Our guest this week on the GeekWire Podcast is Salesforce CEO Marc Benioff, who says he has never been as excited about anything in his career as he is about the latest developments in artificial intelligence — AI agents that can autonomously reason, plan, and take action on behalf of businesses.
“We’re going to do this in a large-scale way and give customers a level of success with AI that they’ve really never had before,” he said. “I’m super excited about it. I think it’s going to be incredible.”
Benioff is almost as strong in his negative sentiments toward Microsoft’s Copilot. He calls Copilot the second coming of Microsoft’s much-maligned “Clippy” Office assistant, and asserts that the Redmond company is giving AI a bad name by disappointing customers with underwhelming results and lax security.
There was one thing that diverted him from those topics during our conversation: his dog, Honey, sneaking up behind him in his home office to eat his bagel, tipping over a tray and breaking a plate in the process.
But it didn’t take long for him to get back on topic: “We’ll clean that mess up,” he said. “That’s, by the way, very metaphorical for the mess that Microsoft has made in regards to the AI industry.”
I spoke with Benioff, the CEO and co-founder of cloud computing pioneer and enterprise applications giant, in advance of the general availability of the San Francisco company’s Agentforce AI technology for sales and service.
We also talked about Benioff’s ownership of Time magazine, and what that means for his political involvement; and the status of his past pledge to turn Seattle into Salesforce’s HQ2.
Listen below, and continue reading for a transcript, edited for clarity. Subscribe to GeekWire in Apple Podcasts, Spotify, or wherever you listen.
Todd Bishop: You seem to be everywhere lately. You’re talking about the potential for AI agents to transform business. You’ve been slapping Microsoft upside the head, which we can talk about. You even talked about the potential to rename the company Agentforce. It’s like the Marc Benioff of old, if you don’t mind me saying. So let me just start by asking you, what’s got you so fired up about AI agents, this whole notion of autonomous artificial intelligence?
Marc Benioff: Well, you are right. I am fired up. I would say I’ve never been more excited about anything in my entire career. For the last 45 years that I’ve been writing software and publishing software and working with customers to be successful with software, well, this is our dream, what’s happening now, that we can really see that software can do things that it’s never been able to do before, that it can make us more productive, truly more productive. It can augment our employees. It can improve our margins, grow our revenues. It can improve our KPIs.
Fundamentally improving our overall business performance, this is our dream of why we all got into the software, business because we thought this is where software’s going and we are at this moment right now. And that’s why I’m really totally excited.
TB: When you make the pitch to customers, I can imagine that some of them are still saying, “Wait a second, Marc. I’m still getting used to the notion of just working with AI as a companion. This might seem like a step too far for 2024, to have autonomous AI agents working on behalf of my business.” What do you say to folks when they bring that up?
Benioff: Well, I think you’re taking a pretty big jump into the abyss there, and before we do that, I think you should take a step back. Look, we’ve been working with AI for a long time. For Salesforce, we’ve been doing AI for quite a while and seriously AI for more than a decade, especially with our Einstein platform. We are the largest provider of enterprise AI in the world. We do about 2 trillion enterprise AI transactions this week alone, and that’s across all of our products and platforms.
And it’s been a fantastic journey — machine intelligence, machine learning, deep learning. I mean, it’s been awesome. And this generative AI moment is exciting because it’s another step on our collective AI journey.
So while we’ve been working on AI, I think we’ve always been saying, “Oh, it’s going to make us more successful. It’s going to make us more productive.” But now with this introduction of this concept of these agents — it’s software that is reasoning, planning, taking action on your behalf — and using the generative capabilities that we see in these LLMs that we love. And then grounding that, connecting with our corporate databases, It gives us the ability to really achieve capabilities with enterprise software we just have not been able to do before. And we are in the zone.
[With Agentforce], the early results have been just nothing short of, well, it’s just stunning, what is possible. We’re going to come out of the gate really swinging on this, and that’s what you’re feeling.
AI agents, a practical example
TB: I got a chance to talk with a couple of your executives, Adam Evans and Brent Hayward, so I have a sense for the answer to this question, but can you give me some examples of what people will be able to do with Agentforce that they might not have been able to do with just a chatbot or a generative AI solution?
Benioff: This is about, for the first time, humans with agents driving customer success together.
Let’s take a specific industry, healthcare. All of us have had this experience. I even have a boot right now I’m wearing because I had a bad SCUBA diving jump a couple weeks ago, and I’m going to be going back into an MRI on Friday. …
Healthcare is very much an individual sport in our country where you’re on your own, and in the future that just is not going to be true. Because what’s going to happen is, your phone is going to ring, or you’re going to get a text and it’s going to be, “Hey, did you drink the water? We’re following up for your MRI and just want to know what’s going on with you and how are you feeling, and are you experiencing swelling? What’s going on exactly?”
And you’re going to say, “Yes, I did. Oh, thanks for reminding me. Oh yeah, no, it’s fine.” “Do you need more help on understanding all the things going on with your foot?” “Oh, give me more information. Oh, that’s great. What are the five things I need to know?” “These are the five things you should know. And are you sure you’re OK? Do you want to talk to a doctor?” “No.” “Do you need to reschedule your appointment?” “No.” “Well, we’re going to schedule a repeat MRI for you in a month, so I’ll be contacting you again.”
That last part, it’s all with an agent, and this idea that we can fill in the gaps of our human workforce with a digital workforce. This is a big idea and this is a great thing in this industry, healthcare. Because I might need a repeat MRI, I might need repeat labs, I might need to schedule an appointment with my doctor.
But, look, my doctor does not want to hear from me. My doctor’s burnt out. My nurse is burnt out. They’re in a post-pandemic malaise. You know that’s true. You go talk to any healthcare organization. But these digital workers, they could fill in this gap and they can give us some breathing space between our doctors, who should be working on certain very important things.
Because there’s a lot the AI cannot do, especially in healthcare. But there’s now a lot that the AI can do, especially in healthcare. And our job has to be to put those two things together. And if we can do that, oh my gosh, this is a moment for our customers that is going to be absolutely amazing.
Agentforce vs. Microsoft Copilot
TB: You’ve been outspoken about Microsoft in recent weeks, and I really want to dig into this a little bit more with you than you’ve talked about before. Fundamentally, your issue is that Microsoft has set expectations too high and not met them, in addition to other things. But I don’t want to speak for you. What’s your problem with Clippy?
Benioff: We’ve all had that ChatGPT moment where we type in the query and we get the result and it’s super-search and we’re like, “Oh my gosh, this is amazing. This is magic. This is awesome. How is this possible?” And our first experience with an LLM, we’re all remembering that. That’s so important.
But the idea for them to say, “Now here’s Copilot, and it’s now going to make your enterprise so much better, and here you go,” that’s a fantasy that has not played out. And you just talk to customers and they’ll tell you, Copilot has not transformed their enterprises.
Every CEO was so excited with the vision of, “Now that we have LLMs, AI is now curing cancer and climate change has stopped and your company’s going to be fully automated.” Not so fast. And when we look at Copilot and when we look at this concept that every company has to have their own model or LLM or train and retrain their model, again, this is not how this is going to play out. And that needs to come to an end.
That is really the big message that we have for our customers, which is, we are the largest provider of enterprise AI, and we’ve made you a lot more productive with AI. And now watch this, with Agentforce. With Agentforce, you’re going to achieve the dreams and visions that you’ve always had for AI. And we’re going to show you, just like in my examples with healthcare, where you’re going to be able to fill in the gaps with your customers to have a much higher level of customer intimacy.
Now, I’ll give you another great example of a real customer. One of the first customers that we turned Agentforce on is Wiley. Wiley is the textbook manufacturer, and right now everyone’s going back to school so they’re all trying to buy more textbooks, and they are usually surging their workforce, hiring more sales and service people.
But not now, not at this time of the year, because they are using Agentforce. So they were able to scale their service organization, their sales organization, their marketing organization, their customer outreach, with digital workers, not just human workers. And that is so important. And this idea that Agentforce gives you the ability to deliver a higher level of customer success than ever before, that is really amazing. And that is why I’m so excited.
Microsoft’s many Copilots
TB: Microsoft itself has created some branding confusion here and so it’s important to point out, when you’re talking about Copilot, you’re talking about Copilot for Microsoft 365, which is the thing that they charge an extra $30 a user a month for. Not GitHub Copilot, which frankly has been one of the early successes in the industry in terms of its ability to be a coding companion.
Benioff: Let’s just take this apart.
TB: Go for it.
Benioff: Because I think what you’re getting into is something very important, which is that software developers are more productive using AI, and there’s different tools for doing that, including that.
But then to say, now “My software developer is more productive with AI of this,” and now to say, “My whole enterprise is going to be more successful with this technology” is not true, and yet it’s what they’ve been selling.
And you can go out there and see with the customers, they don’t have success with this idea. This idea that a fundamentally ungrounded AI into your enterprise is going to deliver you more productivity is not true, and that’s where Microsoft has gotten themselves into trouble here. So let’s just put that aside for a second.
TB: Well, but Marc, I should say, they would push back on that. We both know that, and they would say 60% of the Fortune 500 are using this thing, and they have specific case studies where they would show successes as well. I’m not going to go through them.
Benioff: Which customer are you going to say, let’s talk about it. Take one.
TB: I don’t have Microsoft’s case studies in front of me. I can just tell you that there are some.
Benioff: I have the product. It doesn’t work. … I mean, I have it right here. I’m like, “It’s just ChatGPT.”
TB: What then is the difference with Agentforce?
Benioff: What you have to realize is that LLMs are just a component of intelligence. They’re just an ingredient. Like if we’re making a cake and we need sugar or salt, it’s just a component. But an LLM by itself is just an input-output search engine. It has an amalgamated data set with an algorithm that are combined and you put something in, your prompt, and out comes a response. It’s not a data storage mechanism. It doesn’t store your corporate data. It doesn’t have your metadata. It doesn’t have your sharing model, it doesn’t have your security model. Let me give you another example.
I had a bank CEO come and talk to me … and they have this idea that you can just dump your corporate data into an LLM and it somehow is going to make you a smarter, better company. It’s not true.
But what is true is you can build an architecture out through a platform, good old-fashioned enterprise software platform approach, and the combination of an LLM, your corporate data, your corporate metadata, your sharing model, your security model, and also other critical parts of enterprise software that we know are so important, like the apps themselves, because there has to be a seamless handoff between your human worker and your digital worker. All of these things combined can achieve that vision, and that is the magic and the promise of what’s possible today.
Gartner on enterprise AI security
TB: Related to this, you’ve said that Microsoft Copilot is spilling data everywhere, and I went back and found the Gartner study that you were citing there. And so just briefly here, I want to quote what it says. It says, “Microsoft 365 Copilot honors user permissions. It also respects data security controls such as sensitivity labels where these have been correctly applied. However, if controls have not been correctly applied or permissions are too open, Microsoft 365 Copilot can increase the risk of oversharing by retrieving, summarizing, and generating content that the user should not have access to.”
That’s part of it. I gotta say, if I were editing a story that a reporter turned in, and that was the text, I would have a hard time putting the headline on there that it was “spilling data everywhere.” Can you get into more detail about what you mean by that?
Benioff: Well, I can. I think you actually just said it, which is that customers — and you can read that Gartner report yourself — customers basically have reported directly to me that they’re seeing results in queries that they should not be seeing, that they don’t have that authorization as a user.
In the example of banking, which we just went into with the know-your-customer approach, my banker and your banker have different sharing models. Your banker cannot see my account balance and my banker cannot see your account balance. But in this example, Copilot is showing up, giving people who should not be seeing that data, that data. That’s all that’s going on.
TB: Right, and I think the point of the Gartner report is that it is happening in some cases, and it’s actually stopping some companies from adopting Microsoft 365 Copilot, but it seems like it’s more about the permissions that those companies themselves have put on the data and the protections that they have implemented.
Benioff: My point is that’s never going to happen with a Salesforce platform because you’re defining those user roles up front, your sharing model. You’re defining your security model. And when we’re talking about regulated industries like healthcare, financial services, the government, and other pieces, there is no margin for error on these things.
My healthcare data, my banking data, I don’t want that spilling all over the floor of these enterprises, which is what’s happening. And you’re citing the report actually that I read, and there’s also a very good Computerworld article on it, as well, but it’s not just that. It’s this false premise that Microsoft has sold customers that this is what AI was meant to be, and it’s not true.
What AI was meant to be is the idea that we’re going to have a more productive enterprise with augmented employees, with higher margins, with higher revenues, and this is how we’re going to achieve it. And in my example with healthcare, or my example with banking now, well, that’s what Agentforce does.
Like in the example with Wiley. They are achieving all of those things which you cannot achieve with Copilot. And this idea that we’ve been sold for the last two years that Copilot can do this, it’s not true.
So we for 25 years have helped our customers build their sales forces, their service forces, their marketing forces, their commerce forces, they’re analytics forces, they’re data forces. Now, we’re helping to build their Agentforce.
And that idea that workforces are fundamentally overwhelmed, that they have low-value tasks, stalled productivity, fixed capacity, burning out, like I mentioned, in healthcare, well, customers want more. They want to get down to that immediate, zero-hold-time idea, personal and empathetic service, working with experts, scheduling things, taking actions immediately, have an instant response. None of those things happen with Copilot.
And that bigger idea we talked about at the beginning: what if workforces had no limits? What if workforces had no limits? What does that mean for your company? That’s not what we’re getting with Copilot, my friend. And I think you know that is true. That idea, which is the big idea in terms of AI, radically improving our capabilities, what if workforce has no limits, that isn’t what’s going on, and that is what we’re going to deliver with Agentforce.
And that idea that our customers can use a platform that’s trusted and secure, scalable and accurate, easy to customize, that has the AI built in, that has your apps built in, that uses a metadata framework, that has an open ecosystem, all of these basic capabilities. That idea that you can build radically new types of applications to automate your company? That is AI.
That is what AI was meant to be and that is the contrast that I’m playing around this fantasy of not only Copilot, but also this false premise that I hear from a lot of customers that they’ve been sold by Microsoft, that they need to build their own models and retrain their own models. And that’s why we need these increasing levels of hyperscaler capacity. And those things are not true. This can be done in a more efficient software infrastructure just as we have for decades as the proven computer science approach.
Microsoft’s new AI agents
TB: I don’t mean to put myself in the position of Microsoft proxy here, but I do want to point out …
Benioff: No, it’s good. It’s great. You’re doing fine.
TB: … they did come out this week, seemingly in a preemptive move ..
Benioff: Did you use the word panic?
TB: I don’t know who’s panicking. I read half the headlines and they say you’re panicking. I read the other half and they say Microsoft’s panicking. … But Microsoft did announce their own AI agents, about 10 of them, for Dynamics 365. Did you see that as validation?
Benioff: No. You have to understand, number one, Copilot has failed, and this fantasy that every company has to go build and rebuild their own model has failed, and customers are looking for a new opportunity for this level of automation. That is what is exciting. And we can do this. We can offer companies an incredible new capability, and we see that with our customers across the board.
Let’s talk about Disney. Disney is another great example where they have a very complex product line and for their employees, they call them cast members, to get this very intelligent response for a customer is incredibly difficult.
Now, we automate every Disney customer touch point from Disney+ to the service center, to selling the real estate, to this idea that you’re in the park with a Disney guide. And this is an opportunity for their employees to have an even heightened customer experience.
And I’ll give you a very concrete example. I’m walking through the park, something I actually do four times a year, and I’m with a Disney guide, and we’re on our way to Rise of the Resistance, my favorite ride, it’s at Galaxy’s Edge, which is Star Wars Land, and now the guide is notified the ride is down, because it’s a very complex ride to maintain. What is the guide to do? Does the guide, A, know flow control across the park so that it knows every ride how they’re doing and what’s busy and what’s not, and, B, my ride history, everything I have done and not done? Because those two things need to come together in an instant to make a recommendation to me, “Marc, Rise of the Resistance is down. We need to go to this new location.”
But the agent at that moment can say, with now more than 90% accuracy, “Yes, Marc needs to go to Toon Town.” … There’s no way the guide can do that on their own. It’s not possible. But the AI and the agent can, and now the agent is working side by side with the guide. That is agents with humans driving a greater level of customer success to better.
‘Trailblazer’ and human connections
TB: You were talking earlier about those “aha” AI moments, and I can assure you that you’re speaking with a real human here on the other side, as you can tell, and not an AI agent yet. But I had one of those “aha ” moments with Gogle Notebook LM.
Benioff: That is a great product.
So I read Trailblazer when it came out in 2019, and I have a copy on Kindle, and I took the Kindle copy and was able to fiddle with it and get it into Notebook LM and upload it with all sorts of background about the AI stuff and everything. And I asked it for questions to ask you. And one of the questions blew me away.
“You write in your book Trailblazer that you’ve never forgotten what you learned from your dad. “Nothing can take the place of human relationships, the bedrock of any business.’ How do you balance your belief in the power of AI with the importance of human connection and relationships in business?”
Benioff: Well, this is I think going to be the seminal moment, not just for me, but for all of us. Because there’s no question that we have this incredible ability to augment our employees for the first time with digital workers. That idea is so exciting. That idea that we’re going to work side by side with AI, we’re all going to get our head around that. We’ve seen it in the movies. We all saw it in “Minority Report.” We saw a “Terminator.” We saw “Her.” We saw “Space Odyssey.” We saw them all. Well, we’re going to live it and there’s no question it’s all unfolding in front of us.
And that is going to be the existential question of our time, which is, you and I are going to work side by side with agents. I’m sure we’re already doing that at some level already. For example, how many times a day do you already go to Gemini or ChatGPT or Perplexity or you.com or any of these systems to get an answer? Probably quite a bit.
But now let’s just go back a second. Before that, you were going to Google and you were doing Google Search, and Google Search also is AI. And now you have “super search,” is what I’d call with ChatGPT. And then when we combine that technology and we ground it into your company data, add your sharing model, which I’ve now made that point exhaustively, integrat it with your apps, it’s time for some magic to happen. And I think that’s what’s really cool.
We just did this conference in San Francisco called Dreamforce, and we had 45,000 people come. It was awesome. I don’t think you came, but maybe next year. And while we were there, we took 10,000 of our customers, because I brought in 4,000 of my employees just to focus on something called Launchpad, which was help our customers build agents in real time.
I have a dream that I’m going to try to get a billion agents going in the next year. So I wanted to really use it as a kickstart. I did one today in Tokyo, by the way. I did one last week in the UK and in Australia. I’m trying to really get this going globally. And sat down with those customers and said, “We’re going to build this. Let’s go.” And that idea that we can do it and add and extend their system, that is really important.
Saks Fifth Avenue came to the conference, and they are a tremendous Salesforce customer already, we automate all their customer touch points, and in an hour they built a system to manage their returns and their customer service. And if you go to their site today, it’s running.
So that’s a big thought, that they can extend their human workforce in an instant, in a moment. And I think we all realize our workforces are constrained emotionally, physically, spiritually, mentally, and we just don’t have enough people to do the jobs that have to get done.
Even in my own company, I feel that way. Even right now as I’m rolling out Agentforce, I wish I could be doing so much more faster and more scaled. Agents are going to provide that capability, and that’s what it’s all about.
Salesforce in the Seattle area
TB: You mentioned when Salesforce acquired Tableau that Seattle was going to be HQ2. Do you still see it that way?
Benioff: We have a huge workforce in Seattle. We have just tremendous folks up there. We have so many groups all over the world now. I think what I did not expect was, we had this thing, I don’t know if you heard about it, called the pandemic. And when we had the pandemic and we all went home and we all learned how to use Zoom and be comfortable with Zoom, that changed everything for me.
I think HQ2 probably looks a little bit more like our homes and not so much like a specific city. Maybe that’s HQ1 for a lot of companies. I have friends of mine who only have remote workforces. So I think everything changed at that moment.
It’s not that Seattle isn’t really important to us. We have a huge facility there. We have a brand new tower in Chicago, in Sydney, in Tokyo, in New York, in Atlanta, in Paris, all over the world. But at the end of the day, we are more flexible, more virtual workers than ever before, and we’re going to continue to be that way.
Time magazine and the media business
TB: In 2020, as part of your ownership of Time magazine, you declared, I think rightly so from a journalistic standpoint and ethically that you would no longer be making political donations and (personal) endorsements. As we approach November 5th, any regrets about that decision?
Benioff: No. I think it’s been a great decision. First of all, I’ve really enjoyed my stewardship of Time. Here’s this issue on inspiring women, I’m holding it up, and this is something that we brought actually into the brand, which is not only the concept of Man of the Year, but Woman of the Year.
I would say that this has been a moment in the last six years where we’ve tried to focus more on the middle way, and that’s owning Time and being in the media industry. It’s, by the way, very different than the tech industry. It’s much harder. The media industry is very difficult. It’s very complicated. The tech industry, software, I’ve been doing for 45 years, it’s much easier.
TB: What’s the toughest thing about media?
Benioff: It’s the fluidity of the models, and I think that with Time, what we’ve done is, we have the print, which I just showed you. We also have the digital. We have events. We have other kinds of digital products. Even our NFTs, if you remember those. We have our Time Studios. We have a movie studio now. We have many different types of products inside Time. And that idea that we’re constantly building and growing and expanding brands, sub-brands, new products, it’s so fluid and so fast moving. I would say that in some ways it’s even more so than the tech industry.
Subscribe to GeekWire in Apple Podcasts, Spotify, or wherever you listen.
Audio editing by Curt Milton.
https://ift.tt/OtTzR6p October 26, 2024 at 03:14PM GeekWire
Post a Comment
0Comments