone with one prompt can generate any kind of software of any type of complexity, then um the value of
16:28
applications will go down to almost zero. So what does that actually look like?
So today you know in the startup ecosystem in the tech ecosystem there's all these generic generic SAS you know vertical SAS software and any of you
16:45
who's running a small business or even a bigger business you you probably have bought you know dozens and dozens of SAS software just to uh just to run your business even today you're able to replace large parts of those software by using something like replet agent or
17:01
writing your own software uh I think in the next few years. This again this will go from uh you know maybe 15% replaceable to 100% replaceable.
So this will really fundamentally change the software market. Uh just to give you a story uh one of our colleagues at Replet
17:18
uh Kelsey was uh she works in HR uh she's never written a line of software in her in her life and she wanted an orchar software. She had a few bespoke needs like she wanted to connect it to
17:34
ADP, our our sort of payroll software and and she had a few features that that she wanted and she went on the market and she couldn't really find an org shot software that exactly fits her needs. They were very expensive that were going
17:50
to cost tens of thousands of dollars a year. So she decided to make it.
She took a week, less than a week, three days, and she made orshot software that we're using today that we can go out on the market and sell it as a SAS product for tens of thousands of dollars a year.
18:05
So that's like mind-blowing, right? I mean, it's HR professional can make software to run their work.
That's happening today. Try to project that out a couple years later.
Like the software business fundamentally changes, gets
18:20
disrupted. Not only software but I think how we work, how businesses works, how corporations work will will fundamentally change.
Today we have these roles you know uh
18:35
companies like to specialize since the industrial revolution when factories you know became the main mode of creation. the sort of modern uh special, you know, specialization in the economy kind of
18:52
emerged where one person is making one part of the the the product. It goes on a factory sort of um assembly line and another person is responsible for testing it, another person responsible for assembling it.
And so this this um
19:07
specialization has been the way the economy has been trending for a long time. And it sort of makes sense, right?
You want to uh specialize people as much as possible. You want them to be as as replaceable as possible.
And so this is how the modern economy is built. But once your HR professional is also a
19:26
software engineer is also potentially a marketer is also potentially anything because they can learn anything. There are AI agents that can do anything for them.
really, you know, you go into the world where jobs will become less specialized, less um siloed. And in
19:44
fact, we started, we're seeing it today and we're at Replet the way we're structuring our orchard and our business based on this idea. We're building for the first time, we're building a like a actual product team, product management team.
And our product team is actually
20:00
made of designers, engineers, and product managers all almost always in the same person. So we're trying to merge a lot of roles together and create this generalist employee.
So the the arch will start to look more like a network than a hierarchy. So it'll look more like an
20:17
open source project than a than uh it will look like a traditional company hierarchy with a marketing department, sales department. Every employee will like wake up in the morning and their mandate would not be write this
20:35
marketing email or you know make this uh optimize this button. Their mandate would be make the business work, generate value for the business.
So everyone is sort of an entrepreneur and that will really disrupt and fundamentally change how companies work.
20:52
It's a model that really we haven't you no one has really embraced or or or even started to talk about but really you know think it through if everyone has access to a general purpose software engineering agent and sort of agent for every possible role obviously domain
21:09
expertise is still important but is not as important as it used to be. It's exponentially less important and this also affects how people build businesses.
it affects the opportunities that are available for us in the future. One uh really interesting book uh that I
21:27
read this book is was written in the 80s which is insane given how good the predictions were. So I'm just going to read this.
Ideas will become wealth. merits wherever it arises will be rewarded as never before.
In an
21:44
environment where the greatest source of wealth will be the ideas you have in your head rather than the physical capital alone, anyone who thinks clearly will potentially be rich. The information age will be the age of upward mobility.
The brightest, most successful and ambitious of these will
22:01
emerge as truly sovereign individuals. Now, some of these various is a bit dated.
The information age perhaps we call the intelligence age today. But this book predicted things like uh crypto, remote work, all sorts of things like that.
And this idea of like a sovereign individual, someone so uh
22:19
empowered by technology, so empowered by these uh uh agents uh that is able to create enormous amount of wealth uh individually is is uh is is going to be the norm. Think
22:35
think about um someone like Satoshi. Satoshi created a single person created a trillion dollars worth of value.
I don't know what the market cap exactly. Perhaps it's more than trillion dollars of Bitcoin.
But like that's a single
22:52
person. They wrote the paper.
They wrote the software. They put it out there and it became a big thing.
Obviously there's a lot of people. It's a big market right now.
But it was created by a single person and we don't know who they are. And I think that's going to be a common occurrence in the future.
The really
23:07
great thing about it is really um the access to opportunity will be universal. The idea of merit being rewarded wherever it arises doesn't matter if you're in Silicon Valley or anywhere else in the world.
If you can think clearly and you can use some of this
23:24
technology. If you can think clearly and generate good ideas, go into replet, put in those ideas, make the first version of software today, you can start to become more like the sovereign individual.
Again, the way collaboration work will be will be seamless. You know,
23:39
everyone's talking about the you know $1 billion single person company, but I think that really kind of misses the point a little bit. What's really interesting about it is that you'll be able to assemble groups of people really quickly.
You'll also be a be able to assemble uh groups of agents really
23:56
quickly. You'll be able to assemble these companies and also unwind them.
You can create mission purpose, you know, companies or like projects and unwind them really quickly. And in some cases it could happen in a day or two.
And sometimes you might be you might think you're working with another human
24:13
on the internet, but they're actually an agent built by someone else who's out there doing work for them. So the way we work uh and the way people build startups will fundamentally change as the the cost of transaction goes down
24:28
goes down to zero then um the the the reason to hire an employee full-time uh you'll have less of a reason to hire full-time employees. So think about um like getting an Uber today.
The
24:43
transaction cost the kind of effort of getting an Uber is just one button on your phone. I think the same thing will be in the future to to get a developer whether it's a software agent or another human being.
Uh it'll be just like one button. I want this problem solved.
24:59
You'll be able to maybe your agent will be able to go find and interview a lot of different people or agents on the internet and be able to find the best uh thing to solve that problem. And um and so you'll be able to like build
25:14
businesses really at the speed of light. Now you know I I I talked about how kind of application software goes to zero.
That doesn't mean that all software goes to zero. Today you know rapid agent or others the way it works is the agent makes a piece of software the user uses
25:30
the software to solve problems. You can think of those things as intermediate steps.
Instead, agents can just solve problems. And and for Replet, and I'm sure a lot of other businesses to survive, at some
25:46
point, Replet needs to stop being focused on making applications and start being focused on solving problems with software. So, I want to leave ample time for for questions.
So, I'll I'll I'll end here and open it up. My name is Chinat from Stanford.
Nice to
26:03
meet you. My first question is in this future do you see there potentially humans engaging with multiple agents or will there be a unilateral agent and if in the case of like multiple agents um how would we deal with the fragmentation of like data memory and context across
26:20
all these different agents I think multiple agents uh and and the reason I think that's true is because let's let's say I'm someone with true unique domain expertise uh let's say I'm I'm a lawyer who is top
26:36
in the world at solving certain um cases that that are very rare. And so I have this domain expertise that I'm not going to share in the open source.
I'm not going to sell to scale AI so that
26:52
they can sell to to OpenAI or Google all of those. I'm just going to keep this resource to myself.
But the way I would monetize it instead of myself going and selling my services directly, I would like imbue this knowledge into an agent that becomes this very specialized agent
27:08
in this very specialized domain and then I can scale myself. uh and so so I think I think people will be building these agents to work on their behalf and then um there's going to be agents that that
27:24
uh go out there and assembles these teams of agents uh and then there's going to be obviously software development agents and maybe there's and maybe you're running all all this through chat GPT or whatever main interface you have but I think it's going to be a multi- aent world with
27:41
different contacts similar that we have in the world today. When I go to a lawyer, I need to give them my context.
So, uh, and maybe there are protocols and this is why they talked about how MCP really doesn't solve the agent to Asian problem. I think there needs to be
27:57
more interesting protocols in this space that and maybe this is a startup someone builds. Hi, thank you for the insightful talk.
My question is as follows. in the not so far off future where we're going to have AI systems that can automate most if not
28:13
all of meaningful physical and cognitive tasks and there's increasing delegation to agents that work on your behalf and talk to other agents that are working on other people's behalf then what is left for humans to do or like what will our human condition look like because our
28:29
physical and cognitive aspects can all be done by intelligences I I think it fundamentally depends on your worldview and belief of the limits of AI
28:44
versus the uniqueness and premacy of what humans can do. So it becomes a bit of a religious discussion but my view is there's something special about humans and my view is that there's a fundamental limitation with how we do AI
29:01
today and maybe this gets solved but AI today can't truly generalize out of distribution everything AI can do h needs to be represented in the data say I go back to this example of this lawyer that is expert in the world at very rare cases
29:18
uh again this is something that no one else knows how to do uh or can do or whenever there's like a truly novel problem, truly novel case, you still need human ingenuity um to solve that problem. And so I think
29:37
humans will be more in the creative seat and I think agents can be creative as well but their type of creativity are not net new knowledge. It's more like about uh which is a lot of what creativity is bringing a lot of different things together.
And so but
29:53
but this idea of like ideas become wealth uh is um is what gets really exciting about it is like people can generate novel ideas and test them out really quickly which you know I don't think we're going to get to a to a point where you can go tell an agent hey go
30:10
find me a you know business idea and go test all of them. I don't think we'll get there anytime soon.
Thanks for your talk. I've been following Replet for many years and that's actually where I learned how to code as well was on Replet.
So you mentioned the value of clear thinking and ideas being the future. Do you see this as an argument
30:26
more towards a favor of a liberal arts critical thinking model of education instead of a more STEM uh skills-based focus? I don't I don't think they're mutually exclusive, but I do think that the liberal arts will become more valuable.
I I think today engineers
30:43
tend to be uh a little more parochial than they can afford to be in the future because you know what I shot what I showed with the kind of the model for what the future company could look like. Everyone becoming more of a generalist.
31:00
I think today engineers can afford to like not understand even the business they're in. A lot of engineers are just focused on very narrow domains.
Uh so I think people need to have a more broaden worldview and set of skills. So um but I
31:15
don't think they're mutually exclusive. I think you know being being scientifically minded I think is going to be important.
Hi. Yeah.
Uh so I wanted I was more curious as to uh where in the tech stack like is replet making a lot of progress
31:33
because as you said uh replet can do tasks which for for one hour and so given that like replet uses probably closed source models which have no access to pre-training and post- trainining uh where in the text are you
31:49
making that like amazing uh kind of innovation? that gets your models to work auton autonomously for like an hour.
It's what I was calling the habitat of the model. So, you know, the um the
32:05
commercial models can train really great models. They can train them to be as autonomous as they as as possible to be coherent over a long period of time.
But uh us or really any um agent company
32:21
needs to be able to provide the infrastructure for for that agent to to exist in. And so all these components that I talked about.
So one really crucial thing about replet is this idea of uh you could call it trans uh being being transactional or atomic. every
32:40
mutation to the replet computer environment happen in sync with every other uh component of the system. So right now in replet if you go to your history you can see previous checkpoints and you can
32:56
actually go to any one of them and reboot the application in that state and so we think that infrastructure is going to be really crucial for how to make um the models more more reliable. I think there's a limit on how much the training
33:12
can increase reliability, but I think the environment feedback and the ability to try things really fast is the way to get to the upper echelon of reliability. So that's what we're focused on.
Hi. So you talk about the generalist employee and how that's the sort of future of um companies.
I I to I totally
33:30
agree with this vision, but where I find myself stuck is finding roles today that set me up for that kind of future. what what kind of opportunities should we look out for?
What kind of positions should we look out for in startups, in companies that would prepare us with the skills that are necessary in order to be a generalist good employee five five
33:46
years down the line when that finally becomes a thing. I know being a founder is one option but um not all of us want to take that career plunge immediately.
Uh some of us want to work with other people, build teamwork skills and learn all of those other things as well. How do we go about that?
Jo join startups as early as
34:03
possible. Like obviously you can think of it of it as um sort of exponentially decaying uh curve where like being the first being the founder you get the most generalist experience being the you know
34:20
first employee and then by the time you get to the hund I don't know like to the maybe 100th employee you're sort of like you're not getting as much of that journalist experience but like just join as early as you can depending on your risk risk profile and and all of that.
34:35
But even like number 20 at a like a series B company, I think you will get a lot more experience than at a at a fang or something like that. Even if you join that startup, you need to be seeking those generalist opportunities.
So don't
34:51
sit there waiting for people to give you tasks. Have that mindset of I'm waking up in the morning.
I'm not looking at a to-do list. I'm looking at a mission.
and my mission is to make this company succeed or be more valuable. Um, hi, my name is Shivam.
I also wanted
35:06
to ask about the one hour of autonomous like agent development. Uh, specifically like could you like elaborate a little more on how like you and your team approach how long of a time horizon is worth pursuing as opposed to improving reasoning for shorter time horizons.
So,
35:23
so I think the what you're talking about with shorter time horizons is more like uh let's um let's work on reliability um and then longer time horizons like let's work on autonomy um removing the human in the loop and the burden of the
35:39
human to continue to test and give feedback. Um so we're doing both.
When I'm talking about reliability, this is more investing in reasoning and more investing in this parallel agent. um trial and error that I was talking about while we're calling sampling and
35:55
simulations. And then for uh long horizon, it's more about testing, making sure that because as you go longer, the there's like an there's a like a gold drift.
The the agent might start doing things that you don't like,
36:11
but having those guard rail rails of of testing along the way uh will make it so that it stays more coherent uh over time. And then as we collect more data about what fails and what doesn't work, you can either uh like go and fine-tune
36:27
that or you can just like continue to improve the prompts uh and add more guard guardrails to to make it better. So I think both are important.
Hi, I'm Sophia. Um have been thankful for your talk and I've been following you met at AI for developers when you
36:43
were talking about ghostriter and the work behind it. Um, but I'm curious to hear more about how agents are kind of over oversaturating um, uh, certain set certain sectors and um, whether or not that should kind of
36:59
you should consider that when you're uh, working on them or joining a startup that's working on something. I think certainly software is like really tricky.
software engineering agents. There's like there's a lot of people that want to do that.
And if you're coming in late, you want to have
37:14
a truly novel idea to be able to like compete there. But, you know, there's a lot of things like who's who's building the agent for HR or finance.
Uh I know one company's doing accounting. Uh there's a lot of companies doing SDR for
37:30
whatever reason. That's very crowded.
What I would start with is what are you interested in and what where do you have domain knowledge? So the best way to start an agent company is that if you yourself you're you yourself you're like a
37:45
compliance officer, you start uh a a compliance officer or you're passionate about compliance. I don't know who's passionate about compliance, but if you're passionate about compliance, uh, go start an Asian company because you're going to learn the most about it and you're going to have the most domain knowledge and domain knowledge is the most important
38:02
thing to build an Asian company. Hey, um, so if the cost of software and building software is going to zero, then by extension, the platforms which build software like Replet, like the value capture will be going down to zero.
So, how are you planning to make money long term and how are you
38:18
going to compete with like the other competitors like Bolton and lovable? Yeah.
Yeah. So, notice that I I said not uh old software.
I said like application software specifically. So, I think software will continue to run our lives but a lot of it will be autonomous.
So
38:34
for example, I build a lot of personal software using Replet and a lot of it is around managing my life and my family and like you know uh doing a lot of quantified self stuff a lot of like you know data about my sleep and and and all
38:51
of that stuff and then I spent a lot of time like plotting that data and doing all of that stuff like instead I should be able to tell replet agent here are my goals you figure out what kind of software that needs to that we built and you figure out how to how to uh operate
39:06
it and you tell me what wearables I need to buy and what um and and what do I need to log in the morning, what do I need to do and she'll be able to go make the software uh acquire the things that I need in my home, what kind of sensors and then
39:23
solve the problem for me. I think I think Replet needs to become a universal problem solver for our company to survive.
And I think for a lot of the others, you know, I I think it's already, especially the companies that you talk about in the prototyping space, it's already getting really crowded there. I think what replet where really
39:39
Replet excels today is the fact that it's full stack. It can go from idea to a deployed and scaled software.
Hi, uh my name is Emma and I'm really intrigued by your vision of this future where all code is written by agents. But I'm also kind of concerned because there is this kind of known problem where if
39:56
you train a generative model on data that is generated by another model, you get an issue of like accumulating error, accumulating noise. So my question is in this future where code is written by agents, it's tested by agents, is approved by agents, how do we kind of prevent this exploding error problem
40:11
while still allowing these models to grow and evolve? My bet is that pretty soon we're going to move into more of the alpha zero style of training where um you have a more traditional LLM that's trained on all of the internet.
40:27
Um but but then the way to train the next generation of it would be to give it a reinforcement learning environment where uh it's generating a lot of problems and doing like selfplay where it's solving these problems getting feedback on them and doing it in this
40:44
like massively parallel way. I think this is how we're going to get the next generation of software agents.
It's not going to be trained on human code because like you said there's not going to be human code and so we have to solve this otherwise we'll plateau very hard. Hi.
Um, I'm
41:01
quite interested in some of the systems report uh support required for these agents. Um, and I find the universal package manager that you've released and your use of Nyx quite interesting.
And you mentioned this um copy on write
41:17
snapshotting and and uh uh forking and merging. Uh and I'm working on a similar thing.
Well, you should come work out. Uh I was wondering if any of this is publicly available or something.
I think you might be thinking about open sourcing. Um yeah, I mean we open sourced uh some of our package manager work.
We're big
41:34
contributors to Nexos. So we use Nex OS which is a transactional operating system generator is the best way I can I can describe it and and possibly the file system stuff will well at minimum talk about it but this is like active active work right now.
Um but yeah come
41:52
like intern at Replet and learn all this stuff and then go build it yourself. Thank you.
All right. Thank you everyone.