- il y a 16 heures
Founder’s Story Building Teams for the GenAI Era
Catégorie
🤖
TechnologieTranscription
00:01Bonjour, tout le monde.
00:03Bonjour Paris, c'est génial d'être ici.
00:05Je m'appelle Julien Beck, je suis un partenaire à Sequoia Capital.
00:10Aujourd'hui, nous sommes speaking avec Gabriel Hubert, le founder de Dust.
00:14Nous avons été heureux d'être un partenaire avec Dust,
00:18maintenant, un mois depuis le Seed round.
00:21Et c'est juste incroyable de voir comment vite vous avez effectué.
00:26En juste un mois, ils ont obtenu 20 clients en France.
00:30Nous avons déjà entendu que nous avons des clients en France aussi.
00:34Donc, nous sommes vraiment heureux de l'exécution.
00:36Et aussi, je suis très heureux, comme un français,
00:40de montrer mes collègues que, en France, nous ne faisons pas juste des strikes,
00:44nous vendons produits.
00:45Donc, Gabriel, nous sommes retournés à votre début,
00:50à l'entrepreneur et à la foundation de Dust.
00:54Oui, merci beaucoup pour me, Julien.
00:56Bonjour, tout le monde.
00:57Je vais juste demander de l'audience pour un grand show de hands,
00:59de comprendre ce niveau nous devons avoir la conversation.
01:02Si vous êtes un peu de founder ou de start-up,
01:05raisez vos handes.
01:07Si vous êtes un peu d'investissement,
01:09raisez vos handes.
01:10Et si vous travaillez pour un grand company,
01:13et vous êtes intéressé à Gen-AI,
01:14raisez vos handes.
01:16Potential customers, great.
01:19So, I was lucky enough to finish my studies in California at Stanford,
01:24and I think it's the first time where I got to see that the companies that were being started
01:29in the U.S. that everybody was hearing about were companies that were actually started by humans,
01:34that weren't perfect, that also made mistakes.
01:37Not by the eyes.
01:38And so I think for France at the time, this is way back when,
01:45it's going to date me, but this is 20, sort of closer to 20 than to 15 years ago.
01:50It was the first time that I could see that people starting companies were just real people.
01:54And that made it a lot easier to consider or project myself in there.
01:59And then quickly after that, my then friend, now serial business partner, and I decided to give it a go.
02:06Great, very exciting.
02:08I remember you were very opinionated at the beginning of Dust about how you wanted to build a company.
02:15So tell us about those unique insights that were the foundation of Dust,
02:19and tell us maybe a little bit more about what Dust does today.
02:23Yeah, so before starting the company, having had our first business acquired
02:29by Stripe, joining it when it was about 100 people and seeing it grow to 3,000 people in under
02:35five years.
02:36And then for my co-founder to work at OpenAI as a research engineer,
02:40and for myself as a product lead at Allen, a healthcare insurance business in Europe.
02:45I think the insights were twofold.
02:48One, it was clear that the product layer, the application layer needed a lot of work and a lot of
02:54attention.
02:55And this is, you know, 2021, early 2022, before ChatGPT was all over the place.
03:02And we really thought that investing in that early could yield large dividends.
03:06And two, I think the other insight or level of conviction we had in this specific area of technology
03:12was that if you trust people, a lot more change can happen rapidly.
03:16I think that this technology is very empowering, and companies that let their employees use it,
03:23discover use cases, and deploy use cases quickly have an advantage over others.
03:28So Dust allows fast-moving teams to deploy assistance on their internal workflows quickly, securely,
03:37with a very bottoms-up approach.
03:39Anybody in any team can access the context of the company, continuously updated by us,
03:45and use the best models, whether they come from Mistral or OpenAI, Anthropic, et cetera, et cetera.
03:51And it's the combination of those capabilities that really allows to unlock concrete workflows quite quickly,
03:58which we've been pretty happy and successful with.
04:01Yeah, no, thank you for the explanation.
04:03I think the thing I find most powerful about Dust is that it really helps automate the 50% of
04:10the tasks
04:10I really hate doing in the office.
04:12I think we can all think of a few examples, and it's really become a superpower for a lot of
04:17the users at Sequoia.
04:18So for the other investors, don't listen. We need to remain the most productive firm.
04:24But just going back a little bit, you're, you know, French-born.
04:28You joined a scale-up, but in the early days in the U.S.
04:31You came back to France to join Allen, but then when you decide to found the company,
04:37unlike many founders who just go to the U.S., you decide to do it in France and in Paris
04:42in particular.
04:42Tell us why.
04:44I think the easy, obvious answer is that that's where my co-founder and I were living.
04:49So we had to ask ourselves what would justify moving over.
04:53We were lucky enough to have among our investors very successful founders that were many steps ahead of us in
04:59their journey,
04:59and to ask them the question.
05:01So Olivier Pommel, the CEO and co-founder of Datadog, started the company in New York because he was living
05:07in New York.
05:08And when we asked him, like, should we start in the U.S. or should we start in France?
05:11He's like, you should start where you are, because every step you take that's removed from that is just wasting
05:16time
05:16and adding complexity to your life and potentially your business.
05:20But he did add, listen, I wouldn't give the same advice if it was 2010.
05:24I think I'd consider maybe moving to the U.S. a little more.
05:28But the level of ambition and the quality of the operating teams you'll be able to find in France now
05:33is quite different from what we had 10, 15 years ago.
05:37And I think that added to the idea that, you know, having been lucky to experience what we did in
05:43the U.S.,
05:44we can contribute more actively to the ecosystem here.
05:48And Europe is only going to be in the race if people actively contribute to keeping it in the race.
05:54It was an opportunity to see it as relatively straightforward, not a huge handicap.
05:59I think we can still grow commercial and a number of distribution efforts abroad when we need to.
06:05And then a great place for talent with a very different point of view on ambition.
06:10I think that the number of people who've come back from five years at Airbnb, five years at Uber,
06:16who've seen intimately how the sausage is made at these very successful companies
06:22can now contribute actively to this new wave and generation of, I think, ambitious startups
06:29and hopefully soon scale-ups.
06:31Yeah, no, that's very true.
06:32I think I want to unpack two things you mentioned.
06:34One is the access to talent in France and in Europe.
06:38And the second is the skill of ambition.
06:41What we ran a study last year at Sequoia, a talent team mapped out engineering talents globally,
06:46and it found that AI talent in particular has the highest density in Europe of all places in the world.
06:53And Paris and London are two of the most dense nodes.
06:56So I think that really answers to as kind of tangible data to what you were describing.
07:02How have you experienced that?
07:03That would be my first question.
07:05And the second is just to unpack the scale of ambition.
07:08Do you find that has radically changed with founders and with employees in particular?
07:14So on the talent front, absolutely.
07:16And I think you can see it.
07:18The next report I would look at is just salary inflation for early engineering teams in startups in Paris.
07:26People can see it as bad news if they're paying the salaries.
07:29We tend to see it as good news because that means that we're actually in the game for top-tier
07:34talent
07:34and that these people who have obviously a wide variety of options available to them
07:42are picking French-based and in some cases French-based and French-headquartered companies,
07:47which I think is great.
07:48To the level of ambition, it's really quite different.
07:53You know, two and a half years ago, Stan and I were in a restaurant with three researchers
08:01working at Meta and DeepMind who wanted to start a company
08:04and who were thinking about ways to potentially start in France.
08:09And they were Timothée, Guillaume and Arthur and they started Mistral
08:14and the rest is sort of slowly but surely becoming history or I don't even know if it's slowly.
08:19But the way they were understanding how they'd like to grow the company back then,
08:23despite being from a different generation, like quite a few years younger than I am now,
08:28was a great testament to their early vision and ambition setting.
08:34And I think we're seeing that in a number of companies.
08:36You have invested in, Sequoia Capital has invested in a few in France and, you know,
08:4110, 12 years ago, I don't think that was the case.
08:43I don't know that you were looking at the French market in that way.
08:45And that these companies have been able to attract American firms and their American partnerships,
08:51in some cases, to invest in their companies is, I think, proof in the pudding.
08:55Yeah, and I couldn't agree more. And I wasn't there 15 years ago, but I can tell you that nine
08:59years ago, we were looking at the rocks.
09:01You were, hey, you were maybe not investing yet.
09:03Exactly. But that's a really interesting point.
09:06So now we're basically saying that Europe has both the access to talent, but also the scale of ambition.
09:12You've set up initially an in-person team, right?
09:16So walk us through the logic of doing that as opposed to a distributed organization and walk us through also
09:23how you think about scaling that
09:25as you become more successful as a business and need to grow your team.
09:29The only thing you have when you're starting is velocity. It's the only thing you have.
09:33And I think when you try to operate as more than one human, what you need is velocity and trust.
09:39So for Stan and I, it was really the question, what is going to build and maintain velocity and what
09:45is going to accelerate the ability to really trust each other?
09:48I just don't think that we're genetically wired to very easily trust people that we do not know the physical
09:55height of.
09:56And so when you've been working with people for six months and don't know how to spot them in a
09:59cafe,
10:00I think there's something subconscious that makes it harder to be as vulnerable, as direct.
10:05And in some cases, to take the risks you need to take, to take the shortcuts you need to take,
10:10knowing that people will trust your intent.
10:14I started working two days after lockdown at Allen in 2020.
10:19And the fact that many teams were growing quickly without ever having the opportunity to physically connect
10:26was clearly an impediment to building the level of informal trust that is going to keep you moving fast.
10:32So right now, we're, you know, sub-20 people.
10:36We don't need to be thinking about hiring in other places because it's limiting our talent pool.
10:42If we had 500 engineers and we were looking to double that in the next 18 months, I think we'd
10:47have a different conversation.
10:48I don't know that we would have the remote conversation.
10:51It's not something that we're particularly excited by, but I think, you know, opening up different hubs would obviously be
10:57very, you know, very credible.
10:59I'll say a lot of the founders that we meet as customers now, some of them only started their companies
11:04a few years ago, but started them in the pandemic when, you know, you started a remote company or you
11:08didn't start a company at all,
11:10have been telling us that they envy the ability to have started off with an in-person, in-office culture
11:17just for that aspect of velocity and trust.
11:20And we don't have a, you have to be here from Monday to Friday policy.
11:24I think FaceTime is just the worst cultural rule, but the flexibility you need, you take once you've built the
11:31trust with the rest of the team and it's going to be the summer Olympics soon.
11:35I don't expect many people to make it to the office every day because we're in the center of Paris
11:39and it's actually like pretty hard to get about for some of the events, but that's fine.
11:44I think we've built some of the initial trust that helps us move fast.
11:48Yeah.
11:48And I'm sure that when people see you in person, they're always surprised by how tall you are, right?
11:53Well, I actually, Ed, our lead designer is incredibly tall and people, when they meet, it was like, ah, I
11:59get the designer confidence now.
12:02Interesting.
12:03Um, and how do you think about the culture and creating a winning culture as you scale your organization?
12:09What are some of the principles that you want us to have that you, you want the company to have
12:14as it scales?
12:16We sort of wrote them down very straightforward.
12:19The one is we will build AGI and that's just trying to set the stage for ambition.
12:24I think that we believe the product layer will actually be the one that stays when the competition around the
12:30models has made it easy to switch them in and out.
12:33And so by owning that experience with users, we want to be the place where the innovation starts in deciding
12:39what we determine as general purpose, um, AI.
12:42The second one, uh, is, um, left lane blinking headlights.
12:47So in, in, in French, which kind of gives an idea of how fast we want people to move.
12:53You know, there is no try, just do, don't ask for permission and bend the arc as some of the
12:59ways in which we describe that.
13:00Um, and then the, the last one is, um, A team.
13:04Uh, and so by that we try and describe that pushing each other is loving each other.
13:09Um, we also, you know, continue to hire the best and, uh, the last one we say is that we're
13:14not saving babies.
13:15We can do serious things without taking ourselves too seriously.
13:19And I think keeping some distance and humility with what you're doing is also a way to build a culture
13:23that, um, yeah, doesn't drink the Kool-Aid too much.
13:27Right.
13:28We love that.
13:29Um, and tell us a lot of, there's a lot of hype, obviously with AI,
13:32but also you guys have been extremely fast at making revenue.
13:36So clearly we're moving not just from, to demos, but to production.
13:41Uh, tell us a little bit about some of the things that you're excited about that you're seeing in production
13:45today,
13:45whether it's at Dust or other companies that you've observed.
13:48Yeah.
13:49Well, one of the great things about building a platform,
13:51there are many bad and terrible things about building a platform.
13:54It's just hard at the beginning.
13:55You don't know if you're building the right abstractions.
13:57You'll only know later on.
13:58But one of the great things is that your users help you understand
14:01where they see the value in much more diverse and unpredictable ways than you could have imagined.
14:06Uh, so a couple of examples, you know, I think I'm not going to list all the use cases that
14:10people are finding across companies,
14:12but the ones that have surprised me have been, you know, marketing teams extracting information from the code base
14:18so that they can understand when to write blog posts without bothering anybody from the engineering team.
14:23That was just a concrete way in which they were being slowed down that has been disintermediated
14:28and that I think is really the shape of what this technology can do.
14:31It's helping internal translation happen very, very seamlessly.
14:35At Allen, personal coaching has emerged where people will design their own assistant to help them work on their projects
14:43and in light of the leadership principles that the company has, give them comments and suggestions
14:48on how they can set their goals or improve their contributions to various tasks.
14:52Um, I think the exploration phase is also just accelerators.
14:57One of the other ones that I think is, uh, is, is funny, uh, Conto.
15:01Conto, um, they've got a very, very small team looking at the German market
15:05and apparently, and I don't know if there are any Germans in the room,
15:07but, uh, company onboarding is quite complex in Germany
15:10and the paperwork involved is pretty, pretty hard to read
15:13and that's if you can read German, which some of the members of the Conto team for Germany can't do.
15:18So, using assistants to automatically translate and extract information
15:22that they need to manage compliance at the time of onboarding
15:24is one example where they were able to get it going in 20 minutes,
15:27test it in two weeks, and it's now one of the most used for that geography.
15:31So, the pace at which the bottoms-up ideation and iteration can happen
15:36is, I think, what's really exciting.
15:38And I found that use case super interesting
15:40because you're really enabling something that we couldn't do before at all, right?
15:45So, if you fast-forward a couple of steps,
15:47we're not just empowering employees,
15:49but we're potentially replacing entire functions.
15:53How do you think about that and the role you want to play in that transition
15:56from, you know, we call it services as a software at Sequoia?
16:03Let's see how it plays out.
16:04You know, I think it does...
16:06Our product manifesto has a comment that's like
16:08augmenting humans, not replacing them,
16:10which we also say like R2D2, not Skynet.
16:13And the idea behind that is really that I think smart, empowered people
16:18are very comfortable getting rid of the repetitive tasks in their day-to-day,
16:22and they want exoskeletons to do the menial stuff faster
16:26and focus on the interesting stuff.
16:29Working for younger companies, which is currently our situation,
16:33allows us to also just plan for the future differently
16:35rather than get into conversations about replacing or removing entire functions.
16:40So, we're not here to automate your customer service,
16:44but at PayFit or at Penny Lane or at Conto,
16:46the customer service teams are reinventing how they do escalations,
16:49how they do ticket investigation, and how they do training.
16:52Thanks to DUS, and I think the feedback we've gotten so far
16:54is that they feel a lot less threatened
16:56now that they see how it can help them day-to-day.
17:00Interesting.
17:01And maybe a final thought.
17:02Now, at Sequoia, we often think of technological shifts
17:05as being overestimated in terms of impact in the short run,
17:09but completely underestimated in the long run.
17:12I think that was the case for the internet,
17:13and we definitely think it's true in the case of AI.
17:17What are some of the things that you think are overhyped today
17:20and underhyped tomorrow?
17:23I think the immediate replacement of entire teams is overhyped.
17:26I think people are not anticipating some of the risks,
17:30including the risk of fatigue that their customers might have
17:33in understanding that they're being dealt with by an automated system.
17:36We've all been mad at customer service,
17:38and I think that there's something people are looking for
17:41when they want to trust a brand and repeat doing business with it
17:44that is not just, oh, you found an automated
17:46and a cheaper way to deal with my question.
17:48So, I think that's overhyped in the short term.
17:50I do think in the long term,
17:52many people will be able to interface with products
17:54much more autonomously than we think.
17:57Multimodal input, voice input,
17:58all these things are going to change experiences quite fundamentally.
18:02And then by and large, I think that what's underhyped
18:05is just how selective employees or candidates
18:08are going to be with their potential employers
18:11when it comes to their tech stack.
18:13Apart from developers,
18:14I don't think many of us are asking our future employers,
18:16what do you have for internal messaging?
18:18What do you have for data analysis?
18:19But I think in five years' time,
18:21graduates will know to ask a potential employer,
18:23how do I understand the business priorities?
18:25How do I get updated in real time on my constraints?
18:29How do I interface with other divisions?
18:31How do I understand what the data I can plow into
18:33is telling me without knowing SQL?
18:36And if you're telling me,
18:37oh, just ask the data team,
18:38or ask the marketing team,
18:39or ask the engineering team like you were doing in 2015,
18:42I think you're going to be compared to people saying,
18:45oh, to push code in production,
18:46you write it on a piece of paper,
18:48your engineering manager will correct it with a red felt tip,
18:52and then you can push it in the mainframe with punch cards
18:54on Saturday morning twice a week.
18:57So that, I think, is underhyped.
18:58The way in which end-to-end workflows
19:01will be increasingly part of our day,
19:03and if you're working at a 500-person company,
19:05maybe 200 of those aren't really human.
19:07But they're still doing the job of a 500-person company easily.
19:11Great.
19:12Well, Gabriel, thanks so much
19:13for raising the scale of ambition in France
19:15and for making all of us more productive.
19:18Speak soon.
19:19Thanks.
19:19Thanks so much for having me.
19:20Thank you, everybody.
19:21Thank you, everyone.
Commentaires