The Six Five team discusses Salesforce Dreamforce: Clippy’s Revenge or AI DiWhy?
If you are interested in watching the full episode you can check it out here.
Disclaimer: The Six Five Webcast is for information and entertainment purposes only. Over the course of this webcast, we may talk about companies that are publicly traded and we may even reference that fact and their equity share price, but please do not take anything that we say as a recommendation about what you should do with your investment dollars. We are not investment advisors and we ask that you do not treat us as such.
Transcript:
Daniel Newman: We were in downtown San Francisco. The streets had been swept. The gates were up. It was pretty clean. It was Dreamforce and 45,000 people converged on the city. Marc said some provocative things. He said that Microsoft Copilot is just Clippy. Was he right? What was your takeaways?
Patrick Moorhead: Yeah, so first and foremost, the city was clean. It’s always nice to have San Francisco at least four or five blocks safe and clean. Whether it’s Marc Benioff or Xi from China, they really did a good job. Now, I think my max walk was two blocks. I don’t know what was happening in the tender line, but I never go down there. Now, listen, we all knew what the news was on Monday before we even hopped on the plane because they did their big announcement on Friday. In tech, there’s always the next new huge thing. At least for Salesforce, it was agents, an offering called Agentforce. There’s a lot of forces in the Salesforce product portfolio, services portfolio, and this is literally the next new thing. It really positioned as what AI should be.
Essentially, agents is the transaction layer that gets stuff done. It reminded me… I mean it’s not the first agent to come out. Microsoft has agents, Google has agents. If I look at what ServiceNow is trying to do, particularly for the transaction layer, while they’re not necessarily calling them agents, it’s AI to get stuff done. As we’ve talked about so many times on the show, AI in the spirit of driving efficiency increased revenue and customer stickiness. I thought the demos were very compelling. One thing that I found very provocative is that Marc Benioff said, “Hey, typically, I’ve got this thing locked two weeks out,” and he said, “We were making changes up until the night before.” I was thinking through what could this possibly be? Well, there was a gigantic Microsoft event that preceded it, where they were talking about things like Copilot for sales, things like this. I have to say yes, Dan, because the shade that Marc was throwing at Microsoft and Copilot, I had never seen that before. I think in our press and analyst Q&A, he threw out expletives, maybe four, maybe five times.
Daniel Newman: They say you’re smart if you swear more.
Patrick Moorhead: I adopt that philosophy. I’m not saying I’m smarter than most people, but I do the same effing thing. Anyways, I mean you dropped two even on the lead in. Not using the words, but the shortened-
Daniel Newman: Insinuated.
Patrick Moorhead: Yeah, PG rated ones. But essentially, he talked about Microsoft Copilot being Clippy. There was also a slide up there, that I think it was time to deploy this gigantic number for Azure AI with one financial services company. So, yeah, I think it did affect it. My takeaway is that the success of Copilot is either slowing down Salesforce deals and/or it’s making a financial impact. Dan, I did a little digging on Copilot. In the last 18 months, and this comes right out of Microsoft, Copilot customers grew 60% quarter-over-quarter. The number of people who use Copilot daily at work doubled, and then out of their earnings call, 60% of the Fortune 500 use Copilot. Copilot customers increased more than 60% QOQ, and then I think more importantly, there’s 20,30 customer stories that talk about the impact of Copilot. These weren’t just across the Copilot for productivity and 365. This was Copilot for sales, for customer service, companies like Virgin, PG&E, Amgen, Air India, Cognizant, Vodafone, Honeywell, FNB, and also citations on the time it took to implement. Of course, Microsoft is going to put low implementation days, but 25 days for a specific client to roll it out. So, it’s great to see the competitive barbs going back and forth, and it’s pretty natural. Microsoft has three plays.
It has an Infrastructure AI play, it has a Dynamics 365 AI play, it has a Copilot that again goes across everything, but it has a productivity play, and then with its Copilot toolkit, you can create your own. I think this was the main point which Marc was trying to make, which is don’t DIY your AI. I think that he has a point on that it should be easier, should be a lot easier, particularly if you’re a Salesforce customer already, and you’ve got your data aligned. You’ve done the data magic inside of Data Cloud. So, I see his point, but it’s clear that Microsoft is putting pressure on what he is trying to accomplish. I think my final comment here is on simplicity. I really like the way that Salesforce not only talks and positions on simplicity, even talking about the benefit.
I’m an ex-product and product marketing person.
So, I’m really into value proposition templates, and Marc very clearly spelled out what are the challenges that his specific customers are having, the impact that that’s causing, what’s the widget that he has, and this is Agentforce, that it’s using to solve those problems. What was the benefit to the customer and how is it different, right? Classic product management and product marketing 101, and then a description of how this works, but it’s not a bunch of marketing BS. They’re one of the few SaaS companies that actually publishes papers on their LLMs, where it shows how it was trained, what data was it trained on, and what were the weightings. So, there’s real science behind this, not a bunch of marketing BS and stuffed animals like they love to put on stage, which I find incessantly annoying as an industry analyst, but I totally get how they’ve created this cultish marketing and sales and service operations culture, how people get into this. It’s like their one show of the year and they’re going all in, baby. They put trailblazers up there. You’re in pictures and slides, I get it, but it’s annoying. It annoys me.
Daniel Newman: Yeah. Listen, first and foremost, it was a very interesting inflection. I feel vindicated in a lot of different ways. As we’ve talked incessantly, Pat, about this AI CapEx boom, we’ve continuously gone round and round about, “Where does consumption happen? When do people use AI and how?” So for me, this last second pivot that Marc Benioff had, the change, the all-in on Agentforce, we heard Jensen get up and we went to his little round table, Agentforce. Jensen was having fun, extra spicy this week. But we have a situation where people are trying to figure out how to get value from AI. I’ve said from the very beginning, it’s going to happen in software. Not just in software in terms of the DIY of having a bunch of data scientists and developers building applications using CUDA and writing to GPUs and creating custom. That stuff is happening, but for most businesses, they want the Copilot experience. This is where the debate’s coming in, is Marc’s talking about, it just doesn’t work that well. He’s talking about hallucinations.
Patrick Moorhead: Yeah.
Daniel Newman: Data being spilled all over the floor, the graph, and of course, Microsoft has the most comprehensive graph because it’s got such a giant software estate. But what Marc’s trying to say and what matches some of what we say is that when you’re a little tighter, a little narrower, and a little more specific on the data that you use, you tend to get better grounding, better accuracy, better tuning, and better outputs. Now, I’m not saying that this isn’t something that Microsoft can do with smaller language models that companies and enterprises can put together using their data and doing it in open Azure services. I think what they’re saying is the out-of-the-box, single-click experience that everybody wants is still too hard. So, that’s the conversation and the debate that we’re having. So, I mean Agentforce, to me-
Patrick Moorhead: Hey, Dan, do you think he was talking about that? Because there’s Microsoft Copilot for sales. That’s a specific product that to me competes directly with what he’s talking about here.
Daniel Newman: I think if you want my perspective, because I only know what I think he’s thinking, is that he’s using Copilot in a universal sense, because I think there are Copilots that are more… Copilot for Finance is very specific to FPNA and the data can be walled off and it can be pretty darn accurate. At the same time, when you’re using a Copilot assistant in your productivity tools and it’s looking across the entire graph, the content it creates can be pretty off from where you might want it to land on the first pass. By the way, this is the same problem I think we’re having with all large language models. It’s like search. We talk about prompt engineering. You’ve got to have the know how to make the most of it. Of course, as these models get better and smarter, it becomes less important how perfectly we script a prompt to get a pretty good answer. That’s why you and I go across different models and we play with it and we see what things come up with. That’s going to be some of what’s able to be differentiated. But in the end, I think they’re very focused on the transparency. They’re very focused on being a leader in open communications of how they’re handling these models, safety, governance. I think that’s the story they’re trying to tell.
I think it’s risky to go after Microsoft. I don’t think it’s necessarily something that will win out in the long run, but I think what they’re trying to say, and this is, I guess, where I’ll end this, because we could talk about this for hours, is the abstraction layer that I’ve said for the longest time, whether it was ServiceNow, whether it’s going to be using SAP and Oracle, whether it’s going to be on Salesforce, whether it’s going to be on Microsoft, is people are not going to use AI from 10, 15 different platforms. What they’re going to want is their data, their fabric to be universal. So, all these enterprise softwares become different database silos. These different silos become the back end and you’re going to have a limited number of front end. So, the strategy with Salesforce is the agents are this new front end. So, instead of worrying about the SaaS and the application, you go and you build these agents and a lot of those sort multiscreen RPA activities that we’ve been trying to build out for more than a decade can start to work, because now we’re taking what the promise of RPA was and we’re combining it with neural networks and reinforcement learning. So, instead of being rigid and programmed and unable to be evolved, and Pat, remember when RPA was a big thing like five, six years ago?
Patrick Moorhead: Yes.
Daniel Newman: We were talking about it all along. Who talks about it? Nobody. Because an AI powered intelligent agent has replaced RPA. The special RPA companies, the Automation Anywheres, the Bluesky or whatever, Blue Prism, UiPath, they’re all sinking right now because you can do all this stuff with agents. So, Salesforce is simplifying it, they’re putting it into a new abstraction, you get this cool generative AI builder. I spent some time with Clara Shih, the CEO of Salesforce AI. She showed me the demonstration of how this works. Build an agent, build a generative tool. By the way, Pat, you know when I talked about and eventually you’re able to talk to it and it’ll abstract a screen for you? They’re starting to get to the point now where you could tell it what you want and it will build it right in front of you to your liking. That is where I start to see AI being consumed in a really meaningful way.
