- il y a 1 jour
The AI Toolbox: Shaping the Future of Work & Human Potential
Catégorie
🤖
TechnologieTranscription
00:00Bonjour, tout le monde. J'espère que vous avez apprécié votre premier jour à VivaTech.
00:06Je suis très heureux de vous être ici aujourd'hui. Je suis Julie, je suis le co-founder et CEO
00:11de Polen,
00:12une professionnalité de formation spéciale sur l'AI en un côté et sur le leadership en l'autre côté,
00:18avec des professionnels expert-led.
00:21Et nous sommes ici aujourd'hui pour discuter comment l'AI est changement de façon que nous travaillons,
00:27concretement, comment est-ce que les organisations sont adaptées,
00:32comment construirent les nouvelles capacités sans perdre de l'expérience humaine,
00:37et qu'est-ce qu'il prend pour faire l'AI partie du travail d'aujourd'hui,
00:43pas seulement une headline ou un projet de pilote.
00:46Pour explorer ce sujet, je suis très heureux d'être ici avec deux leaders au cœur de la transformation,
00:53Mathieu Birac, à commencer parmi, à mon côté,
00:57chief people officer at Dr. Lib,
01:00et Stanislas Polu, co-founder of Dust.
01:04Nous regardons ensemble à comment l'AI est reshaping le travail quotidien,
01:08de construire des assistants de Dust, qui nous expliquera comment tout ça fonctionne,
01:14à l'équipe de l'équipe de l'équipe de l'équipe de l'équipe de l'équipe de l
01:17'équipe de l'équipe.
01:18La semaine d'entre eux, nous avons réalisé étudiant avec Pauline et Edflex,
01:23et un test qui est Notif, c'est que aujourd'hui 79% des professionnelsuls選nent le travail de l'AI,
01:31Mais seulement 10% des entreprises se sentent vraiment préparées et prêts pour cette transformation.
01:39Donc nous voulons aujourd'hui comprendre d'où vient cette frappe et comment nous pouvons améliorer cette frappe.
01:47Nous allons aussi regarder ce qui signifie l'intégration de l'intégration responsable de l'AI,
01:53et comment faire assurer que le potentiel humain reste à cœur.
01:58Mathieu, allons-y.
01:59Vous avez été à Doctolib depuis 9 ans.
02:03Quand vous avez participé à Doctolib, nous étions juste discuté.
02:05Vous étiez seulement 200 personnes.
02:07C'était une petite, pas assez petite à 200, mais encore une start-up française.
02:12Maintenant, vous êtes près de 3 000, opérons à un niveau global.
02:18Vous diriez, pendant ces 9 ans que vous avez expérimenté à Doctolib,
02:22que l'AI transformation que nous avons expérimenté est la plus grande transformation que vous avez vu.
02:29Et que vous avez vu beaucoup.
02:31Oui.
02:32Bien sûr, je dois dire oui à cette question.
02:33Otherwise, we can't close the panel.
02:35So, yes, of course.
02:36Yes, please, it's the first question.
02:37No, no, for sure.
02:38Let's start well.
02:39Well, AI is, of course, the biggest transformation we've seen at Doctolib,
02:42and I guess it's common to everyone in the room.
02:44I think, and it was said before in a couple of panels, the fundamental change is it really reinvents how
02:52we work.
02:53And I think it's the last coffin on the nail of what we used to call knowledge work.
02:57And now we have to switch to something new, which is, I don't know, wisdom work or whatever we call
03:02it.
03:02And so, you know, it asks us a lot of questions.
03:05So, it's kind of a down thing, everything that we have to figure out.
03:08But I feel it's also exciting.
03:10I think it's been a long time since we haven't had the chance collectively to really think about what is
03:16work.
03:17And internally at Doctolib, the way we pitch AI and DUST, which we are working with,
03:23is that, thanks to that, you could claim back 20% of your time, or even a full day, right,
03:29in the week.
03:30And now the big question is, what do you do with the day that you have just claimed back?
03:35Stan, so I mentioned you're the co-founder of DUST, Agentic AI.
03:39Can you explain us in simple words, what is Agentic AI, and what does DUST do concretely?
03:47Yeah, I mean, Agentic AI, I guess the definition, I mean, boils down to the definition of what is an
03:52agent.
03:52An agent is basically a program where some decisions are taken by an artificial intelligence.
03:59that's going to be an LLM in the current technological wave.
04:03And so that means that you've got programs that are going to be able to adapt to the situation
04:08and take actions based on what is required of them.
04:11And so they're going to be able to take many more tasks than what programs would have been able to
04:16take before.
04:16And so at DUST, we're trying to build the agent operating system for companies, letting companies bring them their knowledge,
04:25bring them the tools that those agents can take on the company, and create agents and operate them on the
04:33product.
04:33With the main use cases, I think where we really differ from many other platforms,
04:37that we really let everybody in the company create their own agents and really try to incentivize kind of an
04:43organic emergence of use cases across the organization.
04:47Mathieu, you mentioned that targeting 20% of productivity gain is kind of the North Star deploying AI.
04:56Can you share with us one of two concrete use cases that you're the most proud of?
05:01Because it has so far has the largest impact. And how do you measure this impact?
05:05Yeah, sure. So again, the idea is we are not expecting people to work more with AI.
05:10We are expecting them to work differently.
05:12And so we looked at most of the time consuming low value tasks.
05:16And we try to find, you know, kind of quick wins so that people could embrace AI.
05:20So one of the things that is working really well for us is a tool which we call Cortex.
05:26You know, we are in healthcare, so we have to use that kind of nickname for our tool.
05:32So Cortex is basically our own internal wiki, Wikipedia, right?
05:36Where there's the entire knowledge of the company.
05:38And in the past, we had a nice website where people could actually kind of browse through the website.
05:43And we kind of transform it into a tool which is both a search tool, which you can, like conversational
05:50search.
05:51You can ask questions. And partly agentic, where you could ask, how do I, you know, how can I ask
05:58for a time off?
05:59For example, you know, holidays. You'll have the, you know, the instructions in terms of how to do it.
06:05And then you can also ask the tool to do it for you, right?
06:07So this is between one hour to two hours saved per week for all of our employees.
06:13We have another tool, which is our sales coach tool.
06:16But basically, our entire sales team can ask questions to prepare a meeting, an appointment or whatever.
06:23Same, we have like dozens of messages to that tool every hour at Dr. Leap.
06:29It's probably one of the most used ones.
06:31And we do have a couple of other tools.
06:34And the key for that was to find tools that are super easy to understand and that integrate well in
06:40the workflow of our employees.
06:44Stan, you've helped a lot of companies deploy AI through agents over the past years.
06:52What has been the biggest challenge that you've experienced in getting AI adoption by teams and by companies?
06:59I think the challenge that evolved if you go back, let's say 18 months ago, people would not have adapted
07:07to the idea that those agents are stochastic.
07:10So stochastic, sorry, it's a big scary world, but stochastic means random.
07:15It means that they're going to have a behavior and each time you call them, they're going to have a
07:18slightly different behavior.
07:20And so sometimes like humans do, they do mistakes.
07:22They sometimes hallucinate a little bit.
07:25And 18 months ago, the reaction was always, oh, I've tried it.
07:30It just failed once.
07:32It's no good.
07:32It's no good.
07:33And I think most of the companies and the people we interact with have really, we've seen that adaptation over
07:40the past 18 months,
07:41realizing that they do sometimes make mistakes.
07:43You do have to verify what the agent is doing, but you see the value even if there are some
07:48mistakes that using them will like really accelerate you a lot.
07:53And people have really adapted to that.
07:54I think the second biggest challenge is around the nature of the organization.
08:01Our products requires kind of to have a slightly, I mean, kind of an open minded in terms of what
08:08people will be able to do for your organization.
08:10You give a lot of powers to all of your employees.
08:13And so we've seen that with companies like Dr. Deep that are tech first, forward looking, there's really an adoption
08:21that feels natural because they have what it takes organizationally to reorganize themselves around that new technology.
08:25But if you talk to larger, maybe slightly more older companies that are maybe not less techie, we've seen a
08:33lot of difficulty seeing them adopt the product.
08:36They kind of think in terms of very strict use cases that they want to roll out and kind of
08:41don't see the value of unleashing the cracking of agents on their whole organization yet.
08:46Just yet.
08:48To come.
08:50Mathieu, we know that the AI transformation only works if people start using AI and start experimenting and start iterating
08:58with AI.
09:00Training is key in that matter.
09:0287% of companies make it a priority, but only half of employees have been trained so far.
09:10So there's still a big gap.
09:11How do you address this at Dr. Lib?
09:14And how do you make training at the heart of this AI transformation?
09:18Yeah, it's a good question.
09:19And I don't know that we have cracked it fully because today at Dr. Lib, our employees, 100% of
09:24our employees, they use AI on a weekly slash monthly basis.
09:28But when you look at really poor active users, so people using AI multiple times a day, we are only
09:34at 25 to 30%.
09:36So it's not like we have cracked the topic.
09:39But I think we've tried different things.
09:41We've tried the kind of classic learning approach where you sit people down like that and they listen to some
09:46people talk.
09:47That's a good kind of eye-opener.
09:49But what has worked the best for us is to actually allow people to test.
09:54Because I think there's a blocker, you know, when it's the first time to interact with a blank page.
09:59You have this kind of weirdness where it feels weird to talk to a computer.
10:03And so we put people in a room and we did hackathons where basically they could work and build solutions.
10:09And I think that was the most drastic solution in terms of adoption.
10:14So in my team, for example, we did hackathons a couple of weeks back and we've seen a tremendous kind
10:19of jump between, you know, before and after.
10:22And before we were kind of average in the company in terms of usage.
10:26And after that hackathons, if I look at the numbers of users in my team compared to the rest of
10:32the company, now we are by far one of the first team.
10:35So the most useful thing was to actually, you know, learn by doing in the end.
10:40And make it job specific to see concrete examples in your day-to-day job rather than generalist.
10:48And I think that's also the way you approach training.
10:51We both know that, you both know, that beyond productivity, AI is changing the way we take decisions in a
10:58company.
10:59Whether it comes to hiring decisions, product decisions, strategic decisions.
11:05How do you make sure that human judgment stays part of the process?
11:10Maybe Stan?
11:11Yeah, I don't think we, I mean, it probably changes the way we make decisions, but there's no,
11:16I don't have yet an example of agents actually making very important decisions other than what's next to do to
11:24achieve the task.
11:26But the way it changes decisions is that it accelerates decisions a lot.
11:29There's an example at this, we obviously, big users of our own product was,
11:34there's a project that we wanted to think about and we just unleashed an agent on the problem.
11:40The agent went through every transcript of every course we had, went on every support email that we had about
11:47the subject,
11:48and kind of created a summary with quantitative information of who's asking for that, for which reason,
11:54what is the value of that whatever project we were working on.
11:58And I think that work here would have probably taken someone probably a week or two.
12:04In a startup like ours, we would probably not have done it at all.
12:07And we would have kind of guesstimated what we should do here.
12:10And so, there you see the, not only it accelerates the capability to take decisions,
12:15but I think it probably helps when you are, when you have the right setup to take probably better decisions
12:20as well.
12:20Patio, you want to add something?
12:22Yeah, I think it's super interesting to pick.
12:23You know, at Doctolib what we do is we build solutions and softwares for patients and healthcare professionals, right?
12:28So, we, and we truly believe that technology is a great way to help, for example, doctors to make better
12:34decisions, right?
12:35Because in healthcare, a lot of what you have to do prior to make a decision is to collect information
12:40and to treat that information
12:42and then to pick the right treatment, the right diagnostic, right?
12:45So, we believe that technology can empower doctors to do all of those steps faster and better.
12:50But then, the critical moment of the decision making stays within the healthcare professional.
12:56And same for the patient, right?
12:57So, I think it's a good analogy in terms of what we think, the way we think about it internally.
13:03AI is not here to replace the decision making.
13:05It's here to kind of simplify all of the steps that come before decision making.
13:10And I think in a lot of organizations, people tend to mistake making a decision with actually, you know, compiling
13:17data
13:18and, you know, waiving the pros and cons.
13:22This is not actually making a decision.
13:24So, AI can accelerate the entire process.
13:27But then at the end, there is this kind of, you know, critical moment where a decision has to be
13:31made.
13:31And for us, we believe it's still our job to do that.
13:35Putting the right governance in place is one of the key challenges in fostering AI adoption.
13:43In your experience, what is the right AI roadmap governance in what you've seen and what has worked in your
13:51organization at Dr. Lib and what you've seen at clients you've helped, Stan?
13:56You want to go first?
13:57Yeah.
13:58So, we did ask ourselves the question, who should lead that roadmap?
14:01Should we do a PMO, tower of control, KPIs?
14:05Is it IT or HR or finance or whatever?
14:08I think these are good questions.
14:09But in the end, same, it's not what is important, at least to me.
14:13I think what you want to create in your organization is the right momentum for the thing to take off,
14:18right?
14:18So, it's more about setting the right conditions internally for it to take off.
14:23And for that, you need internal communication, you need legal, you need HR department, finance.
14:30So, for me, the key is more, can you create that momentum?
14:33And then, once you have the momentum, once your people are willing to use, then it's going to almost work
14:40its way into your organization.
14:42So, I think people today, they focus too much on the structure of it, the governance of it, rather than
14:49creating the conditions for people to actually test it and build stuff, right?
14:55And I think in terms of the governance of the adoption of AI inside companies, there's always stuff that might
15:01feel scary, like bringing a lot of data, that data being digested by models, who has access to which data
15:09through the prism of agents and stuff.
15:11And so, I think here, the most efficient way is really to try, and that's something that companies sometimes fail
15:16to do, is really to be very incremental in kind of the scope of the data you make available to
15:22those systems.
15:22By starting small and unlocking use cases, you actually discover very nicely with your organization where the needs are and
15:31what are the kind of pockets of data that you need to make available.
15:34And so, taking that very incremental approach, I think, is something that is really important.
15:39Because otherwise, you'll be stuck in legal discussions about compliance and stuff like that for months.
15:44And we have nothing against legal, right? Of course. I don't know if there's legal team member in the room,
15:48but of course.
15:50We're here in a very tech-savvy event, Viva Tech, we are all, I think, curious about what AI can
15:58bring to our life.
15:59But we shouldn't lose sight of the question and the challenges that lie behind.
16:06Many, who are not probably here today, are afraid of what AI might create in terms of job description.
16:15When we look at entry-level job, some studies point out that it could destroy up to 25% of
16:22entry-level jobs.
16:24How do you address that? And how do you help companies tackle this challenge?
16:31Yeah, I think it's a big question. I have kids. I mean, they are still relatively young, so I don't
16:36have to care about their studies yet.
16:38But I think it poses a big question of the kind of skills that are going to be useful in
16:44the future.
16:44I think, again, in the past years, a lot was about one's ability to digest information and then being able
16:54to re-transcript it and so on.
16:57Now, I guess the skills that we will be looking for are different. The way we think about it at
17:01Dr. Lib is that I think it's an opportunity because it broadens a bit, you know, the types of people
17:08that we can hire in the end.
17:10I think that in some jobs that we have, including entry-level jobs, I think that the way that those
17:15jobs are going to evolve is going to allow us to hire even more diverse types of people.
17:22I still believe that companies have a kind of, you know, a social mission of being a nice place for
17:28people to start their professional career.
17:30And for me, I see it as an opportunity as long as we are able to hire for the right
17:35types of mentality.
17:38There was some, you know, mention of adaptability, resilience, ability to learn. I think this is key.
17:43So how are you able to find that in candidates? And then how are you able to develop that internally?
17:49So I'll give you a concrete example. We have an onboarding academy. So that's how we onboard people in our
17:54company.
17:55And a lot of it is still based on, you know, sharing tons of knowledge. So we try to pack
18:02it. It lasts a week.
18:03And so we try to pack a lot of knowledge in just a week to our new joiners. And it's
18:08kind of a passive kind of moment, which we like because we build it for a lot of years now.
18:13But now we are thinking about it differently. It's how can you put people in a more active mode right
18:20from the start?
18:21You know, and I know you want to talk about agency, but how you want to, you know, instill agency
18:26right from the first days into a company?
18:28That's a good question. I think we can all ask ourselves.
18:33Yeah, I mean, I think the parallel between the AI revolution and the computer revolution are very good to have.
18:40It's going probably faster than the adoption of AI is probably going faster than the adoption of computers.
18:44But there's been, I mean, some jobs are going to be transformed drastically. Some jobs are going to be less
18:50transformed.
18:51Some jobs are going to be so transformed that they won't look the same at all anymore.
18:56But in terms of entry level, I don't think it's about there's going to be less work.
19:00It's just that it's going to be better work because a lot of that work that's used to be done
19:04at entry level will be doable by the machine.
19:06And so the one skill that I think really becomes extremely important is agency.
19:14If you even if you're a new joiner in the company, you with AI available, you don't necessarily have to
19:22pay the price of learning the hard skills of doing this or that,
19:26because the machine will be able to assist you and drive you much faster.
19:30But then it's going to be a question of creating the environment for you to take actions that will benefit
19:36your team, benefit the company as a whole,
19:38and really being the actor of taking of enacting these actions.
19:43And so being a doer, having agency, I think, is really the key skill that people have to build in
19:49an environment where work is going to be accelerated a lot by this.
19:55Mathieu, you mentioned that you kind of had to evolve the skill set that you are looking for for the
20:01new hires with more greed, resilience, adaptability.
20:07And how do you do for the existing workforce? How do you upskill your workforce to make sure they are
20:14still adapted in this new AI world?
20:17Yeah, I think you need to make it clear first, you need to make it clear that that's, you know,
20:22the direction of the company.
20:25I think also you need a lot of pedagogy and empathy.
20:28I think we've seen we've seen some examples of US companies mostly, I mean, and European companies actually try to
20:35go really fast in that adoption, which I think is always good case studies.
20:39You know, Clarna, which is a Swedish company, or Duolingo, which is a US company, they've really tried to push
20:47the boundaries and go really fast.
20:49I think it's very interesting. Those are interesting case studies.
20:53That's in killing job.
20:55Yeah, yeah, yeah. So basically saying that, so Clarna, for example, it's a fintech company that decided to shut down
21:01a lot of their customer service jobs,
21:03because they say that you don't need customer service anymore because the, you know, machine can do it for you.
21:10And so they almost killed their entire, replaced their entire customer service team.
21:16So hundreds of jobs, thousands even by agents. They had to revert back. So they did it, and it failed.
21:25And so now they had to kind of, they almost issued an apology and they had to revert.
21:29So they hired back customer service agents. So I think those case studies are interesting, because you want to be
21:36super clear internally about that's the direction that work is going to take.
21:39But you also want to have a lot of pedagogy and empathy for your workers. So it's, again, about to
21:45be finding the right balance. And this AI cannot do it for us. So again, it's a leadership job to
21:52make the right call.
21:56Maybe to wrap up, one last question for each of you. Stan, I'm not going to start with the easy
22:03question.
22:04If you had to place a bet in five years of what AI can't do now, but it will be
22:10able to do, what would it be?
22:13Wow, five years is a long-term horizon. It's like a century. No, I think five years is almost impossible
22:22to guess. We don't know if the technology will keep accelerating, we don't know if the technology will plateau eventually.
22:26But let's say maybe, I think about 18 months or two years, I think the one thing, the bet that
22:33I'm making is that we might see a world where software developers actually don't write code anymore, which is going
22:42to be very weird.
22:43That means that they're going to be, and we see that internally in our team today, you spend your day,
22:49you say, I have this task, I'm going to start four agents on the task, and four agents on that
22:54other task, and four agents on that task, and then I'm coming back five minutes after, and I'm picking the
22:58best solution from each of those groups, and move forward with that.
23:02And in kind of ten minutes, I've accomplished three things that would have taken me maybe one hour each, sequentially.
23:08And so entering a world where software developers don't write code anymore is very interesting, because they're going to have
23:13to build as a software developer skills that are more about not necessarily managing, but being able to verify the
23:19work of agents and interacting fast with them.
23:22And I think in the case of Klarna, it's the same, it's the real trade-off is not removing the
23:27human, but more giving the ability to the human to say, okay, ticket, go for it, ticket, go for it,
23:31ticket, go for it, and then come back, say, looks good, looks good, no, doesn't look good, you have to
23:35change it.
23:36Is it a better job than writing the tickets? I don't know. Is it a more intense job? Probably as
23:41well. But at least it's a job where you're creating much more value than you would be for, for sure.
23:47I would be very happy to have this conversation again in five years to see how we stand.
23:53Mathieu, to finish, at Poland, we see a lot of companies who are at the beginning of their journey in
23:59terms of AI transformation.
24:01You've started it pretty early, being a tech company, being able to move fast also, because that's in your DNA
24:09and you're still only 3,000.
24:12What would be the one advice that you would give to companies at the beginning of the AI transformation journey?
24:21Yeah, so I think it's a critical issue. I think that in the next years, if our company, I'm going
24:28to talk about my company, if we are not able to embrace that, what's going to happen is that we
24:31are going to lose the race for talent and we are going to lose the race for our business, right?
24:36I think it's going to be impossible for our company to keep attracting the best talent if we are not
24:41able to embrace everything that's been discussed for the last minutes.
24:46And also, just from a pure business standpoint, I don't think that we are going to be able to remain
24:50one of the world leaders in healthcare if we are not able to embrace that.
24:54So, it's a mission critical type of thing. And at the same time, there has been no revolution or no
25:02wave that has been successful thanks to governance.
25:08Governance is not going to make or break everything that we discussed. I think it's going to be about ability
25:12to take risks and try to have fun also in the process.
25:18So, yeah, I think take risks, have fun, and then, you know, we will see each other in a couple
25:22of years to share best practices.
25:24But I guess that's the only advice that I can give at this point is the only thing that we
25:28know is that we'll have to try it.
25:30Choose your try.
25:32Yeah, exactly.
25:33Thanks a lot, Stan and Mathieu, for this fascinating conversation. It was great.
Commentaires