Dreamforce 2024 | Dreamforce Analysis

Clips
News
More from Dreamforce 2024

George Gilbert

Principal Analyst, Data & AI

SiliconANGLE & theCUBE

Christophe Bertrand

Principal Analyst

SiliconANGLE & theCUBE

Rob Strechay

Dir./Principal Analyst & Host

theCUBE Research

Three insights you might have missed from Dreamforce ’24

Organizations in today’s digital landscape rely on interconnected systems, meaning integrating platforms and unifying operations is critical. As AI becomes a bigger part of modern systems, technologies such as Salesforce Agentforce AI are drawn more in focus.Practitioners, meanwhile, have proven to be the biggest benefactors and influencers of data-driven markets. With its leading role in automation integrations, Salesforce Inc. has had its eye on the broader enterprise AI market. The company’s plans around automation and for Salesforce Agentforce AI were a central focus of the recent Dreamforce event.Christophe Bertrand and John Furrier of theCUBE talk during Dreamforce 2024

The future of Salesforce’s platform and AI ecosystem: theCUBE’s Dreamforce event wrap-up analysis

Salesforce Inc. is at a crucial turning point. Beyond customer relationship management and other business-related software solutions, the company is navigating a path to infuse business processes with cutting-edge agentic AI.George Gilbert, Rob Strechay and Christophe Bertrand provide exclusive analysis at Dreamforce 2024.“I think there’s a reason why [CEO] Marc Benioff said this is the most important Dreamforce ever,” said George Gilbert (pictured, left), principal analyst at theCUBE. “It’s very hard to turn a platform company into an application company or vice versa. If they pull this off, they’ll be one of the largest to have ever done it and they’ll transform the company

play_circle_outline Building on top of Salesforce platform

play_circle_outline Platform transformation challenges

play_circle_outline Hybrid cloud and partnership with IBM

play_circle_outline Importance of data for AI and agents

play_circle_outline Impact of agents on business differentiation

play_circle_outline Empowering Domain Experts: Tools for Expressing Thought Processes in Secure Data-Driven Organization

play_circle_outline Value in building custom apps on Salesforce platform

Info

Dreamforce Analysis

George Gilbert

Principal Analyst, Data & AI SiliconANGLE & theCUBE

Christophe Bertrand

Principal Analyst SiliconANGLE & theCUBE

Rob Strechay

Dir./Principal Analyst & Host theCUBE Research

HOST

search

Rob Strechay

>> Hello and welcome back to Dreamforce 2024 from the NYSE. Our studio is built right up the street from the Moscone Center. We're wrapping things up with day two of Dreamforce, which is actually their day one. There's been a lot going on today, a lot of keynotes, a lot of platform talk. We've been talking a lot about ecosystem. We've heard from a lot of the ecosystem they're building on top of Salesforce as a platform, doing things with AI and agents as well as some of the partners where they're selling in their marketplaces such as IBM and AWS. It's been a fascinating full two days to put it mildly, and we're really excited to break this down. Today, to help me wrap this up, I'm once again joined by Christophe Bertrand, who's one of our principal analysts with theCUBEResearch and George Gilbert, who's one of our principal analysts also with theCUBEResearch. So, guys, and I'll kind of throw this to George first, we've been talking about platforms and data. We had some really great guests from Salesforce on with MK and David, and we're really digging in. What's your takeaway so far about the platform aspect?

George Gilbert

>> Well, I think there's a reason why Marc Benioff said this is the most important Dreamforce ever. This is now 25 years since they founded the company because it's very, very hard to turn a platform company into an application company or vice versa. If they pull this off, they'll be one of the largest to have ever done it and they'll transform the company. That's sort of a cloud-level way of saying, and no pun intended, of saying that all the data that Salesforce has accumulated and help their customers shape and inform. So it's not just the customer 360, but anything that customers choose to connect to the customer 360 that can activate used to be like activate a marketing campaign or nurture a lead. Now it activates an agent and an agent embodies business processes that were too complicated to specify. Now they learn and they can... Now, Salesforce becomes the platform on which others can build agents.

Rob Strechay

>> Yeah, no, I agree. I think one of the things, and Christophe and I have been talking to a lot of the ecosystem who I think ecosystem has been out in front of this a little bit on Salesforce and building on top of the platform because ultimately having used Salesforce myself at multiple different startups, it really is difficult to program. When you programmed into force.com and what have you on top of Salesforce, it became very static and making those changes. So to your point, I think if they can pull this off where it becomes a more dynamic app platform, more personalized, really focused on those processes all the way back to the data, it's going to be a clear win for them and especially where they've abstracted and they're not a cloud. They're sitting on cloud, they're sitting on IaaS, and I think that's a huge win for them is that they are SaaS from that perspective. So, what about yourself, Christophe? What have you been seeing?

Christophe Bertrand

>> Well, I'd like to go a little further than this and say it's not just clouds, a hybrid cloud, it's also Db2 and mainframe. It's working with IBM for example. Great conversation we had with the head of global partnerships on this topic. All of this is also very interesting from the perspective of how nimble it makes the ecosystem. I mean, you're right, a platform, George, is something you build on. That's the definition, right? But it has to be pervasive, and I think by being so open to many data sources, multiple hybrid cloud type of environments, that's going to be the key. Now, the challenge that I see, also for every opportunity there is a challenge or more, will be how do you really keep things compliant? How do you keep things governed? That's really the biggest question I have because the play here is a scale play and with scale comes a lot of responsibility and complexity.

George Gilbert

>> I do have one comment to add on that from previous interviews with MK who's president of the developer platform, which is, we used to build data platforms with role-based access control. Like here, you belong to this role, this role gets access to these tables or columns or whatever. Now, the most advanced data platforms like Salesforce have a policy-based security model where you can put tags on the data and then formulas as to who gets access to data by what tag. I know that's dropping into the weeds, but it's part of the theme that it's not just good enough to have a data lake with a very high performance SQL database anymore. It's the meaning you put on top of it, the policy you put on top of it. To your point, the low-code tools and Salesforce had the low-code tools now by putting the agents in them. You can program those low-code tools either with English or just giving them a goal and they figure out the context from the data and the available actions.

Christophe Bertrand

>> Well, that's going to be key. I will tell you why. Because you're going to have all of these AI workers, for lack of a better term, and how many are you going to have in any environment? It's going to be ridiculous the number of agents that might be running in a large environment as the ecosystem matures and the use cases expand. So, I think it's going to be more than the data tagging. The data level is also going to be managing all those AI workers.

Rob Strechay

>> Yeah, I mean I think building off both what you're both saying is that the fact that these agents are doing business processes, it's doing revenue recognition, it's doing closing of a sales campaign, it's doing a flagging of distressed customers within a service cloud or what have you. I think what they're doing is really trying to enable people. I like Benioff's talk and we heard it from David as well, who's president and chief product officer, David Schmaier. Really, it's about enabling people and I think they weren't getting ahead of their skis with their agents and saying this is going to replace people. Their talk track to it was this is super important to make people more efficient, especially in this market, this economy where you're talking about how if you can find the right people, they're hard to retain. Most of the time you can't find the right people and you can't afford them for that matter. So, I think it's a very interesting dynamic going on in the ecosystem around them where there are some that are saying, "Hey, this is replacement." I mean, data we know is what AI is built on. We've talked about it all this week, we've talked about it weeks before. I think the control of that data and the control of the metadata, it'll be interesting because MuleSoft gives them some advantages and they're zero ETL alliance that they have with AWS, IBM, Databricks, I believe Snowflake's in there and others makes a lot of sense, but the metadata doesn't necessarily travel with that and that context. So, I wonder at what level can they bring and bridge enough of the other data within an organization to really have the ownership and be that platform. That's what I find interesting.

George Gilbert

>> Okay, it's a good point. To that, I would... Oh, I got Siri talking to me. Okay, that's like in the weather reports where the cat hops up on the weatherman. So, I would point out that the idea of this federated query in their data platform, they want to be the historical source of truth. And to do that, some of the data that they can federate out to actually fits into harmonized definitions of the customer, not all of it. So it can gracefully degrade. Actually, the issue of the policy, it applies both to humans and to agents and what we'll likely see is a org chart for agents that complements an org chart for humans. Agents can call on other agents and when they're not confident of a step in a process or an outcome, they can then bounce up to an inbox to a human to supervise them. The point of all this isn't to eliminate humans so much as to capture expertise for the easy jobs and processes. Not full jobs, processes or tasks within jobs, and that you're embodying in this org chart of agents the expertise to perform more and more of the routine functions in an organization, and then it up-levels the humans.

Christophe Bertrand

>> It's interesting what you're mentioning here because I go back to a conversation we had with Own company on backup and recovery and the fact that they literally keep an archive all the data. I mean, they see everything. It's backup and recovery. So, having access to all of this now becomes a way to augment further all of this work that's going to happen at the data level and the platform level, but also a way to go back and really inspect what may be going on, what's happening. So, there are really a couple of bidirectional conversations going on here.

George Gilbert

>> Yeah.

Rob Strechay

>> Yeah, and I think one of the interesting things is, and I think this actually brings it all... We keep going up and down the stack. I think one of the things that was interesting was how they're bringing forward Heroku as well and talking about how the developer life cycle. To your point, I think that people are going to look at AI and agents and they're going to want to be able to control the CI/CD for these agents. They're going to need lineage. They're going to need all of these pieces to be combined. So if you're going to have a platform, you almost have to have that Heroku, which hey, it understands Kubernetes under the hood. It knows how to actually distribute it. It knows how to put the things in the right places and orchestrate that from a Kubernetes perspective. It takes away and simplifies the platform so that they work on developing the agents and connecting the data and again, trying to figure out how is all the lineage work and all of this. That's a whole different story that I didn't hear much about, but I think when you look at it, it's abstracting the infrastructure and I think we briefly talked about that yesterday.

George Gilbert

>> Yeah, your point really shines a light on what they kept trying to say in the keynote, which is don't do it yourself. The whole point about the DevOps that you're talking about, DevOps for an agent is it's not just release cycle, but a continual capture of the exception conditions. The exception conditions become teachable moments, and then you look at those and you say, "Okay, which ones had favorable outcomes? Which ones had unfavorable ones?" Then the idea is that you periodically capture those, you add them either to the prompt database or to the fine-tuning, and then the distribution of outcomes becomes narrower around the best practice for that task. It's a whole different tool chain, but underneath, you still need the data lineage to make sure that the data that the agent is using as context is trustworthy.

Christophe Bertrand

>> Look, it's always going to be about good data. There's no doubt about it. So that's the part, to me, that's the constant. It has been the constant since mainframe. That's number one. I think number two is how do you train agents? How do you train humans to train agents and vice versa? I think we will see in the future how things evolve, but I do believe that while they will be some entry-level jobs, very likely that were not that productive to begin with, that may be eliminated. It's an opportunity for a lot of people to actually go up their own stack, if you will, in terms of professional excellence to be empowered and augmented by agents. I think I can think of 25 or 30 examples a day where I would love to be able to do more by just asking a couple of questions to a smart person when I have to resolve something. So, I think that's one of the promises. Again, I think it's early stages. I haven't seen a lot of metrics on any of this yet, and that is something that bothers me a little bit.

George Gilbert

>> Oh, it's interesting that you should mention metrics because one of the things that we will see right now, metrics are they're isolated dashboards, they're frozen, they're one way, they're like looking in the looking glass. But we're going to find metrics are not just what you measure like human or organizational performance by, but those are going to be the metrics by which agents measure their performance. And so you need to make sure you're choosing the right metrics. They all form a tree. It's not like a Newtonian machine because a little bit of a change here and a little bit of a change there might make somewhat of a change here. It's more of a complex system, but the agents are going to be in their own org chart wired up to this metric of how the organization performs. Now, this sounds like it's a blue sky, but it's actually a journey where you wire up the metrics bit by bit. You get the relationships between them higher fidelity, and then you connect the agents progressively to more and more of the metrics, and that's how you align the agent behavior with the goals of the organization.

Rob Strechay

>> Yeah, I think just building off of that, I thought what was really interesting was how they talked about how people are going to actually add to agents. So, customization of the agents and adding different ways of doing tasks, or to your point about, hey, what are the anomalies that I see in the agent behavior and become a teachable moment for the agent is going to differ company to company. I think this was the thing that I sat there and it was interesting to see and just to compare this towards last week when we were talking about Oracle releasing 51 different agents, which is still 30 some odd more than Salesforce announced this week, but is it around volume or is it about efficiency? Are we going to get agent to death? We were talking to some of the organizations that were coming through and one of the interesting theories was last year it was all about co-pilots, this year it's all about agents. I think people are still grasping at the ROI of AI and trying to see, okay, what work will get achieved by these agents? Now that used to be co-pilots that were very broad, now agents are very task-oriented and focused. What are you seeing when you were talking to other people as well and floating around? Are people looking to use the packaged agents or are they looking that their... They feel like they're snowflakes and that they have... and not the company, but you know?

George Gilbert

>> No, no, that was a good analogy. I think what's going on is that you need to show concrete solution oriented value to get in the door. So there's at least one agent atleast in each of the clouds, whether it's marketing cloud, commerce cloud, revenue and order service, there's got to be at least one agent. Usually there's a couple in each of the clouds. But the critical one then is the agent development toolkit where it's now... With low code it was already a revolution in accessibility to corporate or citizen developers to adapt these clouds to their processes. But now with agents, they can go in order of magnitude, two order of magnitudes farther in terms of usability and crucially in the types of processes they can accommodate. Because when you programmed with low code before or any symbolic programming language, you were basically saying, "I know the answer, and it follows these sequence of steps." With agents, you get to the other 80% where you're saying, "Well, let me just show you some examples." We call it the dark matter of processes where you show it some examples and you give it a goal and it sort of figures out how to do it, but the building blocks are in the applications. That's what Salesforce has, perhaps Oracle, that just a pure data platform does not have.

Christophe Bertrand

>> I'm going to take this back to the business level, kind of a notch up from the technology here. When I'm hearing this, I'm immediately thinking, "Well, if I want to differentiate my business, whatever business I'm in, I'm going to invest in the best agent people who can understand agents, develop them, integrate them, because that's where competitive differentiation will end up happening." I'll be more efficient as a business, and in a sense, it's sort of giving power back to humans in many ways that way. But I do believe that's a fundamental in time, fundamental business move to invest in this type of technology and specifically the agent components at the process level, at the outcome level.

Rob Strechay

>> Yeah, I think actually that's a great segue into the discussion we did have with David around this, Schmaier, and I think part of it was when he brought up that we have these PhDs out of Stanford and all of these people and we have that group and doing R&D. It actually reminded me when you and I were sitting with Ali Ghodsi in Databricks, it was like, "Yeah, we have Mosaic and we have all these really smart people and all these people from Cal." It's almost becoming a university versus university fight from a platform perspective because I think exactly what you're looking for is Salesforce is sitting back and saying, "I don't want to hire..." Most companies don't want to hire Cal, Stanford, MIT, whoever.

George Gilbert

>> They can't.

Rob Strechay

>> Yeah, they can't and shouldn't for that matter. It's not their core IP unless they already have them for doing quant and all of the other ML training and trading. I could see it in the financial services-

George Gilbert

>> The data science.

Rob Strechay

>> Yeah, but if you're a healthcare company, you're an HMO, why would you go and hire all these really expensive... You're not. So you're going to look for something like a platform to your exact point because you can't go find the agent builders. You're going to want to start somewhere with these agents. So when I look at what Salesforce is doing, we had numerous others on... We had some great people talking about how they're helping build agents or building agents for specific things like hey, going through and quoting, configure price and quote. We're going through and helping find different defects in experience for organizations as they go through their customer journey. I think organizations are going to heavily customize a lot of their agents to help them in particular parts. I think that's where it becomes is you have these agent tuners, I guess you could say, out there and agent tuning shops. Which if you look around at who's going to be on the floor, there are a ton of, I would say, systems integrators and global systems integrators that are glad to be down there. I know we talked about one of them was part of their announcement and Agentforce was Accenture and others that are out there. To me, is this moving the bottleneck of agents yet up another level? I don't still think the data part is solved yet. I'm still thinking there's things to solve there, but at the same time I think that how we get to the stack has moved up and the problem is moving up now to how we're going to program these agents and customize them.

George Gilbert

>> Well, I couldn't agree more, although I would say the frontier is moving actually at all the levels. You can't even do agents if you don't have access to the data platform as the system of truth, both the historical one and the real-time one, and it's not even meaningful unless it's harmonized so it's in the same language. Then at the agent development level, you need tools that allow your domain experts to express in their language, their thought processes on how to get a job done. That's how you capture the expertise like you're saying, but the capabilities of the agents behind those tools that capture the expertise, those are improving. Then you need this agent control framework that sort of gives context to the org chart for agents, which is basically think of it as an org chart of microservices that think, and then those are all tied into the outcomes, both the transactional apps and the metrics that measure them.

Rob Strechay

>> And I think too, what we were talking about earlier as well, when you get to below that, I guess you could say below the agent, below the data and you get into that infrastructure and those infrastructure services, they're looking to abstract those for people too. I mean, they go and make the acquisition of Own company tentative to approvals and all of that stuff. It would seem like they're really cementing themselves from that layer up and is that what you were seeing?

Christophe Bertrand

>> Yes, I mean I think literally you think about history and other companies, you mentioned Oracle. Well, that protection of the data component was a big move for them many, many years ago, very similarly. So I think it's a sign of maturing offering and I think they identified in the case of data protection and re-recoverability. So, it's recoverability, it's traceability, it's all sorts of things, but fundamentally it's the ability to recover your data. Everything we've said, nothing happens without the data, period. Let's never forget that.

Rob Strechay

>> Yeah, and I think what's really interesting about what you were just saying there is, and I think this is the interesting part of the platform right now, they're mainly deployed on AWS. Salesforce is, and when you start to look at these other pieces they're bringing into that platform, there's no doubt they'll to other clouds and we know they already are, but we won't go into that. I think one of the interesting parts is that they're building some of the infrastructure services into there so that you can have pro developers with Heroku and some of the infrastructure pieces because they know if their platform has issues, security, governance, data corruption-

Christophe Bertrand

>> Then it's game over, it's game over.

Rob Strechay

>> Then it's game over, so they know they can't. It's interesting where it's almost a trust but verify where they're really building into the infrastructure their own ways of doing all of this stuff because they want to be able to be the ones if the customer comes and points the finger at them and says, "My data is gone, my agents can't transact, you're costing me a billion dollars a day or a million dollars a minute, or whatever the number is." My agents need to be reliable. That's one part where today we're okay with 80% accuracy, 90% accuracy, but we'll get to a point where they're 98.5, 98.7 and stuff like that. How do you see that jiving with the whole platform aspect of it?

George Gilbert

>> Okay, so there's two aspects. There's infrastructure reliability and some amount of that Amazon can build in within region across the availability zones or the software vendor can also build into their software so they can be robust to failures, maybe even at the region level. But then you're also talking about at the agent level, which is agents are probabilistic computing primitives. You're not guaranteed to get the same answer every time you run the same thing, which means if you want to pick a process for the agent, that doesn't require a hundred percent reliability, but what you do want are some high value processes where when you capture more and more the domain expertise and the teachable moments, which are exception conditions, you narrow the distribution so that you're getting higher and higher reliability over time. So it's very different. There's infrastructure reliability and then there's the agent reliability, but also one other thing on abstracting the cloud infrastructure, they are using, as far as I understand, nothing more than Kubernetes. We heard this from a senior exec yesterday on Amazon, which means it's not going to be all that hard should they choose to put some part of the cloud or some of the clouds to another hyperscaler for sovereign reasons or just optionality and price negotiation capabilities.

Rob Strechay

>> In fact, I don't even think they're using the native. They may or may not be using native EKS, for instance in Amazon. Because of the Heroku acquisition and how they do containers under the hood, I would not even be surprised if they're using just EC2 instances and rolling their own there and actually doing that, but they could be using it for load balancers and other pieces. But to your point, that's-

George Gilbert

>> They're just commoditizing the cloud vendor?

Rob Strechay

>> Correct.

George Gilbert

>> You used to build your apps, you used to build your custom apps on the cloud vendor and you used Salesforce for your out of the box apps. But the distinction between those is blurring except that it's not that you're sometimes using the cloud vendor, you're almost always now building your custom apps on the building blocks of Salesforce.

Christophe Bertrand

>> Which is where the value is. The value is whoever runs the business process is going to get the most value as a platform, and that's really the play. It has always been the play. I think that's what we're seeing here. Let's see how it evolves. I think to your point, Rob, there's still a lot of work to do at the infrastructure level, which is absolutely the condition for anything else to really work well, but clearly the dynamic is the train has left the station, let's face it.

Rob Strechay

>> Yep. So how about this final thought?

George Gilbert

>> I guess final thought is there used to be a distinction between dessert toppings and floor waxes, and it's kind of my favorite analogy. Now, which is it? You're getting both platform and application from Salesforce. Now, their exec said yesterday, it's going to take over the next year for this thing to get sort of shaken out, the bugs shaken out. He didn't say bugs, but for it to mature. But this has been a dream for decades for the major application vendors that they become application platforms. Let's see if they can pull it off.

Rob Strechay

>> Yourself?

Christophe Bertrand

>> Well, very much building on this thought is, so who's going to be running this then? Is it the people who are here today like five years from now, or will it be a different profile of professionals and what is that going to do to their ability to keep selling and upselling, et cetera? Because now you're talking about a different persona potentially. Or because of the heritage and really training, are you going to help people evolve and actually to the point I was making earlier, augment their own careers by leveraging AI and by going along with the same journey? To me, that's the interesting part here.

Rob Strechay

>> Yeah, I agree with both of you and I think again, this journey for the next year, I think again, it's ambitious. It's super ambitious. I love the mojo they have going. I love the energy they have here. Again, you start to see, we go to a lot of these conferences and there's just a lot of definitely mojo going on. And I think you know what? This is a great place to leave it, so thank you for joining us live from Dreamforce 2024. Partnered up here with the NYSE, they lent us this beautiful studio to set up in and we got some great people going through. Go back and check them out. Also, check out all the news on siliconangle.com. Thanks to Chuck and Brendan for really supporting us this week. I think again, this really has been awesome and thank you guys for being on board with me and we'll see you guys soon. Take care.

Dreamforce Analysis

search

Rob Strechay

>> Hello and welcome back to Dreamforce 2024 from the NYSE. Our studio is built right up the street from the Moscone Center. We're wrapping things up with day two of Dreamforce, which is actually their day one. There's been a lot going on today, a lot of keynotes, a lot of platform talk. We've been talking a lot about ecosystem. We've heard from a lot of the ecosystem they're building on top of Salesforce as a platform, doing things with AI and agents as well as some of the partners where they're selling in their marketplaces such as IBM and AWS. It's been a fascinating full two days to put it mildly, and we're really excited to break this down. Today, to help me wrap this up, I'm once again joined by Christophe Bertrand, who's one of our principal analysts with theCUBEResearch and George Gilbert, who's one of our principal analysts also with theCUBEResearch. So, guys, and I'll kind of throw this to George first, we've been talking about platforms and data. We had some really great guests from Salesforce on with MK and David, and we're really digging in. What's your takeaway so far about the platform aspect?

George Gilbert

>> Well, I think there's a reason why Marc Benioff said this is the most important Dreamforce ever. This is now 25 years since they founded the company because it's very, very hard to turn a platform company into an application company or vice versa. If they pull this off, they'll be one of the largest to have ever done it and they'll transform the company. That's sort of a cloud-level way of saying, and no pun intended, of saying that all the data that Salesforce has accumulated and help their customers shape and inform. So it's not just the customer 360, but anything that customers choose to connect to the customer 360 that can activate used to be like activate a marketing campaign or nurture a lead. Now it activates an agent and an agent embodies business processes that were too complicated to specify. Now they learn and they can... Now, Salesforce becomes the platform on which others can build agents.

Rob Strechay

>> Yeah, no, I agree. I think one of the things, and Christophe and I have been talking to a lot of the ecosystem who I think ecosystem has been out in front of this a little bit on Salesforce and building on top of the platform because ultimately having used Salesforce myself at multiple different startups, it really is difficult to program. When you programmed into force.com and what have you on top of Salesforce, it became very static and making those changes. So to your point, I think if they can pull this off where it becomes a more dynamic app platform, more personalized, really focused on those processes all the way back to the data, it's going to be a clear win for them and especially where they've abstracted and they're not a cloud. They're sitting on cloud, they're sitting on IaaS, and I think that's a huge win for them is that they are SaaS from that perspective. So, what about yourself, Christophe? What have you been seeing?

Christophe Bertrand

>> Well, I'd like to go a little further than this and say it's not just clouds, a hybrid cloud, it's also Db2 and mainframe. It's working with IBM for example. Great conversation we had with the head of global partnerships on this topic. All of this is also very interesting from the perspective of how nimble it makes the ecosystem. I mean, you're right, a platform, George, is something you build on. That's the definition, right? But it has to be pervasive, and I think by being so open to many data sources, multiple hybrid cloud type of environments, that's going to be the key. Now, the challenge that I see, also for every opportunity there is a challenge or more, will be how do you really keep things compliant? How do you keep things governed? That's really the biggest question I have because the play here is a scale play and with scale comes a lot of responsibility and complexity.

George Gilbert

>> I do have one comment to add on that from previous interviews with MK who's president of the developer platform, which is, we used to build data platforms with role-based access control. Like here, you belong to this role, this role gets access to these tables or columns or whatever. Now, the most advanced data platforms like Salesforce have a policy-based security model where you can put tags on the data and then formulas as to who gets access to data by what tag. I know that's dropping into the weeds, but it's part of the theme that it's not just good enough to have a data lake with a very high performance SQL database anymore. It's the meaning you put on top of it, the policy you put on top of it. To your point, the low-code tools and Salesforce had the low-code tools now by putting the agents in them. You can program those low-code tools either with English or just giving them a goal and they figure out the context from the data and the available actions.

Christophe Bertrand

>> Well, that's going to be key. I will tell you why. Because you're going to have all of these AI workers, for lack of a better term, and how many are you going to have in any environment? It's going to be ridiculous the number of agents that might be running in a large environment as the ecosystem matures and the use cases expand. So, I think it's going to be more than the data tagging. The data level is also going to be managing all those AI workers.

Rob Strechay

>> Yeah, I mean I think building off both what you're both saying is that the fact that these agents are doing business processes, it's doing revenue recognition, it's doing closing of a sales campaign, it's doing a flagging of distressed customers within a service cloud or what have you. I think what they're doing is really trying to enable people. I like Benioff's talk and we heard it from David as well, who's president and chief product officer, David Schmaier. Really, it's about enabling people and I think they weren't getting ahead of their skis with their agents and saying this is going to replace people. Their talk track to it was this is super important to make people more efficient, especially in this market, this economy where you're talking about how if you can find the right people, they're hard to retain. Most of the time you can't find the right people and you can't afford them for that matter. So, I think it's a very interesting dynamic going on in the ecosystem around them where there are some that are saying, "Hey, this is replacement." I mean, data we know is what AI is built on. We've talked about it all this week, we've talked about it weeks before. I think the control of that data and the control of the metadata, it'll be interesting because MuleSoft gives them some advantages and they're zero ETL alliance that they have with AWS, IBM, Databricks, I believe Snowflake's in there and others makes a lot of sense, but the metadata doesn't necessarily travel with that and that context. So, I wonder at what level can they bring and bridge enough of the other data within an organization to really have the ownership and be that platform. That's what I find interesting.

George Gilbert

>> Okay, it's a good point. To that, I would... Oh, I got Siri talking to me. Okay, that's like in the weather reports where the cat hops up on the weatherman. So, I would point out that the idea of this federated query in their data platform, they want to be the historical source of truth. And to do that, some of the data that they can federate out to actually fits into harmonized definitions of the customer, not all of it. So it can gracefully degrade. Actually, the issue of the policy, it applies both to humans and to agents and what we'll likely see is a org chart for agents that complements an org chart for humans. Agents can call on other agents and when they're not confident of a step in a process or an outcome, they can then bounce up to an inbox to a human to supervise them. The point of all this isn't to eliminate humans so much as to capture expertise for the easy jobs and processes. Not full jobs, processes or tasks within jobs, and that you're embodying in this org chart of agents the expertise to perform more and more of the routine functions in an organization, and then it up-levels the humans.

Christophe Bertrand

>> It's interesting what you're mentioning here because I go back to a conversation we had with Own company on backup and recovery and the fact that they literally keep an archive all the data. I mean, they see everything. It's backup and recovery. So, having access to all of this now becomes a way to augment further all of this work that's going to happen at the data level and the platform level, but also a way to go back and really inspect what may be going on, what's happening. So, there are really a couple of bidirectional conversations going on here.

George Gilbert

>> Yeah.

Rob Strechay

>> Yeah, and I think one of the interesting things is, and I think this actually brings it all... We keep going up and down the stack. I think one of the things that was interesting was how they're bringing forward Heroku as well and talking about how the developer life cycle. To your point, I think that people are going to look at AI and agents and they're going to want to be able to control the CI/CD for these agents. They're going to need lineage. They're going to need all of these pieces to be combined. So if you're going to have a platform, you almost have to have that Heroku, which hey, it understands Kubernetes under the hood. It knows how to actually distribute it. It knows how to put the things in the right places and orchestrate that from a Kubernetes perspective. It takes away and simplifies the platform so that they work on developing the agents and connecting the data and again, trying to figure out how is all the lineage work and all of this. That's a whole different story that I didn't hear much about, but I think when you look at it, it's abstracting the infrastructure and I think we briefly talked about that yesterday.

George Gilbert

>> Yeah, your point really shines a light on what they kept trying to say in the keynote, which is don't do it yourself. The whole point about the DevOps that you're talking about, DevOps for an agent is it's not just release cycle, but a continual capture of the exception conditions. The exception conditions become teachable moments, and then you look at those and you say, "Okay, which ones had favorable outcomes? Which ones had unfavorable ones?" Then the idea is that you periodically capture those, you add them either to the prompt database or to the fine-tuning, and then the distribution of outcomes becomes narrower around the best practice for that task. It's a whole different tool chain, but underneath, you still need the data lineage to make sure that the data that the agent is using as context is trustworthy.

Christophe Bertrand

>> Look, it's always going to be about good data. There's no doubt about it. So that's the part, to me, that's the constant. It has been the constant since mainframe. That's number one. I think number two is how do you train agents? How do you train humans to train agents and vice versa? I think we will see in the future how things evolve, but I do believe that while they will be some entry-level jobs, very likely that were not that productive to begin with, that may be eliminated. It's an opportunity for a lot of people to actually go up their own stack, if you will, in terms of professional excellence to be empowered and augmented by agents. I think I can think of 25 or 30 examples a day where I would love to be able to do more by just asking a couple of questions to a smart person when I have to resolve something. So, I think that's one of the promises. Again, I think it's early stages. I haven't seen a lot of metrics on any of this yet, and that is something that bothers me a little bit.

George Gilbert

>> Oh, it's interesting that you should mention metrics because one of the things that we will see right now, metrics are they're isolated dashboards, they're frozen, they're one way, they're like looking in the looking glass. But we're going to find metrics are not just what you measure like human or organizational performance by, but those are going to be the metrics by which agents measure their performance. And so you need to make sure you're choosing the right metrics. They all form a tree. It's not like a Newtonian machine because a little bit of a change here and a little bit of a change there might make somewhat of a change here. It's more of a complex system, but the agents are going to be in their own org chart wired up to this metric of how the organization performs. Now, this sounds like it's a blue sky, but it's actually a journey where you wire up the metrics bit by bit. You get the relationships between them higher fidelity, and then you connect the agents progressively to more and more of the metrics, and that's how you align the agent behavior with the goals of the organization.

Rob Strechay

>> Yeah, I think just building off of that, I thought what was really interesting was how they talked about how people are going to actually add to agents. So, customization of the agents and adding different ways of doing tasks, or to your point about, hey, what are the anomalies that I see in the agent behavior and become a teachable moment for the agent is going to differ company to company. I think this was the thing that I sat there and it was interesting to see and just to compare this towards last week when we were talking about Oracle releasing 51 different agents, which is still 30 some odd more than Salesforce announced this week, but is it around volume or is it about efficiency? Are we going to get agent to death? We were talking to some of the organizations that were coming through and one of the interesting theories was last year it was all about co-pilots, this year it's all about agents. I think people are still grasping at the ROI of AI and trying to see, okay, what work will get achieved by these agents? Now that used to be co-pilots that were very broad, now agents are very task-oriented and focused. What are you seeing when you were talking to other people as well and floating around? Are people looking to use the packaged agents or are they looking that their... They feel like they're snowflakes and that they have... and not the company, but you know?

George Gilbert

>> No, no, that was a good analogy. I think what's going on is that you need to show concrete solution oriented value to get in the door. So there's at least one agent atleast in each of the clouds, whether it's marketing cloud, commerce cloud, revenue and order service, there's got to be at least one agent. Usually there's a couple in each of the clouds. But the critical one then is the agent development toolkit where it's now... With low code it was already a revolution in accessibility to corporate or citizen developers to adapt these clouds to their processes. But now with agents, they can go in order of magnitude, two order of magnitudes farther in terms of usability and crucially in the types of processes they can accommodate. Because when you programmed with low code before or any symbolic programming language, you were basically saying, "I know the answer, and it follows these sequence of steps." With agents, you get to the other 80% where you're saying, "Well, let me just show you some examples." We call it the dark matter of processes where you show it some examples and you give it a goal and it sort of figures out how to do it, but the building blocks are in the applications. That's what Salesforce has, perhaps Oracle, that just a pure data platform does not have.

Christophe Bertrand

>> I'm going to take this back to the business level, kind of a notch up from the technology here. When I'm hearing this, I'm immediately thinking, "Well, if I want to differentiate my business, whatever business I'm in, I'm going to invest in the best agent people who can understand agents, develop them, integrate them, because that's where competitive differentiation will end up happening." I'll be more efficient as a business, and in a sense, it's sort of giving power back to humans in many ways that way. But I do believe that's a fundamental in time, fundamental business move to invest in this type of technology and specifically the agent components at the process level, at the outcome level.

Rob Strechay

>> Yeah, I think actually that's a great segue into the discussion we did have with David around this, Schmaier, and I think part of it was when he brought up that we have these PhDs out of Stanford and all of these people and we have that group and doing R&D. It actually reminded me when you and I were sitting with Ali Ghodsi in Databricks, it was like, "Yeah, we have Mosaic and we have all these really smart people and all these people from Cal." It's almost becoming a university versus university fight from a platform perspective because I think exactly what you're looking for is Salesforce is sitting back and saying, "I don't want to hire..." Most companies don't want to hire Cal, Stanford, MIT, whoever.

George Gilbert

>> They can't.

Rob Strechay

>> Yeah, they can't and shouldn't for that matter. It's not their core IP unless they already have them for doing quant and all of the other ML training and trading. I could see it in the financial services-

George Gilbert

>> The data science.

Rob Strechay

>> Yeah, but if you're a healthcare company, you're an HMO, why would you go and hire all these really expensive... You're not. So you're going to look for something like a platform to your exact point because you can't go find the agent builders. You're going to want to start somewhere with these agents. So when I look at what Salesforce is doing, we had numerous others on... We had some great people talking about how they're helping build agents or building agents for specific things like hey, going through and quoting, configure price and quote. We're going through and helping find different defects in experience for organizations as they go through their customer journey. I think organizations are going to heavily customize a lot of their agents to help them in particular parts. I think that's where it becomes is you have these agent tuners, I guess you could say, out there and agent tuning shops. Which if you look around at who's going to be on the floor, there are a ton of, I would say, systems integrators and global systems integrators that are glad to be down there. I know we talked about one of them was part of their announcement and Agentforce was Accenture and others that are out there. To me, is this moving the bottleneck of agents yet up another level? I don't still think the data part is solved yet. I'm still thinking there's things to solve there, but at the same time I think that how we get to the stack has moved up and the problem is moving up now to how we're going to program these agents and customize them.

George Gilbert

>> Well, I couldn't agree more, although I would say the frontier is moving actually at all the levels. You can't even do agents if you don't have access to the data platform as the system of truth, both the historical one and the real-time one, and it's not even meaningful unless it's harmonized so it's in the same language. Then at the agent development level, you need tools that allow your domain experts to express in their language, their thought processes on how to get a job done. That's how you capture the expertise like you're saying, but the capabilities of the agents behind those tools that capture the expertise, those are improving. Then you need this agent control framework that sort of gives context to the org chart for agents, which is basically think of it as an org chart of microservices that think, and then those are all tied into the outcomes, both the transactional apps and the metrics that measure them.

Rob Strechay

>> And I think too, what we were talking about earlier as well, when you get to below that, I guess you could say below the agent, below the data and you get into that infrastructure and those infrastructure services, they're looking to abstract those for people too. I mean, they go and make the acquisition of Own company tentative to approvals and all of that stuff. It would seem like they're really cementing themselves from that layer up and is that what you were seeing?

Christophe Bertrand

>> Yes, I mean I think literally you think about history and other companies, you mentioned Oracle. Well, that protection of the data component was a big move for them many, many years ago, very similarly. So I think it's a sign of maturing offering and I think they identified in the case of data protection and re-recoverability. So, it's recoverability, it's traceability, it's all sorts of things, but fundamentally it's the ability to recover your data. Everything we've said, nothing happens without the data, period. Let's never forget that.

Rob Strechay

>> Yeah, and I think what's really interesting about what you were just saying there is, and I think this is the interesting part of the platform right now, they're mainly deployed on AWS. Salesforce is, and when you start to look at these other pieces they're bringing into that platform, there's no doubt they'll to other clouds and we know they already are, but we won't go into that. I think one of the interesting parts is that they're building some of the infrastructure services into there so that you can have pro developers with Heroku and some of the infrastructure pieces because they know if their platform has issues, security, governance, data corruption-

Christophe Bertrand

>> Then it's game over, it's game over.

Rob Strechay

>> Then it's game over, so they know they can't. It's interesting where it's almost a trust but verify where they're really building into the infrastructure their own ways of doing all of this stuff because they want to be able to be the ones if the customer comes and points the finger at them and says, "My data is gone, my agents can't transact, you're costing me a billion dollars a day or a million dollars a minute, or whatever the number is." My agents need to be reliable. That's one part where today we're okay with 80% accuracy, 90% accuracy, but we'll get to a point where they're 98.5, 98.7 and stuff like that. How do you see that jiving with the whole platform aspect of it?

George Gilbert

>> Okay, so there's two aspects. There's infrastructure reliability and some amount of that Amazon can build in within region across the availability zones or the software vendor can also build into their software so they can be robust to failures, maybe even at the region level. But then you're also talking about at the agent level, which is agents are probabilistic computing primitives. You're not guaranteed to get the same answer every time you run the same thing, which means if you want to pick a process for the agent, that doesn't require a hundred percent reliability, but what you do want are some high value processes where when you capture more and more the domain expertise and the teachable moments, which are exception conditions, you narrow the distribution so that you're getting higher and higher reliability over time. So it's very different. There's infrastructure reliability and then there's the agent reliability, but also one other thing on abstracting the cloud infrastructure, they are using, as far as I understand, nothing more than Kubernetes. We heard this from a senior exec yesterday on Amazon, which means it's not going to be all that hard should they choose to put some part of the cloud or some of the clouds to another hyperscaler for sovereign reasons or just optionality and price negotiation capabilities.

Rob Strechay

>> In fact, I don't even think they're using the native. They may or may not be using native EKS, for instance in Amazon. Because of the Heroku acquisition and how they do containers under the hood, I would not even be surprised if they're using just EC2 instances and rolling their own there and actually doing that, but they could be using it for load balancers and other pieces. But to your point, that's-

George Gilbert

>> They're just commoditizing the cloud vendor?

Rob Strechay

>> Correct.

George Gilbert

>> You used to build your apps, you used to build your custom apps on the cloud vendor and you used Salesforce for your out of the box apps. But the distinction between those is blurring except that it's not that you're sometimes using the cloud vendor, you're almost always now building your custom apps on the building blocks of Salesforce.

Christophe Bertrand

>> Which is where the value is. The value is whoever runs the business process is going to get the most value as a platform, and that's really the play. It has always been the play. I think that's what we're seeing here. Let's see how it evolves. I think to your point, Rob, there's still a lot of work to do at the infrastructure level, which is absolutely the condition for anything else to really work well, but clearly the dynamic is the train has left the station, let's face it.

Rob Strechay

>> Yep. So how about this final thought?

George Gilbert

>> I guess final thought is there used to be a distinction between dessert toppings and floor waxes, and it's kind of my favorite analogy. Now, which is it? You're getting both platform and application from Salesforce. Now, their exec said yesterday, it's going to take over the next year for this thing to get sort of shaken out, the bugs shaken out. He didn't say bugs, but for it to mature. But this has been a dream for decades for the major application vendors that they become application platforms. Let's see if they can pull it off.

Rob Strechay

>> Yourself?

Christophe Bertrand

>> Well, very much building on this thought is, so who's going to be running this then? Is it the people who are here today like five years from now, or will it be a different profile of professionals and what is that going to do to their ability to keep selling and upselling, et cetera? Because now you're talking about a different persona potentially. Or because of the heritage and really training, are you going to help people evolve and actually to the point I was making earlier, augment their own careers by leveraging AI and by going along with the same journey? To me, that's the interesting part here.

Rob Strechay

>> Yeah, I agree with both of you and I think again, this journey for the next year, I think again, it's ambitious. It's super ambitious. I love the mojo they have going. I love the energy they have here. Again, you start to see, we go to a lot of these conferences and there's just a lot of definitely mojo going on. And I think you know what? This is a great place to leave it, so thank you for joining us live from Dreamforce 2024. Partnered up here with the NYSE, they lent us this beautiful studio to set up in and we got some great people going through. Go back and check them out. Also, check out all the news on siliconangle.com. Thanks to Chuck and Brendan for really supporting us this week. I think again, this really has been awesome and thank you guys for being on board with me and we'll see you guys soon. Take care.