- il y a 1 jour
Engage with the future of robotic AI assistance with Salesforce & Enchanted Tools
Catégorie
🤖
TechnologieTranscription
00:00Hello, hello everyone. Thank you for being so many of you today. We're very excited to have you for this
00:09specific session. So let's get started.
00:14As you probably already know, Salesforce is a publicly traded company and we recommend our customers to base their purchase
00:25decision on the public product and not roadmaps.
00:29OK, so today we are the three of us on stage. Plus, obviously, you noticed the wonderful robot. So I
00:39am myself, Sébastien Jamon, very pleased to be with you. I'm at Salesforce, an AI director. I am today joined
00:47from San Francisco, Carlos Lozano, PM Agent Force.
00:51Bonjour, good morning.
00:54And we are super pleased to welcome Sacha Duke on stage as well from Enchanted Tools.
01:00Thank you, bonjour.
01:02And of course, Miraka, right?
01:06Yeah, Miraki.
01:07Miraki.
01:08Miraki.
01:08OK.
01:09All right, you guys. So, first of all.
01:14Yes, a short introduction about Enchanted Tools. So, hello, I'm Sacha Duke. I'm co-CTO at Enchanted Tools. I'm in
01:22charge of human-robot interactions.
01:24Enchanted Tools is a French company based in Paris. It was founded by Jérôme Monceau, a pioneer in robotics who
01:35created now on paper robots.
01:38Today, over 100 engineers, developers, designers, artists work together, all driven by the mission to re-imagine what the robot
01:51can truly be, and especially how they can behave in social environments.
01:57Let me introduce you to Miraki. We wanted to do more than build a smart and useful robot. We wanted
02:09to bring them to life. Most robots are called mechanical task-driven.
02:18And at Enchanted Tools, we believe we should, robots should also inspire, interact. So, that's why we created the Miraki.
02:30Not just a robot, not just robots, companions, not just tools, characters, not just efficient, but also magical.
02:42Thank you so much. And today, we are collaborating with Enchanted Tools. And us at Salesforce, we provide AI for
02:52business. So, Carlos, why don't you talk about it?
02:54And we're coming from the Agent Force side. So, I'm coming from the Agent Force authoring side, which is the
03:00ability to create this business agents.
03:03And so, what's beautiful is that we want to show you how Miraki, Miraki, and Agent Force can actually show
03:09the next generation of facility management experience, consumer experience, and user experiences.
03:16And with that, we're going to show you how we were able to essentially hook the Agent Force intelligence, and
03:23together with Miraki's intelligence, we're showing you a vision of the future.
03:29All right. This is the purpose of our demonstration today. Why don't we bring the best of business AI with
03:37the best of robots out there, right?
03:40What do you think? Should we do it?
03:42All right.
03:43Let's do it.
03:44Let's do it. Okay. So, let's do, like you said, an example, facility management. Okay. And let's imagine we are
03:53at a building, you know, and we are entering the building, and we are welcomed by the robots, and our
03:59objective right now will be to try to book a room. Okay? So, I let you continue the demonstration, Sascha.
04:08A short role play.
04:09Okay. Hello, Miraki. Hello, Miraki. I'd like to book a room.
04:20Come again?
04:22I'd like to book a room, Miraki.
04:29Okay.
04:29I didn't understand.
04:32Maybe we have trouble with the connection, but okay.
04:37Okay.
04:37Plus fort.
04:39No, we have trouble with the internet connection.
04:43Ah, internet connection.
04:46And what are we showing here? On this dashboard, actually, what we have is basically the intelligence of Miraki and
04:51how he's basically engaging with Salesforce objects.
04:54And this is actually a conversation that we had before.
04:58Yes. Yes.
04:59And actually, if we cannot talk to the robot.
05:05Okay.
05:05Okay.
05:06Okay.
05:07So, why don't we do it on our side, right?
05:09Carlos?
05:11We're going to show you, essentially, how it actually works behind the scenes, and how we build that agent that
05:17basically receives you in a co-working space.
05:20As a user, you arrive to the co-working space, and what is the first thing you're going to say?
05:24You're going to identify yourself.
05:25And as you identify yourself, then it actually knows what booking you have.
05:30And because it knows what booking you have, it'll actually take you to that room.
05:35And so, here we are, actually.
05:37If you go back.
05:38So, what we have here, essentially, is the agent builder.
05:41What we use to basically build this intelligence, right?
05:44And so, you can divide it in three pieces.
05:47Imagine you are an agent builder now.
05:49We are all agent builders now.
05:51On the right side of the screen is where, as an agent builder, you're going to simulate the conversation that
05:58the agent is going to have with users.
06:00And so, as you can see, Seb, actually, already started interacting with this agent, essentially letting it know, I need
06:07to book a room.
06:08I am Sasha from Enchat 10 Tools.
06:11And the agent, the robot, is actually telling you, that's great.
06:14Give me your one-time password.
06:16And that is the one-time password that you see there, very, very difficult to hack, right?
06:21One, two, three, four.
06:23And as you actually, that was a bad joke, okay.
06:26As you basically put that one-time password, what's going to happen is that now the agent and the robot
06:32knows who you are.
06:33And it can interact with the business objects for that facility management experience.
06:38And so, it's basically giving you that summary.
06:41You have booked this room at this time.
06:44And this room has these amenities, okay?
06:47You get the idea of the right-side panel.
06:49But what we want to draw, you know, your attention to is how do we create this, right?
06:55How are we building agents today?
06:57And the way we're building agents is actually pretty simple.
07:00This is not a programming that you don't need to be a developer to do that.
07:05But what we're going to do is that we're going to give it a job to be done, a responsibility.
07:10And here, it's on that left side.
07:12You see topics.
07:13So, it's actually simplified the experience here.
07:15It has one topic.
07:17And that topic is essentially meeting room management.
07:21And so, that topic, that job to be done, that capability actually has inside a number of tools.
07:29In agent force jargon, we're calling this agent force actions.
07:33So, if you click on that sub-tab over there, if you scroll up a little bit, Seb, there's a
07:37number of actions that we have given this agent to basically manage that room, right?
07:42So, searching the user, providing, you know, that code verification, managing the booking, and so forth.
07:49And so, now, what we have to ask ourselves is, how is it actually going to invoke these actions?
07:55How is it going to do that?
07:57And the answer to that question is topic instructions.
08:01So, if we go to the left tab there, topic configuration, right?
08:05This is where it knows what this topic is about because it has a title.
08:10It also knows what this topic description is.
08:14This is important because at runtime, it will be able to pick this topic because it is described as such.
08:21It also has a scope of work so that it doesn't go out of rails, if you will, right?
08:27And then it also has this idea of instructions where you tell it what you want it to do, where
08:33you tell it how you want it to use the actions, and where you tell it what you want to
08:38restrict as well.
08:40And the combination of all of that is what we had on the right side panel.
08:45Last piece that I would love to show you guys is essentially, maybe we can even X out, actually, the
08:52topic details panel here.
08:54Here is essentially how the brain is operating, the reasoning engine, and how it's actually operating.
09:02And so, notice that when Sebastian started, he started a session, right?
09:07And then, as a user, right, Sebastian basically asked it a question.
09:11Hey, right, I need to book a room, right?
09:15And so, at that point in time, the system is going to reason, and it's going to ask itself.
09:21It's going to auto-reflect, and it's going to say, what topics do I have available to answer this request?
09:29And when it basically finds the right topic, it's going to auto-reflect again, and it's going to say, what
09:34actions do I have available to fulfill the request?
09:37And it's going to, for example, also pick the information from the user, and it's going to do that autonomously.
09:45Sebastian, as an agent builder, did not program it to basically go ahead and ask for Sasha's information.
09:52It actually did it autonomously.
09:54Why? Because it has an action, but it needs to collect the user info.
09:59And if the user doesn't provide the info, it's going to auto-reflect and say, I need to basically slot
10:05-fill this information.
10:06So, what you have in the middle here is what we call the plan canvas, right?
10:12The reasoning canvas, the different steps at every user request.
10:17What topics do I have available?
10:19What actions do I have to execute so that I can fulfill the user's request at every step of the
10:25way?
10:26Right?
10:28Thank you. Thank you, Carlos.
10:29Why don't we try to spice it up a little bit?
10:33Should I add an instruction maybe to, you know, change the behavior?
10:38Yes, that's a good idea.
10:40So, if we actually go, you know, around the use case, something that actually happened after the user actually was
10:47in the room is that the user called out,
10:49hey, there's actually an issue in the room, right?
10:52And so, the robot and the agent was like, hey, I'm really sorry to hear that.
10:57Can you basically describe what issue is happening here?
11:00And so, then, the user essentially responds, well, the video conference system is not working.
11:08That's really pity, right?
11:09We could have avoided that.
11:12How would have we avoided that, right?
11:14Well, we have to tell the agent and the robot to be mindful of these things.
11:19If the room, when you scan it, has one of the devices that is not working, should you give it
11:26or not to a user?
11:28Well, probably the answer is you shouldn't give it.
11:29You should find another room.
11:31And what we're going to do is that we're going to give it that instruction so that next time around,
11:37for another user, it won't make this mistake, if that makes sense.
11:42So, the way to do that is we're going to actually add an instruction.
11:46So, you can go ahead, Seb, and actually, how do you want to describe this instruction?
11:51You can do it telegraphically, and maybe you can explain what is the purpose of this instruction, right?
11:57So, here, for example, I want to make sure that the equipment should be okay before to book it, right?
12:04Let me stop you there for one second, because notice that equipment has a typo.
12:09Il y a une faute d'orthographe, okay?
12:11So, Seb is kind of like Gen Alpha.
12:13He's writing in emoji before, like B4, like B4, right?
12:17We're not going to use this instruction as is, because this instruction isn't properly formatted for the reasoning engine to
12:27use it at runtime.
12:28And what we're going to do to actually solve this is we're going to use AI to help us build
12:35the AI.
12:36So, you notice this little, like, sparkling stars right here, right?
12:41So, this is essentially an agent builder agent that's behind the scenes that's helping you configure the agent that you're
12:50building.
12:51It's a little bit meta, but you'll get the idea in a second.
12:54So, it's basically going to suggest an improvement on that instruction.
12:58It's explaining what it's doing.
13:01Ensure the instruction is clear, detailed, and free of ambiguous language.
13:06Seb, you can't use Gen Alpha language, and, you know, you have to basically, you know, you have to be
13:11Molière a little bit, okay?
13:13So, use proper spelling and grammar.
13:15So, the AI is telling us how to prompt the AI, right?
13:19And then it gives you a suggestion.
13:22And so, the instruction is what you have here in the, dans le cadre grisé, right?
13:28Ensure that all equipment in the meeting room is functional before proceeding with the booking.
13:33If you like that, you can accept that.
13:35And then that will automatically, now we have prompted an additional instruction to our agent, and we use AI to
13:43build the AI.
13:44So, this is actually, you know, a little bit of beauty of AgentForce.
13:49If you are able to describe it, then AgentForce will be able to do this.
13:54And this is a good example of it, that we call instruction tuning.
13:59And I'll say this.
14:01It's important because this is an iterative process, right?
14:05When Seb is going to deploy this agent again, we're going to observe the agent.
14:11We're going to see how it's doing, how well it's doing, and how bad it's doing, right?
14:16And as we're looking at these dashboards, and maybe we can take a flash of the other dashboard, right?
14:22Like, as we're looking at these dashboards, observing how the AI is performing,
14:27we're going to keep on tuning, and refining, and testing, and deploying, and the loop goes again.
14:34That's the life of an agent builder.
14:36Should we give it a try?
14:38Should we run a session?
14:40We can give it a try, absolutely.
14:42So, go back here, and do the demo again, okay?
14:46We can do the demo again.
14:47And so, what Seb is going to try to simulate here is the conversation,
14:54so that when we arrive to the facility management, to the co-working space,
14:58we're going to identify ourselves as, in this case, right, as the user.
15:06And so, and then, you know, provide, basically, the user verification.
15:12You can do a hard refresh.
15:19And so, Sasha is actually entering the co-working space, if you will.
15:25So, we're going to, basically, impersonate Sasha here.
15:28And Sasha actually has a one-time password, or OTP, if you will, right?
15:34And at that point, notice, at every user request, the planner is reasoning, right?
15:41What is this about?
15:42What is the user intent?
15:44Do I have any topics to fulfill it?
15:46If the answer is yes, do I have any actions to execute?
15:49And so on, and so forth, right?
15:51The system is constantly, look at this one, right?
15:55So, it was able to pick up the code, right?
15:58And it picked up the code autonomously,
16:00and it stores it in a variable that it will use across the session.
16:04So, the middle panel here is a little bit techie,
16:08but it's important so that you can see le trip,
16:11the guts of how the agent and the robot is basically reasoning, if that makes sense.
16:15So, yeah, we can keep going, right?
16:18So, you can book your room.
16:19Let's book your room from 1 p.m. to 2 p.m., right?
16:22Go ahead.
16:22Okay.
16:24Let's do that.
16:26And there's two rooms available.
16:29Which one do you prefer?
16:32This is the way.
16:33Edison.
16:41Oh, see how clever it is.
16:43What is the purpose of the meeting?
16:47Okay.
16:47So, at every user request, right?
16:50It's going to go in planning.
16:52It's going to check out the topics that are available.
16:57And it's going to basically invoke the actions that are available for you to fulfill and to go along.
17:02And that is how we added additional instructions to an agent that we can connect to, for example, a robot.
17:09Yep.
17:09Amazing tip.
17:10Thank you.
17:10Thank you, Carlos.
17:12Thank you, Carlos.
17:12All right.
17:14So, let's keep on going with the presentation.
17:19So, let's go here, right?
17:25Here?
17:25Yeah.
17:26Okay.
17:27So, that was one example, Carlos.
17:30Yeah.
17:30Okay.
17:31So, I guess what we wanted to showcase is that, first of all, we showed you how we are using
17:36AI to build the AI.
17:37And we used one example, which is like topic instructions.
17:41But when you basically instantiate a Salesforce environment or a Salesforce org, there's going to be a few options.
17:49You can start creating agents with AI.
17:51If you go to the Salesforce stand over here, you are able to build an agent using AI.
17:56And this is the idea.
17:57If you can describe it, AgentsForce will be able to do it.
18:00So, we showed one example, facility management, okay?
18:05But there are tons of examples, tons of templated agents for you to mettre le pied à l'étrier, get
18:12started, and start, you know, building your agents from these templatized agents for different types of business verticals, if you
18:22will, right?
18:23And that's on the agent for site.
18:25And together with Enchanted Tools, what type of use cases are you covering, Sacha?
18:30Thank you.
18:32The Miracai has been designed to work on different environments, such as hospitals, nursing homes, pediatric units, stores, and hotels.
18:46There, we have an approach based on three main aspects, which is API.
18:56As a social companion, Miracai can talk with residents, with customers, give information, advice, and lead activities.
19:07They bring joy and peace of mind for residents, for people.
19:14Where we have also travel, means the Miracai can guide people, customers, in different environments, of course, detect danger or
19:25unsafe situations, track objects, and monitor at night.
19:29And finally, the Cari, obviously, the robot can do small logistics tasks, like carrying meals, medical supplies, packages.
19:42All in every setting, the Miracai are scalable, safe, and loved, and it's more than just an helper, it's an
19:54emotional teammate to make life safer, happier, and easier.
20:00Thank you.
20:03That's amazing, and I think that's what we're excited about.
20:06When we're imagining a world where if you go to your favorite fashion retail provider, then you can be intro'd
20:14by a Miracai connected with Agent Force.
20:16If you're actually traveling and you're going to be in an airport, you can be introduced, you know, by a
20:20Miracai and actually Agent Force, right?
20:22Like, think about the London airport that actually uses Agent Force, now it could be using Enchanted Tools.
20:27If you're actually, you know, in the elderly loved one space, right, where you want to have, like, this companion
20:33that not only is able to understand the business,
20:35but that is extremely cute, right, like, and that can help elderly loved ones, this is where we see, you
20:40know, a lot of potentials, and we're excited about it.
20:43All right, you guys, we are arriving at the end of that presentation, so let's summarize what we just tried
20:54to show you today.
20:56Okay, let me try to refresh my screen, if we can go back to the screen.
21:01Yeah, thank you very much.
21:02Okay, so today was all about robots using AI, business AI, so Salesforce AI.
21:11I hope you enjoy the demonstration that we did from a back-office perspective, so the back-end force with
21:17Agent Force,
21:19and the back-end system was obviously Salesforce in our demonstration, but it could have been whatever back-office system
21:27of your choice, okay?
21:28So that was technically what we used, and we are at the end of our presentation.
21:35Thank you so much, everyone, for joining.
21:37There's actually a build-your-own agent right behind you, and we're going to be with the Enchanted Team partners,
21:43also at the theater, in a Q&A.
21:45If you want a deeper dive, if you want to go and have more questions, if you have a use
21:50case that you can describe, that you can imagine,
21:54let's build a business agent together right there, right behind you, and we'll be happy to drive you there and
22:00show you how you can do it.
22:01Thank you so much for your attention.
22:02Just to cut you off, though, because we have five minutes left.
22:05Yeah.
22:05We can certainly get a couple of questions.
22:07That's fantastic.
22:08Everything you've been saying.
22:08Let's do it.
22:09Yeah, great idea.
22:10Any questions from the audience about what you're seeing?
22:12Unfortunately, we didn't get a full demo, right, because of some Wi-Fi issues, but at least maybe if they
22:18have some clarification questions.
22:20We are offering an opportunity to come see the demo at our booth, and hopefully we'll have better Wi-Fi
22:27there.
22:28So if you want to keep going and have the conversation with us, come from 11 to 12, and hopefully
22:35we can do the demo there.
22:36All right, any questions quickly, though, from anyone here, about what they saw, about the service that they're offering?
22:45No?
22:48Okay, both.
22:49Thank you, everyone.
22:50Thank you.
22:50Thank you, Ryan.
22:51See you right behind you in a couple of minutes.
22:53Thank you.
22:53Thank you.
22:55Thank you.
22:55Thank you.
22:56Thank you.
Commentaires