0:00
model agnostic which I think is a is a very interesting aspect of that and that's something that the labs would not be able to do.
Search this video
0:05
Uh if you think about buying your product and your tokens uh at the same place, it's like if you were uh building a plant with machines in there and you would buy the machines from the energy provider and the plug would only work with one energy provider and if you if your energy provider was a Russian gas would be in a pretty bad place right now and you wouldn't be able to connect to nuclear French nuclear makes no sense.
0:34
Um, so you can uh give a round of applause to Stan Hulu.
0:36
[applause] >> Thank you for having me.
0:43
>> I feel like we're super lucky to have him. Uh, he's extremely wise.
0:45
He's going to talk a little bit about building in the shadow of Frontier Labs.
0:50
And then at the end, I want to ask him about company building because Dust has been a company that has been super thoughtful in how they've raised, hired, operated.
0:59
And so I want to be able to share some of this knowledge with you.
1:04
Um maybe let's get started.
1:06
Can you tell us who you are? >> I'm Stan.
1:09
Uh I was lucky to join Stripe pretty early. Spent five years there.
1:15
That was an aqua year when we joined Stripe and then at the end of those five years got interested in into AI.
1:19
uh I'm not a trained researcher but I was uh also lucky again to uh join OpenAI at a time where you could do research as an engineer and uh was all possible and spent three years there working on large language models and maths.
1:37
>> And you decided to leave OpenAI. >> Yes. >> Why? >> Crazy, huh? >> Right. Yeah.
1:46
>> I think the there's a there's a fun fact on this one.
1:48
I think if I stayed at OpenAI, I think it's not true anymore.
1:52
Both series B, but the uh stock option that I gave up were more than the whole company as of six months.
1:58
That's that seems like a bad idea. >> Yeah. Still, do you regret it? >> No, not at all.
2:05
>> So, why did you leave?
2:06
>> So, really uh research is a very specific exercise.
2:08
Uh I'm I'm again, as I said, I'm not a trained researcher and so I spent those three years doing kind of the PhD that I never did. really enjoyed it.
2:16
But I at the end I really wanted to go back to building a product.
2:20
When you do research it's really uh you know you're scratching a surface for weeks or months and then you find something interesting gets you like super high super high but it lasts for 5 seconds literally 5 seconds and after 5 seconds you're like oh yeah that was a you keep scratching again.
2:38
So I think it's a it's a I mean it's something interesting and I and I really enjoyed it but wanted to go back to building something more around the product >> and so you decided to build something uh on top of the labs.
2:52
So you started dust can you tell us more about what dust is?
2:57
>> Yeah sure obviously we live in a world where the definition of the company is a moving targets uh because we're operating on a techn technological substrate that moves super fast for the past 20 years. You've known that.
3:07
I've known that the technological substrate was JavaScript and posgress.
3:11
It was kind of a rock solid.
3:13
You could see what you would be building in years and that's not necessarily the case.
3:17
But I think the dust really the initial uh motivation was very simple.
3:21
It's like those models are are ready uh to like really change the way we're going to work and we want to focus on applying LLMs to the workplace which in 2022 or early 2023 when we were raising was almost considered a niche which was fun.
3:39
Uh but yeah that's the main idea.
3:41
Obviously the what itself evolves a lot.
3:46
>> So you spend a lot of time thinking about work and how work is going to evolve.
3:50
So if you're right in how you're thinking about it today, what do you think is going to happen like two to three years from now? [snorts] >> Wow.
3:58
Two to three years feels like uh >> impossible to predict anything at this stage.
4:03
>> Just try to tell us what work looks like two to three years from now. Your best guess.
4:07
>> Yeah, I'm I'm a bit of a Samman.
4:11
I mean, he said that at some point.
4:11
I think I I I I relate to that.
4:12
think I I I I relate to that. I mean if you think about about our great grandparents um walk was mostly physical or great great grandparents whatever is the number of great uh and if they were looking at us today they would be like you're not walking here you're just sitting in a chair looking at a screen
4:33
all day chilling around that's not work and so I think like two to three years that I'm pretty convinced in a and I'm pretty optimistic that we'll still be walking but that work will feel like not work will be looking at us in two three years and two days us would look at us in two or two three years and we'll be like h that's not really work what you're doing here. >> Um so the the interesting thing though
4:53
>> Um so the the interesting thing though is as you said in 2023 when you started it was the very early days of agents but you still had a thesis on how they would impact work.
5:03
Uh when you reflect back where were you right and where were you wrong?
5:08
wrong? So I think we were deeply deeply convinced that walk would be uh incredibly disrupted by the technology and I think that's something where we were deeply right and I think we're starting to be right only now at the at a three years uh time scale it's only the during the past couple months that work has started to change like as a as a dev the the way we work has completely
5:33
changed since November 2025 so it's been just a couple months but the work has like radically changed and I think we're starting to see that in every other functions and so that is that this technology would be extremely disruptive and and structuring for how we walk we were pretty right there even if for three years uh work was kind of the same uh there were a lot of verticaliz
6:00
product on some particular pieces of work there was uh two years ago it was all about clean which was mostly about asking questions and getting answers but the shape of work was not necessarily changing and so that's the part where we were right in a sense is that this is going to reinvent the way we work the part where uh I was uh wrong um I I think I was more of a believer of an earlier plateau in a sense uh even if
6:33
even if that technology had plateaued like a couple months ago we would still have like like decades of deployment uh and and innovation to do for it to be adopted in every pieces of places where we do work but uh but that that plateau is not happening at all and it's still uh if you look at the way uh we work is being redefined uh right now it's it's it's clearly not happening. So I think
6:56
So I think that's maybe that part where I was uh not as optimistic as I should have. >> Yeah.
7:01
I remember you telling me about this plateau and I'm still waiting for it. >> Yeah. Yeah. It's still not coming. >> Yeah.
7:05
So, uh, I want to make sure we talk about the elephant in the room, which is OpenAI and Anthropic are giants.
7:12
They move fast, they innovate, and you're working right alongside like right next to them.
7:18
Um, how do you feel about this? >> Challenged, I guess.
7:23
Uh, no, it's obviously it's obviously challenging.
7:27
Um, we went for an horizontal platform very early on.
7:31
very early on. Uh again for the past couple years everybody was saying verticalize AI verticalize AI and there were great arguments for that is that when you have a verticalized AI product your go to market is much simpler because you know who to target you go talk to them you wash rinse repeats uh
7:46
it's much more obvious when you have an horizontal platform it's super well documented uh we're not the first to do that notion air tableable all of that the go to market is much more finicky because uh it's these platforms are obviously extremely powerful but people are Ah yes but what am I going to use it for? I don't understand blah blah blah.
8:04
I don't understand blah blah blah.
8:05
Walking in the as you said in the shadow shadow is a bit uh harsh for us but uh walking alongside [laughter] >> sorry >> obviously we're tiny tiny spec compared to entropic and open and that's fine.
8:16
Uh walking uh is is both beneficial and challenging.
8:21
challenging. It's beneficial because they uh they educate the markets uh very efficiently and I think as we're seeing today it's all about horizontal platform like it feels like every product is converging towards the same thing which is some sort of uh new uh uh productivity suites uh like Microsoft
8:42
has been doing or Google Drive has been doing but for human agent interactions and the product they used to be like kind of pretty deep product pockets those verticalized AI, but it seems like everything is converging toward much less depth in terms of product and much more coverage in terms of where it integrates and everything. So, it's a
9:00
So, it's a bit like a in the three body problem novel or or or TV show, you see that thing that that kind of goes around the earth, you almost don't see it, but it's super powerful because it it it covers it covers it entirely.
9:14
So, I think we're competing for that.
9:16
There's a lot of convergence in the uh in the product space.
9:19
I think in that in that in in that aspect and there'll be some different sensibilities in the way you build that bubble of products uh which will create differentiation.
9:28
Uh we are going a lot towards collaboration and multiplayer AI.
9:33
We I could spend minutes and minutes about explaining why it makes sense.
9:37
We're also uh model agnostic which I think is a is a very interesting aspect of that and that's something that the labs will not be able to do.
9:44
to do. Uh if you think about buying your product and your tokens uh at the same place, it's like if you were uh building a plant with machines in there and you would buy the machines from the energy provider and the plug would only work with one energy provider and if you if
10:01
your energy provider was a Russian gas would be in a pretty bad place right now and you wouldn't be able to connect to nuclear French nuclear makes no sense and so I think it makes a lot of sense for those product for the pro player to be actually uh uh uh like agnostic in terms of where you bring your intelligence from. >> So um you're model agnostic today and
10:19
>> So um you're model agnostic today and you'll remain this way. >> Yes, definitely.
10:23
Uh I don't see us uh it's funny because we're model agnostic but at the same time uh at any point of stage uh a lot of the usage goes towards one provider but that's the point is that at for given period of time that provider is the best for doing work at another given point of time this other provider is the best for doing work.
10:42
that still changes very dynamically and so we don't see any moment where we would not be model agnostic.
10:49
>> Yeah, that makes sense.
10:49
So when I think about these giants and what they do to the ecosystem of startup um they impact them in many ways.
10:55
One of them is the funding markets.
10:57
Uh, and there is a risk that I think is an actual thing happening which is they pull out the oxygen uh out of the out of the funding market because you're a VC and you'd rather spend 1 billion in one lab than uh spend 10 times 100 million in 10 different companies.
11:16
Um, you've raised in this environment.
11:18
Uh, so you have experience with this.
11:21
How do you think these giants impact the funding market?
11:24
Oh yeah, I think the so our race was uh so we we were lucky to have a very easy seed, very very easy seed.
11:28
We were lucky to have a very very easy series A and the series B was the first time we raced for real.
11:35
It was complicated and hard as it should be.
11:37
Uh and uh and I think part of that is because we're building from France in a in a moment where it's not it's less from the American perspective slightly less fashionable than in 22 to 2023.
11:48
Uh the other part is that I think a lot of the energy in the market is being absorbed by the labs.
11:53
Indeed as you said uh if I was a gross investors that would make total sense for me to put 1 billion in entropic get two billion in six months and be done with that.
12:04
So uh uh hopefully those companies will become public which will put them in a kind of a no-go zone for VCs which means that hopefully and they'll bring liquidity to the market.
12:16
So hopefully they'll they'll get better.
12:17
So I want to go back to the being in the shadow of French labs but before because you talked about fundraising I want to talk about this because one of the things that when I talk about dust one of the things I say is uh I think Sten and Gabrielle his co-founder are some of the most thoughtful founders I've seen specifically on company building and specifically on raising money.
12:38
You could have coming from OpenAI Stripe uh being a second time founder you could have raised at a crazy valuation and yet you decided to raise at reasonable valuations for your seed and then again for your series A.
12:52
How did you make this decision?
12:55
>> I think it felt very natural to us.
12:55
I've seen some uh some friends gets a little bit burned in the 20 20 21 years raising a little bit too much.
13:02
And I think there is a true coffin corner for companies if you raise too much and do not kind of realize the potential of the money you raised, you end up in a coffin corner where the only outcome is a done run and downturn, sorry.
13:14
And there's a I mean I I've never done that, but I think a downturn is really really bad um uh in many aspects.
13:22
And so we knew from the get-go that we wanted to focus at the product layer.
13:26
We had a mantra that is still almost true. No GPU before PMF.
13:28
We raised our seed at a moment where uh it was fashionable to raise money for training.
13:36
training. So everybody was running 100 million here, 200 million there and we raised five five millions which seems like silly but at the same time we knew that we building on the at the product layer we would not have been able to use
13:50
that money and it would have made no sense and we much rather raise less uh raise as a smaller valuation but be very aggressive into building into that and and creating the momentum and raising and raising another run when we're ready etc. So um that's that's the positive
14:03
So um that's that's the positive view of that.
14:07
The negative view of that is that we're being a little bit too French.
14:12
Uh and I think you want to strike the right tradeoff if you look at I mean I'm always impressed by mistrol in that perspective.
14:20
in that perspective. I've seen uh so uh I've met Arto Mench uh because I knew Giaml very well when he decided to go for building a company and he's extremely smart but he knew nothing about building a company like nothing he was just a researcher no clue no absolute no clue and the way he grew
14:41
into that uh kind of role of CEO of European lab with all its imperfection uh the fact that they managed to build a company that is valued 12 billion is extremely uh recommendable is extremely impressive and there's there's in that there's a bit of let's uh let's try to kick the ball very aggressively. So I
14:58
So I think it's a fine line to to to to consider uh you need to both aspects are important and you need to find your your it's a gray zone basically. >> Yeah. Yeah, that makes sense.
15:11
you talked about uh your series be uh maybe more challenging because you're building in France and it's true like I knew you from your San Francisco days.
15:20
So you knew what San Francisco was like.
15:22
You built there, you hired there and yet you decided to come back to France to build the company.
15:27
So like how did you make this decision and what are your thoughts today on having made this decision?
15:34
uh everything would have been easier from a company building standpoint uh going in the US uh from the get-go.
15:38
I think uh uh but it doesn't mean it being easier doesn't mean that it's the right thing to do.
15:46
And I think we were really I mean I'm I I was and Gabriel as well.
15:51
We really wanted to build something back in France to uh because it was important to us.
15:54
There's a there's an aspect like national preference or sovereignty to it.
15:59
Uh that's an additional constraint.
16:02
So that's uh that's one that you have to be very uh very aware of.
16:03
But I remain convinced that it's when things are going super well that's no problem.
16:09
I mean if you look at uh the lovable of the world the uh Spotify of the world everybody says SF SF SFF the funders needs to be an SF and the same VCs they [snorts] don't give a about the localization of the lovable funders.
16:25
They could be in whatever country when the things are working well the things are working well nobody cares.
16:29
So, uh, it's an additional friction that you take in for the a greater good maybe, but it's a trade-off.
16:37
>> Yeah, that makes sense.
16:37
Um, I want to talk about the the labs.
16:40
There is one specific question I want to ask that may be useful for the uh the audience, which is we've seen the labs uh going after uh new verticals.
16:52
So example of this is Enthropic uh going after Legora in the um law space.
16:58
Um so if you're a founder and that happens to you, how do you how should you think about it? >> Yes.
17:08
So so being verticalized great advantage in terms of go to market being verticalized made a ton of sense because the models were not that good and so you had to build a little bit of scaffolding around that to make them useful.
17:17
that second value just disappears as the models are getting better.
17:23
models are getting better. So I think the uh that doesn't mean that it it's not a good idea to be verticalized but I think almost we we're almost in a world where intelligence is commoditized in a way it's like um the old good SAS and and and internet days it it feels to me
17:40
like the value will bring uh that verticaliz product will have is if they manage to create a different type of mode which is probably the whatever network effect you can find in your vertical I think that will be the because that's the thing that is defensible even in place of perfect intelligence coming in. So if you're
17:58
So if you're building a virtualized product uh in a given industry, I do think that the we're going back to the advices that we used to receive in 2010 almost which is network effects. Find a network effects.
18:11
Find a thing that is defensible because you have a you have value coming from having multiple users interacting with the same platform.
18:17
If you don't have that, if it's just you're delivering intelligence to one and intelligence to other and there is no network effect, I think in that case the defensibility is changing. >> Yeah.
18:26
H how do you think about margins?
18:30
>> Uh margins are tough in those days. Tough very tough.
18:33
Uh >> so and maybe the question is so you know when you're a lab you're uh capturing the margin at the token level and then you build a product and you don't usually add any margin on top of this and so then you're pressured when you've already been taken some of this margin.
18:51
So for you how do you think about charging and what feels like the right way to think about it? Yeah.
18:55
So we used to have a seatbased pricing which made sense for three years which has become because we wanted to favor usage and value creation.
19:03
So it was a blanket price.
19:05
You pay to use as much as you want.
19:07
Obviously over the past six months the usage has exploded.
19:10
The agent loops are getting longer.
19:12
Uh models are just using more tokens all over the place and so our margins have compressed and we are in the middle of transitioning to a cratebased pricing.
19:22
I think that today if you are building a at the product layer there is no way around credit based pricing uh because uh people want to use more AI and they they'll just max use as much as you give them and if you have a a flat price that just doesn't fly.
19:40
So uh that is for kind of how to think about pricing uh in terms of a current product in AI.
19:46
I think it's the only way to maintain your margins because if you have a flat price, the margins get compressed and they'll keep getting compressing at least under the current conditions of models getting better, people are using more and more.
19:58
Uh and then the question about competition with the labs I mean there are precedents where uh companies manage to sell a product on top of an infrastructure that captures the margins and resell the same product at zero price.
20:10
many example box one passport.
20:10
I think there is obviously value in creating a very well thoughtout, very wellcraftrafted, very well differentiated product and that will enable to capture a fair amount of margin still.
20:23
And finally the third piece is that hopefully we have the uh the open source is catching up to some extent and this this will create some pressure on the labs on the margins.
20:33
pressure on the labs on the margins. uh if you think about it uh people I mean when you look at the numbers and you can look at the S1 from SpaceX as well uh the labs are margins are are humongous they're humongous so I'll try to hold it in 10 seconds but uh if you if you think
20:54
about the cost of a model you can look at the latency the latency is a measure of the cost of the model because uh it's for a given model infrastructure the latency gives you the size of the model the energy it consumes etc and if to think about the latency of cloud code versus GLM 5.2. It's roughly the same. 2. It's roughly the same.
21:10
It's the maximum latency acceptable to run in cloud code.
21:12
And so you can compare the margin, you can compare the price of GLM 5.
21:15
2 on base 10 or whatever fireworks to the price you pay for an equivalent frontier model with same latency.
21:22
And the difference is like almost 9x.
21:24
So that means that there is very good chance that those labs are margin kind of creating margin of something like 70 or 80% uh serving us those frontier models.
21:34
So hopefully with open source coming up it'll recalibrate a little bit the market and those margin will go down and you'll have a better chance to compete even if they capture still capture a lot of margin. >> That makes sense. We're at time.
21:45
I'll still ask you one last question because as I said you're very wise.
21:48
So what is there any last word of wisdom that you want to share with this crowd?
21:53
describe being uh entrepreneurs or people that want to be entrepreneurs are young in in their career.
21:59
>> Yeah, I think the uh the only thing that really matters at the end is uh uh uh find the find the the the I mean it's it's going to be sounds so typical but I think it's really true.
22:09
Find the the thing you want to fix the vision you have because building company is shitty on a daily basis like shitty as hell.
22:17
there's some very highs and a lot of lows and so if you don't have that thing that you can really wake up in the morning and say but this is the reason reason why I'm doing it uh it's it's not sustainable so that's the first thing if you don't have that in your mind find it
22:32
today it's fine to reinvent it while you build it's okay but find that thing that will keep you uh uh driven towards an objective despite all the glass uh eating that is involved in the job >> yeah I agree thank you so much and please uh thank him as well. It was
22:47
It was amazing to have you. >> [applause]