Marc Benioff on AgentForce, Slackbot, and why AI companies must be held accountable for teen suicides

consulting and corporate law. These are all valuable pieces of the economy. They are important. Salesforce who we have the CEO of joining us right now. Mark Benoff is in the ream waiting room and we will bring him in to the CBP Ultra. Mark, how are you doing?

He's back.

Great to see you guys. back for round two. I prepared for this interview. I prepared for this. I prepared for this interview. I went

No, we loved it. We loved it. I I will say it was one of the most fun conversations we've had all last year.

It's amazing. Amazing.

Really?

No. Yeah.

Well, then you guys are just not having enough fun in your life. I'm your biggest fun.

Here. Here it comes. Here it comes. The

Don't sell yourself short.

The blogging starts. Uh, but I prepared for this interview by going to Slackbot. uh now powered by Anthropic and I asked what is the single best Metallica song.

It said master of puppets. Agree or disagree.

Okay guys, I know that you are this isn't at the tip of your tongue. So I was ready [laughter] and I was going to flip it for your partner so you can he can get all the the words cuz you know we had some issues last time. Let's go through the details.

Are you going to say name your tune?

Yes. It was not a great moment for your show.

It was a rough moment. It was a rough moment.

It was rough.

Are you gonna sign that? Are you gonna sign that and send it to us so we can hang it in the rafters?

I'm not. It's mine. You [laughter] can get your own.

Metallica.com. We run it actually on Salesforce. We run.com.

You can get to your Metallica album on there and all the other good stuff that you need to kind of

Okay. So, I'm going to I'm going to give the I'm going to give my uh uh recap of John using Slackbot this morning because

he this was an ASI moment for John. He was like, I was very pleased with it.

John uh we do the show all day long. There's a lot going on in Slack and so catching up is

is tough.

This absolutely nailed the summary. Immediately found something that was like very

actually interesting, important. We're like, whoa, we missed that. I probably truthfully had about like 200 notifications across 15 different channels as they build up and it basically put together a deep research report of every little item, picked it out when am I mentioned and it was exactly what I wanted to catch up and then I went to it and I asked okay clear all of the notifications. [laughter] It couldn't do it but it gave me the shortcut shift escape and it didn't it in two seconds. So I was very satisfied. How has the response been from your customers?

Well first of all well there's the response of my customer right there. I would say I'm so excited that we could finally get this out of the shop and into people's hands. I've been using it myself for months.

Yeah,

it's completely changed my life. I think it will change your life. I think it'll change your show. I think it's going to make everything better

because, you know, we we've we've talked a long time about prompt engineering in AI. You know, when you're writing your prompt, you're kind of coming up with your question. Now, you really see the power of context engineering. that is it's looking at all the data about your show and everything you guys have ever said. It can even read your DMs and then it's coming in and it's saying, "Hey, right here, this is what you need to ask Mark Benny off this question. This is going to this is the zinger for him."

And that's the prompt you could give it, which is, "Hey, based on everything we know about Mark, everything that's ever happened in the show, everything that's happening inside Slack now, plus everything that Enthropic knows, everything that everyone knows, what's the zinger for Mark?" and then it's going to come up with it.

Yeah. Uh talk to me about the uh the the the tone and the speed. I noticed that uh it responds like an LLM. It's giving me a few paragraphs. It's not acting like a person who might respond with on it and then okay, here's what I found and then it's like six different messages. Do you is that intentional? Do you like the the the tune or do you see it evolving to your style over time? Where does all this go in terms of the the flavor of Slackbot's personality powered by

This is really different than our last interview. We're going deep in product. Okay, so number one,

we love it. [laughter]

Number one, every company is going to have a customer agent and an employee agent. And that's why we've been working on agent force to build the orchestration layer, the observability layer so that companies can get out there with their agents. And that's what I've been so excited about. Then all the apps like Slack is only one app we make. We have a whole bunch of apps like Tableau, our salescloud, our service cloud, marketing, commerce, and and on and on and on. We have a whole family of apps that help you run your business. and then a huge set of data uh capabilities including Informatica, MuleSoft, our data 360 layer that let you federate and integrate your data which means that it's going to find all the data in your company and bring it together.

Now what you're not seeing yet in Slackbot, but what I have

is that you just you just hit a button and it connects to all your Google sources also

and it connects to all

you're keeping the good stuff for yourself. this a fast takeoff scenario and [laughter] you're like we're keeping this for ourselves. We can't let it we can't let it out.

I my like I was at a conference for the last two days, really cool conference in Honolulu, Hawaii. Okay.

Um I live in Hawaii. So I popped over to Honolulu and it was called the Honolulu Defense Forum. And there's so many of my customers there. It was really cool. But I didn't have time to get briefed on all these customers. So I'm doing the briefing myself. I just say to Slackbot

in real time, I'm meeting with this customer. Tell me everything about the customer. And it's not just getting the experience from the public web,

but it's also then looking at all my proprietary databases and saying, "Boom, here you go, Mark. You're ready." And the prioritization, I I think you're going to have a great experience. I want to hear your exact feedback after you kind of get into it for a few days because it's changed how I work. It's awesome.

Yeah. I I had a I had a great experience. I didn't feel like I needed a smarter model. It did seem like I was on claude three, not four, five, which I've used and is great for some.

You want to know why you feel that way?

Why? Yeah. Why?

Because you're getting the data from Slack.

You have so much data about your business or what you know your show your everyone

that you that's what you've been missing in all of your prompts. Yeah. you're when you go on Anthropic or you go on Gemini or OpenAI or whatever you use. Yeah.

You know, it doesn't know anything about you. It doesn't have your context. It doesn't have your data. And that's what I mean context engineering. It knows you now. It really does. It knows so much about you, your company, everything you guys have done, what you've built, this great show you have.

And now it's able to really tell you, and you should like ask it like, "Hey, what are the three things we should do to, you know, increase viewership? what are we gonna what should we be doing to you know increase our revenue make us more profitable what is my priorization for the week try like because it knows so much about your company now because of the huge investment you've made in Slack

you're going to get a great outcome I'm really confident and you know I don't know if you've seen some of the videos that Mr. Beast has been doing

but he is really in it to win it on this and he runs you know his whole company

people talk about Mr. beast like like he's this consumer guy, but this guy loves enterprise software entrepreneurs as an entrepreneur.

That's true.

Oh, he's amazing. The first time I ever talked to him was a couple years ago. He goes,

I am going to be like Steve Jobs. I am going to be

like Larry Ellison. I am I'm like, what? He's like, yeah, I want to be one of the great entrepreneurs of all time.

So, he really has a mental model that's super unusual. Well, he runs his whole companies on Slack. So now he has the ability to kind of

get to a new level by bringing AI in because when you take AI data and the app together as you see

you get a tremendous outcome. AI by itself the models great fantastic but they're commodities at this point. We all know that.

Yeah.

The data sets

those are commodities on the consumer side but not your data set.

The only one who has your data set is you. And then three, the app. You've chosen Slack. Thank you for that. And so now you put one, two, and three together and bam, you are ready to roll. You know, you can now sleep with one eye open.

There we go. Uh give us a Super Bowl preview. What What's What's been exciting in the past that you've done? What do you want to do in the future with the

What's your most memorable Super Bowl? Let's start there. [clears throat]

I think there is going to be some pretty cool stuff coming from us uh for the Super Bowl. I'm not going to give you all the secrets,

okay?

I may have already given you some of the secrets, okay?

But I don't think I can give you any more of the secrets because it is going to be astonishing what we're doing and everything that we just talked about somehow serendipitously. Maybe you guys already know.

Yeah.

You know, but everything that we just talked about, take all of that and make it into a Super Bowl ad over the last 10 minutes and I think you're going to have something astonishing.

Yeah. Well, what do you think about uh uh model pickers selecting the model that I want to use? How much of that should live with the end user? I go to Slackbot and say, "Hey, I want to use Opus or I want to use 45 versus uh your team deciding what makes sense for that overall uh you know, infrastructure that you've built and then it's below the surface and I don't really care as a as a user."

Well, you're already using a whole synthesis of models.

Okay. Because Salesforce h makes we make a lot of our own models. Sure.

We use uh we use all the all the major vendors

and we bring it all together to get what you need to be successful. We don't want you to have to worry about that.

We want to just deliver a great I want you hit Slackbot and it just says boom here we go.

And I don't want you to have to think about those things. I want to make it just great a great experience. And I want to help you grow your business. you know, I want to help you grow your customers. I want to help you service them. I want to help you make your employees more productive. That that's what I'm excited about.

You guys have uh insane scale as you look across the the the Yeah. To to put it to put it uh lightly.

It is. It is crazy.

Uh but specifically, have you been surprised by any industries that are resistant to AI? like just blanket saying like ah we're

well there's so much dark things to AI that it freaks everybody out. I mean I mean how to know how dark do you want to go? Like we can go

let's go dark. I mean we were talking there was the journal had a statistic earlier. They said only 10% of adults are excited about AI.

I don't think that's as dark as I would go. I would say the rate of suicide that I saw with kids this year because of AI that was the one of the worst things I've seen in my life. It was very reminiscent of kind of what we saw with social media, kind of this unregulated technology dramatically impacting families in the most horrible way possible. I watched an episode on 60 Minutes a few weeks ago on character AI and I could not believe what I was watching.

And I think that that dark part of AI, the unregulated AI, because we know AI is kind of, you know, it's inaccurate. It's, you know, it's very kind of unwieldy. We don't know how these models work.

And to see how it was working with these children and then the kids ended up taking their lives. That's the worst thing I've ever seen in my life.

So that's the darkest part of AI. And I think we have to be careful. You know, tech companies hate regulation. They hate it. You hear that all the time except for one regulation they love, section 230. Yeah.

Which means that those companies are not held accountable for those suicides.

So those social media companies or in this case now the AI companies that that's not they basically say, "Oh, we're protected by section 230. We're a platform company." They they will quickly run to that regulation. But the reality is is highly unregulated in every other area. So, this is kind of a moment. I Sorry, am I going too dark? No,

I'm sorry. You said go back. We We have these kind of conversations. We've talked about all this stuff and uh in general,

I I wrote I wrote about this in our newsletter last week, which is that why would the why would the average America the the the all the interviews that all the lab CEOs have done have just resurfaced over the last couple years? Maybe they did the interview five six years ago talking about how uh the classic line from Sam is uh you know AI will lead to the end of the world but there's going to be a lot of great companies created in the meantime and so when people I saw that exact video yesterday had 100,000 likes on Instagram right so like that is everyday people just seeing this and being like

okay I like they maybe like Chad GBT but they're like I I like my job more or I like you know living in this world and being a human. So, I think we have a re there's a real like narrative problem with the industry uh right now and it's going to be

and if we're not about protecting our ch children, what are we about?

So, let's just start there. And because we now have evidence of what happens when it's fully unregulated and because we have the evidence of this huge horrible situation,

we have to take action. And I think some countries, you know, have taken a lot of action, especially when it comes to things like social media. You know, if you're in Singapore, you can't use social media if you're under 16 years old. So it's like it it it will eventually get taken care of. But we live in, you know, in our world here in the United States, we're we're very much like, oh, don't regulate technology because we got to keep the innovation going at all costs. We got to keep the growth happening at all costs. And I'm

Yeah. So what's the real solution? So I get it.

Yeah. Yeah. What what like what what's a pragmatic approach here? Because when it at the same time LLMs can have some negative externalities. We've talked about the example of like, you know, if if grocery stores were invented last year, all the headlines would be like this guy just went insane in a grocery store and that would be a huge story, right? Or like this person was murdered in a grocery store. That stuff happens, but like we don't it doesn't get written about every day. It's not because of the grocery stores.

It's just because

and at the same time AI from an education standpoint, all the focus is on people like not writing essays anymore, but there and there's very little writing on like, hey, this, you know, uh, you know, a 15-year-old might like teach themselves like physics through using an LLM. And maybe that's amazing, right? So, it's like uh cuts both ways.

Well, I own a media company, too. Not as exciting as yours. It's called time.

And in my little media company, we're held responsible for the content that we write and what we say. Oh, by the way, you are too.

But in the case of this technology providers, they are not.

So, I think step one is let's just hold people accountable. Let's reshape,

reform, revise section 230 and let's make let's make it let's try to save as many lives as we can by doing that.

Yeah. Who uh who are you looking to uh in terms of leading the messaging around uh around AI safety? Do you like Daario?

Oh, I'm turning it over to you right now. Uh

I'm passing this over to you guys.

Okay. Okay. Yeah.

You guys can make this part of your narrative and take it forward.

Yeah. Yeah, it makes sense. Uh what about in in the workplace? Have you had to uh contend internally with the idea of someone uh confiding in, as silly as it sounds, confiding in Slackbot about a problem that they're having with in an HR context and how that might be a new territory, a new surface area for you to deal with?

Yeah, we need to talk about Slackbot safety.

I mean, it's it sounds silly, but really, if somebody goes and and they're like, you know, my my boss said he's going to fire me next week. I'm overstressed. I'm thinking of doing something crazy. You're in this but bizarre scenario for the first time in perhaps your entire career.

We are and I think in two areas. First of all, as I mentioned, every company's going to have a customer agent and you've seen like there's 10 since we've spoken like 10 20 30,000 companies. I don't even know what the exact number is today of companies deploying agent force for their customers and we've seen these amazing agents emerge like olive you know William Sonoma site or we're going to see one emerge next week at the world economic forum in Davos called EVA and every c company has an agent to work with their customers to make them more successful and when we deploy those agents we have a tremendous uh architecture to create the guard rails to help prevent crazy things happening. You know, we don't want, you know, we've seen this where customers try to come along and have kind of cyber sex experience, you know, with these agents on these websites like, "No, no, that's not going to happen, you know, on the William Sonoma site. [laughter]

You're that's not what we're going to use the turkey baster here for." Okay.

So, we we have a whole different approach. That's

And so that level of guard rails is super critical. And on the employee side, we h we're doing exactly the same thing. We have to kind of,

you know, put those tools together and work with our partners, but also do a lot of deep engineering work to make it as as great as we can. And that core platform, Agent Force has all of those capabilities built in.

Okay. On the on the employment side, has AI rethought your thinking about employment planning, how many people you need to hire, the scope of the organization? Has it been a roller coaster or do you feel like you've been on one trajectory or the trajectory has changed? Have you been processing just AI making uh your employees more effective potentially?

Yeah, what a great question. I would say that, you know, we have about 80,000 employees now and we'll do a little over 41 billion this year in revenue. We're now the number one enterprise software right we're now the number one

enterprise software company in AI we just passed 11.1 trillion tokens

like crazy

so are you are you surprised are you surprised that companies are not getting valued on token generation yet they're such big numbers you know we we

I am I think I mean We're probably number one or number two on the token side. Um, that's probably has to do with our relationship with Billbo Baggins, but otherwise we've got this figured out. Okay. I would say what we have to do when we think about employment. To get back to your core question,

um, I've changed the employee mix.

I, uh, this year we hired more than 20% more account executives. There still needs to be people like me and you who go meet with our customers to really explain all of the intricacies and nuances of the technology. There's a lot of false narratives and people playing the market and saying things about AI that are not true, especially in regards to the enterprise. And we have to go explain it and talk to it and build the relationship and show them and prototype and deploy. And so those people are more valuable than ever and more important than ever to us. On the engineering side, I've held my engineering headcount mostly flat this year

because I've gotten so much productivity increase. I probably have about 15,000 engineers and

they're more productive than ever. I'm so proud of them. Look at what we just delivered for you guys. You know, they're like doing great. And um we need to go go even farther.

Yeah.

But you mentioned Enthropic. You know, we own about 1% of Enthropic. I would say that you know those coding tools uh you mentioned cloud code other things cursor we use that inside we do things to have a more efficient more productive engineering organization and we it turns out we've reduced the number of cos to one because of [laughter] slackbot we can do more because uh we can deploy that AI technology the cos did uh I mean you've spent your entire career in technology did this did the chatbt moment did the AI I boom. Uh, were you expecting it or did it hit you like a flashbang? Was it just was it just a just a massive flashbang that just hit you like a ton of bricks? [laughter]

Hit me like a ton of bricks. Um, I would say

we have a flash. In fact, I had to use it.

10 years ago, I had the nastiest dream about AI and so I started a whole AI initiative at Salesforce called Einstein.

Okay. and we became the number one enterprise AI.

And so we were building models, doing these things, making all it happen. I always believed that this time would happen, but it was more about predictive AI and machine learning and machine intelligence and these kinds of things. And it was it, you know, we still use all that, by the way, to make this work.

And then when the large language model emerged and prompt engineering emerged and by the way, you may or may not know, but I'll give a little credit to my engineering team because they deserve it. But prompt engineering actually emerged out of Salesforce research

and then fantastic

all of these companies you know have new companies came along to say we're just going to focus on the large language model and it's been a huge incredible success story.

So um I am just so impressed with what's happened in the last three years but it's just another core part of our infrastructure now and I think you see that in Slackbot. So yeah,

getting back to Slackbot, it was just a seamless upgrade for you today.

Sure,

you have Slack. We added Slackbot. When we added Slackbot, we added

all the AI and all the data capability. And now actually in the new version of Slack, you're going to see a lot of new CRM capabilities as well. You'll be able to manage your customers information without even having to go to Salesforce.

You will be able to manage your service organization. But here's the secret. Salesforce is behind the scene in doing all of it. Let's go.

And at some point, if you want to upgrade to the bigger, more advanced adult versions of all those things, you'll be able to do that, too.

Okay. I I I want to talk about that because uh we've we've talked to a lot of companies where they they have a single product there, a point solution, but because of AI, they're able to add on other surface area. Everyone's bleeding into everyone else's territory. Everyone's expanding right now. What is it like running an organization where you might have two products that are now bleeding into each other? And like how do you have to go and play tiebreaker a lot? Do you have a pattern for this? Like how do you how do you work through all that?

I love it. I I think everyone should, you know, innovate and create and compete with each other, especially inside the company. Yeah.

Some people stay in their lane, some people don't. I'm not going to interfere with that. You know, we have these tremendous leaders like running Slack or

our sales product. Our sales and service product are 10 billion dollar products each. They are the some of the largest products in the software industry and the new versions are incredible, but the new versions are all Slackbot first and you haven't seen them yet, but if you want to get a demo later after the show,

you'll see Slackbot and Slack sitting on top of Salesforce and that boom, you can have all this incredible productivity, but it's the lineage of Salesforce coming forward into the present moment reality of these unbelievable new advancements in AI. Yeah.

And that is what I've been working on. So, a Slackbot first world, the ability for every company to have a customer agent, an employee agent, uh the agent force layer, the all these incredible family of applications we have. We just bought a company called Informatica, by the way, for like $9 billion.

And it was

kind of an incredible acquisition. All right. Well, there we go. But I love I always love the company. Yeah.

But you know they've done things in their new technology that for AI if you don't have your data right you don't have your AI right. Remember the reason why Slackbot is working is we have all your data in there normalized. And so it's able to take all your data and give you really cool answers. Well every company needs to normalize their data. So that's why we bought Informatica. Everybody needs to integrate it together. And then you have to go and connect to all the other data sources in the enterprise like snowflake or datab bricks or big query or red shift or even IBM mainframes. We call that zero copy. So our system goes out grabs all that data in all these other places brings it together like it is for you already in Slack

and then everything sits on top of it and you get this really powerful AI experience. So that's what's going on. That's what's exciting for me every single day.

Let's talk about M&A. Uh you talked about Informatica. There's uh you know the SAS apocalypse has been widely reported. I'm sure you look at some charts and get a little get a little

I'd rather not look at the charts. I can't figure them out. I mean our revenue and our profitability and cash flow go up and then the the stock is all over the place because people think

there's some weird end of SAS. But I would just ask you guys did did did Slack become a lot more valuable for you today or less?

Way more. way more way and I'm not talking about that. I'm talking about you looking for value. You've done so a ton of M&A like you must look around at at companies that you feel like are oversold.

I think there's a lot of incredible opportunities right now in the market

and um you you if you were not looking at look we do about 15 or 16 billion in positive cash flow a year. We're going to use that. We're going to buy back some We're going to buy back some stock. We're going to do some dividend action, too.

Fantastic.

You know, but what else are we going to do with our cash? I think we should buy some companies also, you know, because I love organic innovation, but I love inorganic innovation, too. I bought about 100 companies.

Give us a give us a give us a a threeinut M&A master class. There's a lot of companies whose market caps have inflated. They've got a bunch of cash on their balance sheet. They're, you know, 1% of the size of Salesforce, but maybe they're starting to think about M&A. What have you learned about how to do it well in a few minutes?

And this is the last question.

You got to, you have to innovate. Innovation is our core, one of our core values. What are our core values? Trust, customer success, innovation, and quality, sustainability. Those are our five core values. We can talk about any of those, but when we talk about innovation, we talk about organic and inorganic innovation, right? Organic is those 15,000 engineers making Slackbot. It's 100% organic.

But then, well, actually, but it's not because we bought Slack. Remember five years ago when I bought Slack? Because I love Slack. I thought they did such a great job. I thought it was a one-time trade. I don't think anything was going to replace it and I thought I could make it a lot better.

I thought I could put it on top of all of our products. I thought it could be a user interface for the future. I thought AI would be incredibly benefited by the data inside Slack. We just proved that to you with Slackbot.

So, we bought Slack. Well, it was a huge purchase at the time. One of the biggest ever in the software industry. I think it was like $25 billion.

Our biggest.

But guess what? The value that it's provided now it's, you know, it's not a subbillion dollar company like when we bought it. It's a multi-billion dollar company. I think it's like almost $3 billion or more. or I don't have the exact number in front of me, but you can just go, "Wow, this is like incredible the value that Salesforce and Slack together provide."

And that's what I'm looking for. Or Informatica plus Salesforce together or Mulesoft. I mentioned we bought it when it was a $300 million,

you know, revenue company and, you know, now it's a huge multi-billion dollar software engine as well. So we have so many of those and our job is to buy them and make them work. Not all of them are going to work. Acquisitions are very risky. You got to be really careful, you know, and then you can have it all worked out and it still doesn't work out. But it's worth it to take the risk because innovation is risky and organic and inorganic innovation you have to do both. And

what is a good what is a good one hit rate? What is a good hit rate at at scale? you said you bought something around a hundred companies. Uh but for uh when you're talking like with your M&A team like how often are you okay with them just you know and the company just swinging and missing?

I have such a cool team there because not only do they that do that they also run a $5 billion investment fund. You know we have invested in so many amazing companies and taken so many amazing companies public. We just sold Whiz that we helped start to Google.

We took we helped grow Snowflake and took it public and made one I think we made a billion and a half dollars on that. I think we made a billion half dollars on whiz.

I think we have now about

you know a point of anthropic which is a few billion dollars or so.

Fantastic.

I you know and it was started you know when it was a tiny company. So we love investing number one. We love buying things and we think all of that is innovation.

I love it.

And it's all in one big innovation bucket. And I'm, you know, but we're trying to do something that others have not done is we're trying to innovate in the enterprise at scale.

Yeah.

And that is the hard thing because look, as I said, we have 80,000 people and more than 41 billion and we're starting going to start our next fiscal year out on February 1 and it's going to be a it's going to be a massive year bigger than that. So

I'm like going, "Whoa, these numbers are huge." But I've got a huge dream. I'm trying to go as fast as possible, you know, to 100 billion in revenue, but while maintaining this incredible profit. Oh, we're one of the most profitable enterprise software companies, too. We're delivering more than 34% margin this year.

I have I have one last question coming last question. Are dreams an underrated Ares underrated source of uh business alpha? You said you said you had a had a had almost a nightmare about AI and that started Einstein. Do you look for signal in your dreams?

I I pay attention to my dreams. I pay attention to my my visions, my insights. And you know, Salesforce started because I was swimming with a pod of dolphins outside the the coast of Hawaii where I live.

And I was in about a hundred dolphins and I was one with the pod. And when I was one with the pod, all of a sudden in my mind, I saw this vision of what Salesforce could look like. And I came into Larry Ellison and I said, "Hey, I think I'm going to quit my job." And he gave me $2 million and we started Salesforce. And that was that. That was 26 years ago.

It's amazing. Well, thank you so much for sharing the story with us.

Thank you, Dolphins.

Thank you, Dolphins. Thank you [laughter] to the sound. The sound is amazing. Look at that.

Thank you, Dolphins.

Thank you for coming on the show. Take care of the dolphins.

We will talk to you soon. Have a great uh super fun. The chat's calling him the Steve Jobs of AIDNED. I think he's a strong candidate.