327 | Breaking Analysis | Salesforce After Dreamforce - How $CRM can grow beyond its own interface

This is Breaking Analysis with Dave Volante. >> Salesforce's next growth opportunity may come from customers spending less time in its interface and having agents do more of the work. Coming out of Dreamforce 2026, we believe that is the shift that is worth exploring. We've seen the progression from command lines to graphical interfaces to browsers and mobile and that's now entering another phase. An agent can generate an interface around the task inside Claude, Slack, or another client. The customer no longer has to start with the application's screen. The starting point becomes a customer outcome. But generating an interface, well, that's not the same as understanding the business. This is where our system of intelligence SOI framework comes into play. The system of enga of engagement S O S O S O S O S O S O S O S O S O S O S OE connects people and agents. The system of intelligence supplies the business context and the system of agency the S SOA turns that understanding into action. Salesforce's larger ambition is to connect these elements through what it calls an enterprise AI harness. the data, the business knowledge, workflows, and controls that let models do useful work across systems. This goes well beyond answering questions. The customer evidence from our good friends at Qualitate suggests an exciting opportunity for Salesforce. But the data also shows a company in transition. Qualitate's latest channel checks find that most organizations interviewed remain in agent force pilot or proofs of concept. Yet among customers who have modeled or experienced the financial impact of headless access, 75% expect Salesforce spending to increase, not decline. Working outside Salesforce's interface could actually mean doing more work on its platform. So that creates both an opportunity and a potential challenge. More consumption can generate more revenue, but it does not automatically generate more customer value. Pricing must make sense as these pilots scale. And if customers increasingly work through their preferred AI environment, Salesforce has to demonstrate why its business context and execution capabilities remain an essential underpinning of the customer experience. The point is the interface can move elsewhere, meaning the next purchase may or may not stay within Salesforce. Welcome. Welcome to breaking analysis number 327. Salesforce after Dreamforce. How CRM the company can grow beyond its own interface. In this episode, George Gilbert joins us to map Salesforce's postdreamforce direction into our system of intelligence and new AI stack framework. We combined his firsthand observations from Dreamforce with new qualitate customer research to examine headless access, agent force maturity, competition from AI native agents and the changing relationship between pricing and business outcomes. So today we explore not only the question whether or not Salesforce can sell more AI, we think it will. The real issue is can Salesforce grow beyond its own interface by becoming the system that makes AI most useful to the business. George Gilbert once again welcome. >> Good to see you Dave. >> You were at Dreamforce deep inside the analyst program. Um you also did some cube hosting. So we're going to pick your brain on that. You always come out of these with a dot connecting um sequence. So excited about that. Eugene, if you'd bring up the first slide, this is our our our AI stack framework. One that if you followed this program, you've seen many, many times. You know, George, what we like to do, we've used this for Snowflake, we've used this for data brick, we we introduced this, I think now a couple of years ago, and we've evolved it. Um, nothing has really changed here. What's changed is we're now seeing that this framework that you developed along with Jeffrey Moore and adapted to the new AI stack. We're now starting to see the software vendor community start to roll out products that fit into this framework. So take us through briefly, you know, remind us, you know, how we should think about this framework and then we'll dig in to u how Salesforce maps to it and then bring in the customer data. >> Okay. So at the bottom left where it's between brown and orange that's where we started this uh building this framework which is that's the analytic data platform the snowflakes and data bricks of of the world. Um and that's where people have um customers have historically tried to aggregate their data. But despite the fact that they could put all their all their operational data, transform it and put it into one analytic data platform, it's still silos because the the cubes the essentially the way they model the data so that they can slice and dice it. um is aggregated and each each each cube is itself a silo that's on on the left side on the right side on the bottom are the traditional systems of record whether it's Oracle or SAP or Netswuite um or Salesforce itself um and the layer above each governance metadata those are the cataloges that uh allow the data to start making sense to both humans and agents to start making sense The governance metadata is usually um people, places, and things. It's not column names and table names. It's it's now human and agent readable. But to go to the next layer above um where it's in green, this is this is the core. We keep saying it's the most important piece of real estate and enterprise software for the next 15 years because this is what connects um general purpose intelligence in the form of the LLM or the agent into um how your business works, the physics of your business and how the that that model of your business will improve over time. at the at the most basic level it's metrics and dimensions that's what we know from BI this is you know a metric like revenue or churn or average recurring revenue but when you get to the full system of intelligence these are the processes um like how do you do quote to cash or how do you onboard a new employee um and the system of agency on top plugs into to that system of intelligence to do work. The system of agency reads that map to understand the state of the business, reasons about it, makes a plan and then operationalizes it through the system of intelligence. The system of intelligence, the workflows then become tools essentially. And um we'll come back to this theme, but for the first time we saw a vendor connect the well-known concept of an agent harness, which is usually like a a standalone piece of software and that system of intelligence, which is what um for Salesforce is data 360, agent fabric from Mulesoft um and the application 360 semantics. They're now calling that their agent harness. And what they're saying is you take whatever model you want and we make it useful and we allow it to compound in in in value over time. >> Got it. Okay. Um, let me take a a beat here. I want to reintroduce qualitate to our audience. A couple weeks ago, prior to the CrowdStrike event, uh, Falcon, we gave you a little taste of of Qualitate. Qualitate for those of you who didn't see it is a firm started by Sara Kadakia. Sara was uh one of the early uh data scientists at ETR uh and form qualitated about two or three years ago. Actually I think it was three years ago. The platform is is unbelievable. Uh essentially it's an agentic platform that that builds and leverages an expert network. So, so Qualitate has, you know, a network of experts and rather than going out and doing multiple choice surveys. Qualitate actually has built agents to actually do the surveys, the human speaks in human language. It answers the questions. Qualitate captures those and then it what George and I call schema on read. So in other words, it dumps all this data instead of saying, you know, choice one, choice two, choice three, choice four, multiple choice fit you in a box which may not, you know, be a perfect fit for you. It just asks you questions, answers those questions. Qualitate has so much scale that it can do those those multiple choices. It can concatenate them and categorize them, you know, with AI. Okay. So it's that's why we call it schema on read. So qualitate just last week dropped uh along with Salesforce Dreamforce a detailed study um from customer interviews with Salesforce customers. And this one in particular um was 20 participants. So they're in-depth interviews. They're half hour or hourong interviews um that they conducted over five business days. And and I'll just give you just a quick you know quick run through. Um, agent force still early, you know, a lot of adoption, but 75% as we'll show you of the respondents said they're interested, the only 25% uninterested. We'll talk a little bit about flex credits and pricing that sort of attracted more folks. Um, there's no evidence yet anyway of agent force adoption detracting from the Salesforce budget. Headless has universal appeal. We're going to talk about that a lot. and uh and and of the 20 again the end is not huge but it's sort of detailed surveys organizations were allocating between five and 15% of Salesforce spend to agent force um so Eugene if you bring up the next slide um this is essentially a clawed client using you know accessing through Salesforce data through headless 360 and building and generating the interface Okay. So in the upper right here you see we've gone from command line interfaces to gueies to the browser to mobile and now this is aentic UI on the fly and in the right hand side you can see the qualitate data. So so George take us take us through how claude is integrated into Salesforce and how headless works and then we'll dig in a little bit to the data. >> Okay. So let me briefly talk about the context is headless 360 is Salesforce without a user interface where it's basically it's exposed Salesforce all Salesforce functionality through MCP servers APIs um and then a set of plugins which is a collection of skills within uh so far uh claude and that's called claude force um and then slacks slack's own slackbot Um so that each knows how to navigate Salesforce this as we were describing it as a system of intelligence. Um and then this is the really important part. It uses the the capabilities of a coding agent to understand the data the data that's now um has has all the context and definitions around it from Salesforce and it generates a user interface specific to the data and the and the functions that you want to use. So the all the problems of all the years of trying to simplify the Salesforce user interface so you know to attract more users or so that it just exposed the functionality they need. This is now essentially generated on demand. So it's really a generative user interface um with a coding agent behind the scenes. And so far it's Slackbot uh Slack with a Slackbot generating a UI Claude. Um and then in the Lightning user interface for Salesforce it has coworker. Um, and it's just it it's probably the most immediately relevant and useful and impactful announcement from the entire conference because all of a sudden now Salesforce is a semantically harmonized data repository that can unify all your customer related data and serve it up in whatever context um a user needs. And as we'll see also later, um that same headless 360 can serve agents. In this case, it's serving a coding agent that's presenting it to a user. George, you were if not the first, one of the first to really call to our attention and our audience a couple years ago now of the work that Salesforce was doing. The prevailing narrative at the time of Salesforce, they made all these acquisitions Tableau and Slack etc. and they were just a collection of assets and you were really the first to sort of uncover the deep engineering work that they were actually doing. Uh I think it started around um the data cloud I think they called it and then they turned that into data 360 and that and and that is now evolving into what we saw this week at at Dreamforce. Uh and so the point is Salesforce is doing the engineering work. um they're now of course you know integrating with with Anthropic in particular and they have gone from this sort of collection of assets to now really in the mix in this agentic era. Eugene if you bring that same slide up if you would uh let's bring in some of the qualitate data and dig into it. So you can see 90% are of the of the organizations again small end but still it was universal are already you know outside the Salesforce UI or they're planning to be um they're using you know other agents whether it's it's it's claude code or codeex or you know even things like maybe perhaps cursor I mean expect George has a great saying there's more agents than there are cle on a camel's back uh so there's there's a crowded market out Yeah, the other interesting data point was 75% of those those customers actually modeled out the impact of using headless thought that it would increase their spend over time. And we're going to come back and talk around this and and you you see that developers they have an affinity you know whether it's Claude or Codeex or Cursor or whatever you know whatever you know harness or coding agent they like they've consolidated around a preferred LLM which in in some cases could be inside of of Salesforce. So the you can see the points here from the actual verbatims claude affinity is real. Our developers are pretty standardized on cloud code. Okay, so that was cited. But hey, there's we're a Microsoft shop. We're using a lot of GitHub copilot. So, so the headless 360 really takes Salesforce from an application UI into an enentic environment with a new surface, you know, but it also makes, as we say here, the preferred client, you know, very very strategic here. George, any any thoughts on the data? Yeah, the one the one um nuance I would add is that this is an a coding agent that is generating using code a user interface to present to a human to interact with. As we'll talk about later, that same headless uh 360 meaning Salesforce without an AI is also presenting its data and workflows for an agent to take action with. So here we're looking at the system of engagement side engaging with a human but later we'll talk about the system of agency where it's an agent taking action potentially with a human in the loop. Um so one is um custom UI generated by an agent to present to a human. The other is um the data and workflows as context to ground the work of an agent operating on its own. And I tell you, George, this really resonates with me. U I was onboarded to the Qualitate platform on Monday. I've already launched two surveys. So I'm speaking to the platform. I'm speaking to the data. I'm I'm I'm generating surveys. I mean, I'm launching surveys by talking to the platform and then and then collaborating with the platform, structuring a survey, pushing a button, and then off it goes. I'm already here it is Friday. I'm already I'm doing a survey around service now control tower customers and another one in Coreweave because we're going to be a coreweee event in a couple weeks. So, I want to see what the results are. I'm already I've already got results. H have way through the the the the 15 and 15. So I'm doing 30 on each 20 15 on each already halfway through with the results. Absolutely transformative experience. So now Eugene bring up the next slide. This is an example where Slack is a is a collaboration uh system of engagement. It's headless Salesforce with Slack actually generating the UI. Explain this George. >> Okay. So the importance here is when in the last example where you're interacting with Claude, it's it's basically a solo user um interacting with the application and they might generate potentially an artifact to share but claude is not a collaboration environment. Slack is a collaboration environment. So here, same idea where in Slack um in this case, Slackbot generates a custom UI to interact with the Salesforce data and and Salesforce actions, Salesforce workflows, but only here Slack is a surface where you inter interact with your colleagues whether within within the enterprise or or across enterprises and you interact with other agents. So this is um this is a work surface for humans and agents collaborating together but same principle headless 360 in the background and a coding agent generating the UI and it'll it it should make a huge difference in the accessibility of of Salesforce to potentially new users. So in the upper right is the qualitate data. Same three data points, the 90, the 75, and the seven of eight that we just showed you on the the claude uh example previously. Down below you can see the Slack specific data. As our reps start to use Slack or some of of the other tools really outside of the Salesforce UI, the expectation is, you know, more transition volume and possible API consumption. And we we'll talk about that later in terms of what that means for spend and pricing and and so slack code which was announced at reinforce last week if the the qual data is early not mainstream I I looked at this George and I was like wow so say 60% had not heard of slack code I was shocked that 40% had heard of it so it was actually pretty well you knowh exposed >> this was the the stunning stunning thing was so slack code might have been announced before the environment but the Slack surface where it generates the UI that was just announced like Wednesday Thursday >> right and and 40% and I mean we're sort of aligned to it or sort of were exposed to it but that was really the exciting announcement wasn't it >> yeah yeah I mean it was it was really exciting because essentially now Slack is a UI builder for um essentially for for any enterprise data right now they've got the skills in there to navigate um the headless 360 but you can see it become a UI for for other applications a UI and collaboration environment for other applications as well and as we'll come back to I just want to put a pin in the so people um make a note a mental note that right now it's just generating a Salesforce a UI to Salesforce data but when you have MCP servers and APIs to other applications it can generate a UI that maps not just Salesforce data but SAP or Netswuite or anything else for that matter. Now, it's going to take some work to harmonize that data where the user at the end has to be the one to bridge it. But the point is we're we're in an era where the user interface to your backends is generated and that's the first step in essentially bridging all these silos that we've had. And and just a point of of that I want to pull point out from the qualitate just reading through the qualitate study a number of customers indicated that that was really attractive to them. There was another it was almost pretty bifurcated. There was another group that was like well governance is a big thing here. Uh we want to see some proof. You know MCP servers are just not fully baked yet. And so you have that you know some customers are comfortable with it. Others you know may not be probably the regulated industries want to want to see more. Eugene, bring up that same slide if you would. I want to double click on some of this other qualitate data down below. Um, so we talked about 60% had not heard of it. That means 40% had N equals 7 report no interest due to entrenched alternatives you know other sort of collaboration tools and then there was one point about you know um monitoring with some developer interest but the the most of these interviews did not go directly to development but it was you know through maybe perhaps a business person or an IT decision maker. quote, "Our developers are pretty standardized on cloud code and then we're a Microsoft shop. That's what we showed you before." Uh, thank you, Eugene. Okay. Um, let's move on uh to the stack that Benoff showed in his keynote. Eugene, if you bring up the next slide, uh, this is a slide he that that Mark showed. Uh, George, you're you're you're pointing from sort of the green area of the system of intelligence. This red box is essentially the SOI is is Salesforce's SOI. Something that we've as I said before we've you you called two or three years ago even you could see this all coming together. So explain the layers of this slack and this stack and then we'll get into it. >> Okay. So what's important here is that in the age of agents, humans and agents both have to be able to read the data and understand what it means. That's the state of the business and to be able to reason across the possible actions and their outcomes and then to take action. And we've all been hearing about the incredible pace of development with agents really starting with coding agents. And the core elements that people have focused on with coding agents is the harness. And a critical part of the harness is >> memory and and context. um where for a coding agent it's you know the the the the memory and the context are primarily the code base that you're working on. Um and then the tools would be like you know how to read and write um to the file system to edit part of the codebase and I I the reason I bring up these coding agents is that's where most harness development has been focused. What what Salesforce was trying to emphasize this time for the for the first time was we need to rethink what a harness is when you want the agent to do work in an enterprise application. Here memory is now data 360 which is the state of the business. Th this is all your CRM data that tells you where customers are in the you know what their profile is where they are in the in the sales process or or service process. But then the customer 360 these every harness has tools but here a tool is well I want to do a refund or you know I want to move uh I want I want to activate this customer segment um with this type of campaign. That's they're they're trying to make the point that um our in our you know our as you pointed out Dave that there are more agent development kits than there are fleas on the average camel. What distinguishes an agent development kit is its harness and here it's you have the grounded enterprise data and the enterprise workflows. That's what makes agent force useful. Yeah. Now um actually Eugene please bring that slide back up explicit on this slide is not anything around data bricks and snowflake sort of the poster children for for data platforms but from the customer data from looking through just interacting with with qualitate and looking at some of the customer surveys a couple of things you know came out. I remember I tell the story often that Tony Bear and I were talking. I think we were at a Salesforce, sorry, a a Snowflake conference and he said he felt like that that Salesforce needed to acquire a data platform. And I remember sort of talking to him about that. I we talked about this and may maybe that's not the right thing for them. They're better off partnering. Kramer had Shredar on yesterday on his Mad Money and and asked him about that. You know, are you competitors? You compliment each other. and Srio said hey we've been working together for a long time and you know we're partners there's obviously going to be some overlap in the van the point is and George you saw this as well in the qualitate data customers wanted to have composable data platform capabilities meaning if they're a datab bricks customer or a snowflake customer they recognize that those companies frankly do what they do better than than Salesforce outside of Salesforce and maybe even sometimes within Salesforce and so they have an affinity toward those platforms so they want composability. Explain that George. Okay, it's a really critical point and it's you know the qualitate data it's it's really credible data because what we heard um at the conference was customers and and even before customers would say look I've standardized on data bricks andor snowflake like one or the other or both and then they want a customer data platform that layers over that. Now the irony is um data 360 is it's a data platform that is really a a data a customer process model that should layer on top of your data bricks or snowflake environment because it does this it has this zero copy capability where you just point it at the relevant um data that's in snowflake or data bricks and then when you issue data 360 queries it dynamically goes goes out to Snowflake and data bricks and brings that data back or it caches it locally. In other words, it is a layer above them, a a value ad layer above them. But I think the there's an issue with Salesforce where they're trying to make it really simple to sort of enable data 360 in customers who have these existing environments, but they're also having to earn credibility with the technology side of the house which has historically been responsible for the data bricks and the snowflakes of the world. whereas Salesforce has been um really in entrrenched on the business side of the house. And so that composability means um customers need to perceive this as a um a a set of components that work better together if you layer them on your existing technology platforms on the technology side of the house. And they're still working on that. They're still working on positioning it and they're still working on making it easy to get that time to value because as we're going to talk about when we get to system of agency, you really need data 360 installed before you can get any value out of agent force and and so that's a bit of a limiting factor. In other words, >> it's not a prerequisite, however, right? But but it is if you it Oh, it is. Okay. >> It is. And now there's a a matter of you know how much you know you need in in place but that's that's been one of the limiting factors with agent force which is it looks to data 360 for its context and so that means you have to the the TAM at any one time for agent force is whoever has data 360 involved uh installed. So depending on to what degree you know you want to take advantage of that capability and we saw this in the some of the qualitate data it can be a heavier lift um and and and as well we saw that there was definitely some hallucination issues >> like any LLM and that's going to get better over time. Um okay our dear colleague David Floyer did some of his best work in in retirement. We we do miss him. Um and one of the things he wrote uh was when we talk about bringing deterministic and stochastic or probabilistic software together, it's it's important because you've got you know one that is sort of estimating based on probabilities and one that is you know very high uh highly structured and deterministic outcome. Okay, you bring those two together and one of the points that Floyer made was determinism in many organizations if you think about it is actually illusory. What does he mean by that? Well, you might have determinism in your HR system. You might have determinism in your CRM. You might have determinism in your in your your logistics system, in your financial system. Try bringing those all together and see if you have determinism. So, Eugene, bring up this next slide. You know, this is where the enterprise harness uh and the system of intelligence comes in and harmonizing all that data across different departments, different applications, injecting the process knowledge and different workflows and that really is all about breaking down the silos. George, take us through, you know, the intent here. Okay. So, this is a organizational view of how we're trying to break down the walls between what were different departments. um that you know formerly you had people and their processes specialized into functions or depart you know or departments and as as you were saying David Floyer had this brilliant analysis that no matter how deterministic your processes were you know procure to pay order to cash um the fact that they were bridged you might have had you might have those hardcoded into applications But they were bridged by people. That's that's how you bridge those silos before. And that made it um non-deterministic because people make a point of saying agents are non-deterministic. We should we should actually move to the next slide because this is where we make the point. If you look at the um upper left, a lot of people say, "Well, why can't my agent just talk through an MCP server or through a direct API to all my applications?" And you know it goes to town across those because the a an agent needs even more explicitly than a human a way like a a harmonized map to tell it you know that customer 103 over there in one system is the same as acme in another other system. And the way you calculate um revenue in one system is needs to map to how it's done in another. And so you you need a layer above all those silos and because as we were saying the silos today give you essentially non-deterministic out outcomes because you've got people bridging them and to the extent you can start putting a harness over it which is it's the first time we've seen a vendor now refer to a harness not as just a single agent you know with memory and tools but it's how does an enterprise work you know the state of the enterprise and the workflows. Uh that's that's the point here because now you start getting um deterministic bridges. Now the deterministic bridges in the harness, they don't dictate everything that must that needs to happen, but they put guard rails around what an agent must do or must not do and controls. That's what the new enterprise harness looks like and acts like. >> And I just I'd like to get comments on this. When you talk to practitioners who are actually deploying uh AI, they will tell you AI is really good at doing, you know, mundane data entry types of tasks. It's getting really good at that. It's when you get to those higher levels of agency and and workflows that it becomes more challenging. And you know, based on your previous comments, it seems like this is where Salesforce, we've talked in the past, should have an advantage because it's got that underlying application logic and process knowledge. AI thus far has not been that great in in in delivering on that promise. Would you agree? >> Yeah. You know, and it's interesting part of the we we would have expected or I expected Salesforce to be among the the first big winners uh with agents and the irony is even though like the platform has all the sort of harmonized semantics for you know how your customer data um is defined and and the processes by which you move by which it moves through your organization. The irony was there's still there were lots of gaps in the flow definitions, the metadata that defined how things worked. And those gaps were okay when it was a human looking at a screen because they could bridge this the gaps essentially, but it took a lot more work to clean that all up so that there was no ambiguity for an agent to make it. And that's why I want to tie it back to what we were talking about earlier with the clawed Cloud Force UI and the Slackbot um Slack surface. It's much easier to generate a UI for a single person on a small set of back-end functions than it is for an agent to go to town across all these workflows. That's why I think the generative user interface is a much bigger near-term um is much of a much bigger near-term relevance to Salesforce customers than you know um endto-end workflows and their outcomes executed by agents. >> So to pick up on that and and just to sort of refresh everybody's memory, I mean we have been talking about this notion of a system of intelligence for for quite some time. the importance of harmonizing data. Not only harmonizing data but also uh the the tacet knowledge of the enterprise through workflows and potentially coming up with new ways to work. We've also, you know, this year really started to highlight the importance between the emerging new client surface, whether it's Claude, Co-work, uh, Coco, Datab Bricks, Genie, whatever that that user surface is, and the backend system of intelligence, that closed loop where the back end learns from the front end, learns from the reasoning trace of humans on acce on exception and then and only then can you surface and trust if you a governed you know data set that has that tacid knowledge and then you can take confident action. So if you bring up the next slide Eugene this is really you know we're sort of moving up the stack here and we're kind of using one of the Salesforce slides. I got to say George they I mean Mark gives amazing keynotes. Um yeah he's very performative. I I love him. I I think they're great keynotes. A lot of people maybe may think it is too performative, but I thought he was fabulous. Great marketing. Uh and here you see the agent force, you know, up the stack and and then AI force above that. The point I want to make and then have you pick up on it is there's a lot of agent builders out there. We saw from the qualitate data people are using whether it's bedrock or other agent builders. Um but the idea is you got to build those agents. they've got to be governed, you know, and and then you've got to have an agent control framework that's trusted. And that's basically what Salesforce is talking about and and kind of showing here with this little robot and you know, one of its um you know, one of its uh perhaps trailblazers. Take us through the system of agency, the S SOA, how that relates to what you heard at Dreamforce. Okay. So the the what's shown on the slide is a series of role specific agents. Piper which is inbound um uh pipeline generation Hunter outbound sales. KC help which is for service cloud um and then page which is for ITSM and H and HR. In other words, this is the whole goal is faster time to value so that you don't have to go into agent force and custom build each one. It's more configure and make sure you have enough uh of the backend knowledge and maybe tune what each agent does. But there's other important things that are going on now um that for the first time um there's more model choice. um you can of course uh choose whether you're using claude or open AAI and they had both Dario and Sam at the keynote but they also introduced their own reasoning model um and this was kind of interesting because this was they took um an Nvidia open weight model and they postrained it on Salesforce so it knows how to reason through and interact with a Salesforce environment. This gets back to the point that if you take a midsize model and you post-train it on an environment, it is much more uh costefficient and performant in that environment. And so this is the whole idea that you're going to see a family of models. For the most the hardest reasoning, you'll go to the frontier, but for everyday work, you now have a Salesforce, you know, native model. Um and then and then as you alluded to there's a but what's not on the slide there's a whole uh governance fabric that's coming into place um for discovering orchestrating um governing and observing and observing is is critical here because with agents they're always going to be learning. They're going to learn from the outcomes of their actions. And so you go back into the their um the agent traces, the reasoning traces, the tool calls they made and then that's a a new type of analytics because you look at which traces led to successful outcomes and then you improve the agent whether it's in uh prompting it better um constructing better context or eventually fine-tuning it itself. But this is going to be critical because you're you're tracking tracking this observability data and that's going to be the breadcrumbs like in the in the consumer online era user clicks fed all the the you know the Googles and the Facebooks to make the matching engines work better the recommendation engines. Now your agent traces are going to be what make your agents smarter. >> All right, good. Let's press on. I know you got a call as do I. Just a quick note is that that was the Neotron model that Nvidia has has launched. They're really trying to help us shore up its open source position. It seems to be everywhere. My sources indicate the folks have used it. It's a little bit behind still, but knowing Nvidia, they'll catch up. Let's go on to the next slide here. Uh Eugene, the Finn acquisition. Explain the Finn acquisition and how it's it's relevant. It's the it's the alternative to to to Sierra. We we've we've talked about Sierra. Um uh but but take us through this. >> So this is the other side of that discussion where you have this agent this enterprise agent harness which we talked about as data 360 customer 360. Um but the the uh the downside was that you had to install data 360 and for some customers that was too heavy a lift to get started with agent force. And so, as we were saying, the agent force TAM is really the data 360 install base. And so, for all the other customers who want to get started with something quickly with a a a PC or a pilot and all they need is access to a couple Salesforce objects and some limited data, whether from Salesforce or from data bricks or Snowflake, Finn is their answer. So now you don't need to install that whole enterprise harness. This the answer here is you can get up and running really quickly like with a Sierra or Decagon. Um and it's really I I I spent some time at the booth. It's really just a bunch of scripts, a natural language development environment and then just easy connections. Um just like you would be in Chat GPT or Claude connecting to some tool same way you connect to a few API endpoints. And so it's it's a very compelling offering for people who say, "I want to get something up and running with my Salesforce environment, but I don't want the heavy lift of putting the whole platform in place." >> Yeah. And of course, Brett Taylor, ironically was uh was at Salesforce for a while. He was the chief product officer. he was the COO and eventually became the co-CEO and then co-founded founded uh Sierra uh and of course is uh I think uh uh obviously you know prominent within OpenAI. Um let's let's bring up the next slide. Let's talk about pricing models. pricing models. I I I made the point, George, when we were prepping for this is software companies, SAS companies, they're always really good at preserving their revenue. Whether you, you know, when you go from core counts and cloud onrem to SAS, seat pricing, consumption, they always figure out ways to make sure that they preserve their their value. What we're showing here is a slide from BCG um which is a maturity model and Benov talked about this on on I think it was to Kramer or maybe he did it in his his keynote talked about all the different pricing models that they're experimenting with seat pricing consumption pricing outcome pricing he even talked about gain sharing what I call gain sharing where they'll take a piece of the the benefit um Salesforce has enterprisewide license agreements ELA's They they make it sound in their marketing. They may make it sound like an all you can eat. It's not really there's restrictions on there that procurement will run into. But George, take us through this and then we'll close with the the qualitate data. >> Okay. So, really quickly, I haven't actually been able to find in one place all the different sort of contract contract pricing options. Um but basically we're still more on the um predominantly on the left side of this chart where usage based in terms of resources this is um how many this is like token consumption agent-based is how many agents do you deploy and it's it's like a proxy for seats um usage based in terms of interactions is um how many conversations did I have and by the way that that at one point was the original agent force pricing outcomebased is, you know, how many customer uh help requests did I resolve? Um, and then outcome based financial pricing is capturing some of the upside. That's it's almost like valuebased pricing. We're sharing in the value created. And it's the the more to the right you get the harder it is because we don't have all the measurement capability in place and agreement even on on the or desire to share um financial outcomes but Salesforce is trying to be on very u flexible and meet customers where they are and offer uh sort of different options where they where they do have this resource-based meter Ed um agent-based agent-based interactions um and they they actually for the first time at least the first time that I saw they're having um outcome based based on jobs completed in terms of conversations or resolutions which is what for instance Syria offered um originally. I just want to add one thing Dave that to tie this into the business implications that we were talking about earlier with AI force. This is when the agent is consuming your um your data and process uh model. But with with headless 360 this the assumption here is that this is your agent. you're you're charging because it's your agent on your platform. But what this does not reckon with is what is what if you have headless 360 from Salesforce and eventually say headless 360 from SAP and you're building your agent with data bricks or snowflake then it's their agent and what type of pricing are you getting then is that you know you're you're just a data feed one data feed essentially among many that is a business issue. that this does not reckon with >> right and and then you've also got you know startups there was there was certainly some conversation in the qualitate um data around startups and you and I have talked about no schema or schema on read um in the software market which is kind of an interesting concept where you know if your if your system of record is Salesforce and you're deeply embedded into Salesforce it's not likely you're going to go off there but if just using Salesforce for a system of record and that's it. You're not doing your workflows and you're not deeply embedded then maybe you know you can access that data and and and create equivalent functionality. This is where you know the SAS apocalypse is not binary. There's definitely portions of the software industry that are going to going to feel that. Eugene, if you bring that slide back up, I just want to you know point out the the qualitate data. It's it's it's very unclear how this is all going to play out. Flex credits. We've seen the success that uh Crowdstrike has had with flexible credits, you know, flexible u security pricing. Five to 15% of this base. Again, N was 20. Um they they were not stealing um from other Salesforce spend. They were they were it was incremental of spend at around 5 to 15%, that was 18 out of the 20 said that. Uh so nobody was shifting uh budgets. 75% that had modeled it said they expected higher spending as a result of headless. That's something that procurement is going to watch. So we'll see over time if if that if that holds and whether or not that the new AI spending is offset by lower legacy spending and or maybe features that are that are retired and at 44% you know again plan those spending increases. Um thank you Eugene George. We got to leave it there but I'll give you the the final thoughts. We're obviously going to write this up clean up some of these slides. we're all redeyed and uh and and write up our action items uh over this weekend, but please give us your final thoughts. I think Salesforce has told a very compelling story and pulled together the pieces to show where the industry is going that it's not just a raw data platform with some agent building tool that is really only useful for querying disparate databases and then not making sense of that data because it there you don't have um something to define how the data maps across systems. They really have been doing the work over the last several years to put in place a system that not only makes it easy for agents to read the data but to take action across these silos. now they still have work to do with earn with earning credibility with the technology side of the house with customers um and you know to make the product sort of just work in the Steve Jobs sense they clearly aspire to the we're not going to give you a bunch of piece parts we might make it composable but it should just work that's that's the value proposition now they just have to deliver and they have to convince their customers to trust them with that. >> Yep. And we'll see if that translates, you know, into consistent growth. They certainly guided that they're they're going to get out of the 50s, 50 billions into the 60s. Uh but that's but but we'll see what those levels of growth rates that are sustainable. George, thank you so much. I appreciate it. >> Thanks, Dave. >> All right. And thanks to our friends at Qualitate, you know, awesome job. Really timely data. Check out qualitate.io. This is Dave Volante for breaking analysis. We'll see you next time.