Good morning,

This week’s Stratechery Interview is with Salesforce Founder and CEO Marc Benioff. Salesforce is one of the most important companies in tech history. After a rocketship ascent at Oracle, Benioff realized that the Internet would enable an entirely new way of serving enterprise software, via the cloud. That, though, would require a new business model, a new way of marketing, and, of course, an entirely new approach to building software. Salesforce represented all of that, and more: over the ensuing decades Salesforce has transformed into a platform for enterprise applications of all types.

Today Salesforce’s big initiative is Agentforce — a new wave of AI agents, built on the Salesforce platform — which I wrote about earlier this month in this Update and in Enterprise Philosophy and The First Wave of AI. In this interview we first retrace Benioff’s background and the founding of Salesforce, and then get into his vision for AI, and what many people — including yours truly — get wrong. Benioff’s vision is compelling, and the case he makes for why Salesforce is well-positioned to win is a strong one; more importantly for me, this interview is a lot of fun!

As a reminder, all Stratechery content, including interviews, is available as a podcast; click the link at the top of this email to add Stratechery to your podcast player.

On to the Interview:

An Interview with Salesforce CEO Marc Benioff about AI Abundance

This interview is lightly edited for clarity.

Background

Marc Benioff, welcome to Stratechery.

Marc Benioff: Oh, it’s great to be with you, of course.

So I know that AI is the topic of the day and a major focus for Salesforce, your new Agentforce initiative, but before we get to that, I always like to understand a little bit more about the background of the people I talk to, even the famous ones. It’s your first time on Stratechery, so you’re going to have to bear with me here.

MB: I’ll bear with you.

Okay, great. Question number one, I always like to go back to your childhood and I was reading up on you a little bit—

MB: Do you want me to lie down while we’re doing this?

I have to get an honest answer to this question.

MB: Do you want me to sit on the couch? I don’t know what we’re talking about.

(laughing) Your first app was called How to Juggle, which you allegedly sold for $75.

MB: Allegedly.

I need to know who the buyer was.

MB: CLOAD Magazine in Goleta, California. Do you know what the command CLOAD is?

Yes.

MB: What is it? We’ll test you, how good you are. CLOAD, what does the command mean?

You had to actually put the application from the disk onto the drive so you could run it.

MB: Aha! Fail! Cassette load.

Oh! Okay, yeah, there you go. Man, I just got owned.

MB: (laughing) Sorry. I’m sorry. Well, you asked for it, Jesus.

I did, I walked right into it. Walked right into it.

MB: CLOAD was — you’d get this cassette in the mail every month with all this software that you would load into your TRS-80 Model 1, and I was in high school and 15 years old, I think, and I just started writing software and I really love this idea that I could write my own software, and I sent them this thing and they’re like, “You know what? We’re buying this from you and we’re putting it on our cassette this month for $75”, and I’m like, “Right on!”. I actually have the agreement somewhere, I don’t think it’s here, but I have that agreement.

Yeah, that’s one worth framing on the wall. I mean, you created an Atari game company called Liberty Software.

MB: Liberty Software.

You wrote assembly code for the Mac as an internship. It sounds like you had some serious chops, why didn’t you just continue being a software developer?

MB: Well, what do you think I do for a living?

You do develop software at a large scale, but I’m curious, what drove this transition?

MB: I am the CEO of a large software company. I know we haven’t really met before, but I’ll just fill you in on the details at a high level.

(laughing) The No Software logo confused me.

MB: (laughing) No, you’re right, I started when I was 15, I have a huge love of software, I programmed that first game. Well, I think it was just written in BASIC, it wasn’t like some amazing assembly language thing, and then I wrote a bunch of games with my friend in high school, Steve Fisher, who’s now the head of our engineering efforts at Salesforce. We had this high school software company and we wrote games like Escape from Vulcan’s Island.

Well, maybe if I just said cartridge load then it would have been less embarrassing.

MB: (laughing) And then when I was in college, now we’ve fast-forwarded, I’m watching this 1984 commercial in my fraternity room and I’m at USC, I’m in the entrepreneur program learning how to be an entrepreneur now, and I watched this ad that Steve Jobs did on why 1984 won’t be like 1984, and I called Apple the next day and talked to Guy Kawasaki.

Nice.

MB: And Guy Kawasaki said, “Oh, we want people like you writing software, I’m going to get you all set up”, and I’m like, “I’m an assembly language programmer now actually”, I had just finished a cartridge for a Romox called Flapper, which was a video game, and I’m like, “I’d like to write some assembly language on the Mac”, and he’s like, “Okay, well, we’re doing 68000 and you have to buy a Lisa”. And I’m like, “Well, no, no, no, I don’t want to buy a Lisa, I would just want to do this natively, don’t you have something that runs natively on the Mac?”, and he’s like, “Okay, listen, you get two Macs and you can put a cable between them and you compile on one and you debug on another and this is how it’s going to work and it’s called the MDS 68000 Development System and Bill Duvall is going to be the architect of this and it’s great”.

So I went ahead and I’m like, “I just buy these two Macs?”, “Yes”, “You buy the cable?”, “That’s all”, “Then you plug them together?”, “Of course”. Then I got the whole thing and I got their code and then it wasn’t working and so I called them back, and this is now maybe May of ’84, and I say, “Mr. Kawasaki, you remember me? I’m just your friend here at USC who’s a sophomore in college and I’m trying to write software for your computer, but your assembly language system does not work”, and he goes, “Could you come to Apple for a few months and help us with this?”, and I went, “Wow, that’s quite an offer”. I’m like, “Well, it’s only a few miles from my parents’ home where I’m going to be living this summer, so I’m in”. I showed up at work and it was a fucking trip, man. That was Apple, 1984.

And yet you ended up at Oracle. Although, to be fair, Oracle was a very different company then than people think about it today, Larry Ellison absolutely in his prime as it were. What made you want to go there? Because you started a technical support role, you weren’t doing assembly language there.

MB: Oh, yeah. That’s not what happened. What happened was that I was writing all these games and then I pivoted to this Apple platform, then I stayed at Apple and worked in different kinds of groups, and I even went to an external group to work in their VAR channel and was having a pretty good experience while I was finishing up college.

I’m still one of those people who thinks a degree is worthwhile, and I’m working on getting my degree and I’m working on getting a Bachelor of Science in Business Administration from USC, Marshall School. I’m at this new job at Apple now, this external job, and my manager comes to me and says, “You’re pretty good with the technology”, I go, “Thank you”, “But your sales skills kind of suck and you need to think about when you graduate, you’re not going to just go out and program”. And I said, “You want me to get a sales job?”, he’s like, “Yeah”.

Then totally coincidentally, my entrepreneurship professors, the three of them, brought me together. These are some great people, Tom O’Malia and Dick Buskirk and others, and they said, “Now, Marc, we’ve been reading your business plans and watching you in class and looking at how you’re writing, and you’re always arguing with us about these different things and it’s great, but we do not want you starting a company yet”, and I said, “Why? I’m ready to go”. They’re like, “No, you are not, you need to learn some sales and customer and communication skills”, and one of the guys who I was working with, and a guy who I was working for at Apple said, “You need to call this guy Gary Gibson at Oracle because that would be the ideal situation, they’re great sales executives”.

So I called him, and that was the beginning of the relationship where I was like, “Okay, well, I’d love to round out my business skills, I don’t want to just be a programmer, I want to learn how to sell and market too”. And that was it.

I mean, it sounds so unlike what you would expect from particularly the early ’80s where you’re a programmer, you’re a developer, you’re working on assembly code, what was it that the people saw in you to push you in this direction? Was it that you were just vocal, you want to start a business, you want to do XYZ, and they’re like, “You need some help if you’re going to get there”? Why did all these people come to you and say, “You’re good at what you’re doing, but you need to do this other thing”?

MB: I think it’s really because I self-identified at an early age as an entrepreneur without even using the term, and they wanted me to be a successful entrepreneur and to be a successful entrepreneur, you have to have more than one skill. You have to be able to do a new technology model, you have to do a new business model, and you have to have core values. They didn’t see a complete picture in me at the ripe old age of 20 there so they were like, “Listen, you don’t have it yet, you’ve got to go do this other piece”, and it was because my university professors and the people that I was working for at Apple both said the same thing. It was like, “Yes, you’re doing fine on the coding, but no, you’re not ready to be an entrepreneur yet because you don’t have your sales and marketing act together”.

It was Bill Jost actually at Apple who was at this group called PCMA, and I was in this group and they’re like, “Call my friend, Gary Gibson”, and that’s when I called Gary and then Gary’s like, “Okay, come up”, and I sat in that Oracle lobby for three hours waiting for Gary Gibson. He had totally forgotten about the interview and then when I came in, I just pitched him for a couple hours, showed him my games, did this, did a little razzmatazz, and he’s like, “Okay, we’re going to hire you”.

I started answering the 800 phone number at Oracle, that was my job, just picking up, talking to people, then moving into sales, then I was Rookie of the Year my first year in sales at Oracle, then I moved into marketing, then I started working almost directly for Larry Ellison, which was very weird and a lot of fun, and he started mentoring me very aggressively. That happened for 13 years until I started Salesforce in 1999.

That’s amazing. It reminds me, I remember I had a manager at Microsoft who I was in a meeting and we came out of the meeting, he’s like, “Ben, you are the only one in that room that actually understood the problem and the solution”, but then he’s like, “But you have the worst communication skills I’ve ever seen. Everyone’s on A, and we need to go from A to H, and you have to learn to walk people to A to B to C to E”, and it was a transformative sort of moment where now that’s basically what I do, is walk people from A to B to C to E, and it does speak to the power of people that will see what you want to do and tell you pretty plainly to your face that you’re terrible at it.

MB: I think that that’s really the whole point, which is that, look, I think my mother said to me, “You want to play better tennis? Then play with A players, play with people better than you’re at, and you’re going to become a better player if you’re playing with A players”. My mother is one of the top 200 Bridge players in the world, and she’s amazing, but every day she’s playing with these top players and that’s how she gets better, and I think that this is important for all of us, that we have to surround ourselves with great people and then you’ll become greater.

Building Salesforce

So there are so many things about Salesforce that are innovative, the No Software, software over the Internet, the subscription model, your whole PR campaign, and something that I’m curious about is it seems to me that all these things actually work in concert, they’re all necessary pieces of the whole pie, and you’ve talked about that and about how those different things go together. Was that clear to you when you got started? Or was it just like, “Look, the internet is a thing, applications should run there”? What was the key insight that was like, “Salesforce is what I need to build”?

MB: I was sitting in my office in Oracle in 1995, and I’m using a browser and I’m working on a number of new products that I’m building for Oracle, because I’m a software developer, by the way, and then I’m working on these products and designing them and the whole thing, but then I am using the browser and I’m like, “Oh, this is really different, that’s actually really different”. Then I was using Amazon and I’m like, “Oh, this is very different, how did they build this?”, because they’re a huge customer of ours, and said, “Now what did they do?”.

Then I got so freaked out, okay, that I went into Larry Ellison’s office and I said, “Listen, think I need to leave the company because there’s this huge thing happening in the world and I’m not sure what I’m supposed to do about it, but I know it’s something”, and he goes, “Oh, Marc, listen. Your problem is you just need a sabbatical, you’ve been here for ten years, you got here right out of college in 1986, now it’s 1996, take three months off and enjoy yourself and go study this Internet thing”.

I’m like, “Okay, I’ll go and do that”, and Larry at the time was very focused on the network computer, if you remember. Network computer, that was a whole different idea, or I guess you could say, I don’t know, I have a network computer, it’s called my iPhone.

Yep. It’s the world network.

MB: Larry was very set on network computer, but I was really focused on that software was going to change, applications were going to change, the way we delivered software was going to change, that there would be a new technology model and a new business model, complete different system, and the other thing that I noticed is that in the software industry, when there is a new technology model and a new business model, everything is up for grabs, so that’s a very scary thing as an entrepreneur because you have to be ready then to jump and go.

I went and did my research and I went off and I rented this little house in Hawaii, and I’m in a little white sand beach and it’s very cool and I’m there hanging out and I came up with a ton of ideas. The other thing is that at that time I had been doing venture investment for many years, so a number of my companies were going public at that time, and it was very cool and I was making some money and looking at the Internet, and I’m like, “Where is all this going?”, and so I bought this URL You.com, and I held onto it for a number of years. I recently gave it to a friend of mine who started a company so he could use it for a search company, an AI company, which is amazing. If you haven’t seen You.com, it’s kind of great job.

That is quite a gift, I can’t imagine what that would cost these days.

MB: I have a lot of URL stories and gifts I’ve made, including appstore.com, I gave it to Steve Jobs.

That’s a good one too.

MB: That’s a good one too, I have a lot of good ones. But I would say that Bill.com was another one, you probably know that company. I mean, there’s a lot of, for whatever reason, I was really into URLs for a while. What would all these things be and trying to conceptualize the future.

Then I was just like, “Oh, wow, this is really great”, and then I came back to Larry and painted the vision of what the future would be for my perspective. Tell them about You.com, and I was going to want to build a cloud HR company called You, and he was like, “Listen, it doesn’t seem like you’re completely through this process, do you want to take a little bit more time off?”, I’m like, “Good idea”.

I went to India, and I went to India and hung out with gurus for a few months, and that was very cool, and really unusual, and exciting, and crazy making all at the same time and they were also, then the gurus were saying things to me like this, “Oh, Marc, you’re not quite ready to start your company yet, are you?”, and I’m like, “I guess I’m not, I don’t know”. “But just remember that when you do, you don’t want to just build another technology model and business model, right, you want to do something that’s also going to contribute positively to humanity, that you need to actually do this”.

I even got a text from one of my gurus recently, I was so impacted by it, I saved it, I’m like, “Whoa, this is incredible”, because they were more interested in, like I said earlier, you can build a product, you can build a business, and then what are you going to do with your values? I think probably because of that idea that not only was I going to build Salesforce, but I was going to have the 1-1-1 Model, and the 1-1-1 Model was we took 1% of our equity, our profit, and our time, something now that has been copied by 18,000 other companies with our Pledge 1% Initiative, and we’ve been able to give away about a billion dollars, run 100,000 nonprofits for free on our service, and also we have done, 10 million hours of volunteerism through the company so far. So, that’s cool.

I think that that was very important that yes, we did the new technology model called cloud, and we did social, we did mobile, we did AI, we did data, and we also did a new business model, we did subscription, and we’ve been doing consumption as well, and now here we are. We also did the 1-1-1 because we looked at a lot of these other companies, even other software companies, and I’m not going to use any names in this segment, but they weren’t the values — well, the values were not at the level that I was comfortable with. I just wanted to have a company that I would be able to look like right now, I look back at 25 years of leadership at Salesforce, and I’m very comfortable with our values and the leadership provided.

This is a question I was going to ask you at the end, but I’ll jump to it now. You definitely seem to have a burr under the saddle again, or sort of the fire in the belly. Is that a return that is about this AI stuff? Is there a bit where because you infuse the philanthropy aspect into Salesforce, that has kept you energetic and going? I mean you model yourself after Amazon, Amazon’s founder is on a boat somewhere, and you’re sitting in an office talking to an analyst in Taipei.

MB: Well, actually I think Jeff [Bezos] has done a brilliant thing putting [Amazon CEO] Andy [Jassy] in place, because Andy’s doing the operational things, and Jeff is able to do a lot more strategic things, and also Jeff has some interests outside of Amazon, which is nice, launching some rockets and doing some other things in the media world, and other personal things.

It’s a funny story, I was recently actually at a resort in Bora Bora, and I was walking down the street and then Jeff Bezos was there, it’s not the only time I just literally walked into him. Even when I started Salesforce, I was in a hotel in Seattle and I was coming down to have breakfast, and in the restaurant there’s only one person sitting there, it was Jeff. And I’m like, “Jeff, this is amazing, why do we keep seeing each other?”, it was really kind of a bizarre thing, but he’s a great entrepreneur. One of the best.

One more question about Oracle. When you look back, is there a bit that Oracle missed the cloud? Or just they weren’t at the stage in the company to take advantage, and because they were a big company by that point and they were doing big things, and it needed to be something that was seized on by a startup, it couldn’t have been done inside of a company like theirs?

MB: Look, I really think if you want to understand what that situation is, and I try to make sure that my company does not get in this zone, because it’s easy to get in the zone. You should read a book of somebody who is passed on, is an amazing person, amazing. His name is Clayton Christensen, and he wrote this book called The Innovator’s Dilemma, and basically what it means is you can be an innovator like me or whatever, is coming up with something new, and then all of a sudden after two decades or three decades or four decades, that thing that you built, “theoretically” isn’t great anymore.

Now somehow because, and I’ll tell you how we did it, if you’re interested, but somehow the idea that you can take something forward into the future and keep it modern, that’s not always true. Sometimes what you have to do is bifurcate it and then create a new business. The best example in software is probably like Microsoft had built Windows, they created this monopolistic situation. They added Office, they started this aggressive bundling, they looked to disintermediate their nascent startup competitors by doing very nefarious business practices, it’s all written very well in a consent decree.

So that story, then they transcended the Windows Innovator’s Dilemma, which of course we all know, and Windows is not exactly where everybody works, per se anymore, and now, but they have another product line, they have Azure, so they built another pro platform, basically. So that’s not our story. For good or bad, our story is that the code that we wrote 25 years ago, and I can go through our vision of a metadata model, combined with a multi-application framework, combined with a deep sharing model for users, can transcend technology paradigms and by doing that, we have been able to get to this moment where we’re going to talk about at some point what I’m super jacked about right now.

Agentforce

To what extent was this moment actually on your timeline? You walked through in your Dreamforce presentation the different waves of AI, and when you look backwards-

MB: Thank you for watching my keynote, that’s totally awesome.

I felt like that was when I got the first sense you really still had, or it felt like, not to say you didn’t have fire, but there was a new fire, is certainly the perspective that I got from you.

MB: Well, I’ve never been more excited about anything, let’s just say that. I mean, from a business perspective or a technology perspective, I’ve never been more excited.

Walk me through why.

MB: Well, I think we’re about to change business itself. I think that for enterprises, they’ve bought software from companies like mine and others for many years hoping that they were going to be better companies, lower cost companies, easier-to-operate with companies, companies with stronger employees, more capable customer relationships, and we’ve been able to make progress on all those fronts, for sure. But now everything is going to completely change.

Well, so the theory of the change though, and this is why I thought this was pretty interesting, and you’ve obviously been on, if we’re going to mention other company’s names, a bit of a tour these days comparing and contrasting, I think is a nice way to put it, your approach to Microsoft’s. By and large think you are onto something, in that you go back to the original wave of IT and it was all back rooms, and it was top-down implementations and actually, the most jobs that got eliminated by IT arguably happened at the beginning, and so you come out and you put out this wave idea of AI, and you have this brief segment on Copilot, you sort of skip over it and you’re like, “No, we’re actually to the agent wave”. A lot of the demos were, I mean, you said some nice words about people supervising them, but they were sort of job elimination demos. Is that a fair way to characterize it?

MB: No, so I think that that’s not true, and I think that I can just walk you through some of that myself.

Go ahead.

MB: How I look at it, what it means, what that idea is from a tech perspective, but there’s a lot of other dimensions of this reality, and I can walk you through it. I can give you an example, right? You want me to give you an example?

Yes, go ahead.

MB: You see what I’m wearing here?

You’re wearing some sort of brace/casting on your ankle?

MB: Yeah, because I was in Fakarava, do you know where Fakarava is?

I do not know, and I’m going to be super honest with you from now on!

MB: It’s not so far away from here. It’s in French Polynesia, it’s in the Tuamotos, and it’s a great place. It’s like 50 million years old, very old, fully degraded volcanic zone that’s called a series of atolls, like several thousand atolls and I’m in Fakarava, and I just jumped off the boat, and all my scuba gear on and blah, blah, blah, blah, blah and I have these high performance fins, and for whatever reason, I just snapped my right Achilles. I’d had a lot of Achilles issues, but only on my left side, so now I have this snapped Achilles and I’m putting it back together. Now I’m coming back to here in Hawaii, and now I have to go to the hospital and get an MRI. So you’ve ever had an MRI?

Yes.

MB: Okay.

I had to get one on my wrist, it was one of the worst experiences of my life, because I had to be propped up on my side to do it, it was awful.

MB: And you get in the MRI, and maybe they use contrast or maybe they don’t, so they can get a really good picture of what your wrist looks like or my Achilles looks like. And then they kind of put you in a boot, or your cast, or a frame, or whatever they did to you and sent you home, you probably had that experience.

I did.

MB: And let me ask you a question. Did somebody call you the next day and say, “Are you feeling okay? And did you drink all the water we told you to flush out the contrast? And are you taking your meds like we told you?”.

Of course not, and I’m also in the Taiwan national insurance, so it’s definitely not happening.

MB: So now when you went to and said, “I’ve got to reschedule my doctor’s appointment, and I think I need a repeat MRI because I think I just did something else, and things are not going exactly right, and I need to get my labs drawn,” how hard was that to do and get all of that done all in one motion?

Very difficult.

MB: It’s not like you and I are the most dysfunctional human beings out there. And then once you went to your doctor’s appointment, did he call you the next day and go, “Great seeing you, and let’s just review again what I said, because you’re probably a little bit traumatized in the office because you had a kind of white coat syndrome, I noticed your blood pressure was a little high, but let’s just go through again what I said, and do you need me to call you again and repeat it again?” Did he do that?

No, definitely did not.

MB: But imagine this. Imagine a healthcare system that was architected like that. That all of a sudden I’m coming back from Fakarava. Do you know what the international motto of Fakarava is, by the way?

Nope.

MB: It’s, “I don’t give a Fakarava”.

(laughing)

MB: Basically working with the agent here and saying, “I need to schedule an MRI, I’ve ruptured my Achilles”, “Okay, we’re going to schedule that for you, and by the way, do you need to see our orthopedic surgeon before so he can check it and do the Thompson test?”, “Okay, yes, I’d like to get that scheduled”, “And have you had your labs done in the last 90 days? Because we’re going to want to see that”, “No, I haven’t”, “Okay, we’re going to arrange the lab, please go down to the lab before you do it, everything will be scheduled, thank you very much”.

This is all great vision though.

MB: Oh, no, that’s not vision. That is not vision.

Well, what gives you the confidence that it can be pulled off right now?

MB: Because I have a customer that is doing that and I want to use that customer’s name, but they’re very cautious right now about me using it because they don’t want to trigger everybody starting to really look at exactly how our technology is working, but I hope very shortly that in that specific healthcare example, because I’m just going to contrast it to what you said.

Yep.

MB: Think about all the things in our life where we wish there were people and there weren’t, and I think in healthcare, which is a pretty big industry in the US — oh yeah, it’s the largest — that it’s surprising that in the largest industry in the US, we don’t have enough people, and there isn’t a cost structure to be able to do all of these actually somewhat important tasks so that you can see, I’m taking my drugs, I’m drinking my water, I’m doing this. Yes, it’s okay, stop bothering me, I kind of want to have that experience.

So this is the response to the business model concerns. We don’t need to worry about seat elimination, this is going to be usage based, but it’s going to be additive.

MB: Let me just stay with my example first. I just feel like in healthcare, I feel so on my own, and everything I do in healthcare, I feel like if I’m not advocating for myself, the system doesn’t work.

Yep.

MB: That metaphor that there’s going to be some kind of co-intelligence with us, or an agent or whatever, Agentforce or whatever we’re going to call it, that is going to change. I think that we’ve even seen, even in the podcasting industry, we saw some crazy stuff from Google in the last couple of weeks where all of a sudden we thought, wait a minute, this sounds like a podcast that humans are doing, but it’s agents and agents are going to have a role in our world.

Now, what the problem has been, or is actually, it’s more about is than has been is these AI priest and priestesses that I love and I’m friends with them, they’re great people and then they’re like, “You know Marc, AI, oh, it’s cured cancer, by the way, oh, climate change pretty much over”. And I’m like, “Okay, well, that’s really good, and what movie was that?”, because there’s a lot of fantastical ideas of AI, I don’t have to go very far. I can pull out the Minority Report movie, it was written by our futurist, or he’s one of the writers, Peter Schwartz, or War Games or Space Odyssey or Her.

But we all know there’s parts of that that are kind of happening, but it’s not exactly that, and the funny thing is that I guess you could say some of this might’ve been oversold in the last couple of years, these LLMs, oh yeah. By the way, LLMs are the bridge to AGI, this is very important. Pretty much by Christmas we’re having AGI because we have LLMs. Explain to me, you mean these canonical word models that are where things are connecting this, this, and then that’s going to — and that’s AGI? Wow.

Well, I guess it wasn’t the AGI that we were expecting because I think that there has been a level of sell, including Microsoft Copilot, this thing is a complete disaster. It’s like, what is this thing on my computer? I don’t even understand why Microsoft is saying that Copilot is their vision of how you’re going to transform your company with AI, and you are going to become more productive. You’re going to augment your employees, you’re going to lower your cost, improve your customer relationships, and fundamentally expand all your KPIs with Copilot. I would say, “No, Copilot is the new Clippy”, I’m even playing with a paperclip right now.

AI’s False Prophets

I used that line, it is a good one. But there’s a bit where it is kind of interesting because you talked about in the healthcare model, there’s a bit where that is sort of Copilot-esque, but the placement of agency is in a different place. You framed it as the healthcare organization being able to proactively reach out and help out customers, whereas the Microsoft vision is customers using the Copilot of their own volition.

MB: I’ve yet to find any customer using Copilot at all, so that’s very impressive that you’d even say such a thing and give them any credit at all. But I’ll just say this, I think at one level where customers have been misled or maybe there’s two places customers have been dramatically misled, one is that LLM is AI.

Right.

MB: And two is that these models, each of you are going to have your own model and train and build your own model and that’s what we’re going to have to hype. Screw all this infrastructure and light up everyone into a nuclear powerplant because everyone’s going to need their own model. Let’s get training, get more processors, let’s train.

I just think we have to kind of break apart these two ideas. I’m just saying this as, oh, by the way, Salesforce will do more enterprise AI transactions this week, about 2 trillion, than any other vendor in the world. We are the second-largest software company in the world today but I do not believe in some of this narratives and rhetorics that’s going on and I think it’s a disservice, and I’ve been wanting to say this, and you also know two out of the five top performing models in the world, we wrote most accurate, most capable, we created prompt engineering, we’ve done a lot of the work in deep learning, we’ve done a lot of the research.

But when I hear some of this other stuff, because some of the consumer apps are great, I use them myself on my phone, I’ve got all these consumer apps, and I’ll be like, “What are the five ways to make sure that I improve the health of my healing of my Achilles?”, it’s very good at that because it’s super search, it’s basically what Google should be today but isn’t.

Well, you’re a person with very high agency. You’ll go out and try to solve your own problems.

MB: Yeah, that’s how I use Google also. If I didn’t have ChatGPT right now, I’d be on Google going, “What are the five ways for me to improve my Achilles?”, and I’d probably end up with some article. But instead I’m getting a kind of a form narrative that comes out of an amalgamated text model where they have trained this amalgamated text model and then the algorithm I put in my search and out comes the data.

But I’ve even had friends of mine, and I even heard on this crazy podcast, the All In Podcast, them try to explain that LLMs are data storage mechanisms and then that they have new ways to manage data, enterprise data and I’m like, “God, I need to know about that because I still use disk drives”. So let’s actually talk about how you actually do store, I think there’s a broad disservice — I’m not trying to pick on All In or anybody else.

You are trying to pick on Microsoft. But sorry, continue.

MB: Microsoft. Look, if you’re the number two software company—

Oh no, I have the quote, I’m going to drop this this on you. You said in your book, “Position yourself either as the leader or against the leader in your industry, every experience you have, a journalist or potential customer must explain why you are different and incorporate a clear call to action” — I mean, I read that, I’m like, “Oh yeah, he’s doing exactly what he said”.

MB: Hey, I think that’s very good advice for an entrepreneur. I think an entrepreneur could have really good advice by willing to make an aggressive statement like that.

By the way, I’ll make an aggressive statement on that on Twitter or wherever, I like making my little tweets, I actually test out my marketing messages ahead of time on Twitter because I’m a little bit like a comedian who’s going into a small club trying to try out jokes, and then of course I’m going to get a couple trolls or whatever, and you’re going to come at me no matter what. But sometimes when I’m like — and by the way, that’s how I figured out that Copilot wasn’t working, I was going after it, and then no one’s pushing back at me going, “I love that thing”, “What, are you kidding me? I use it 24/7, it’s amazing”, “They’ve changed my whole company, do you know what they’ve done to my cost structure?”, “Oh, and my customers tell us they’ve never had a bad experience with us, oh my employees, they love it”. No, no, no, didn’t get any of that.

That’s where I was like, okay, I’m in the zone. But I think that the issue is that for us, the real opportunity in AI, because we’ve gone through — I would say first of all, ten years ago I thought this was about to happen, and I went to have this existential crisis in my own leadership where I was like, “Well, the AI wave is here and I need to crank it out at Salesforce”, and I built something called Einstein. Which is our core, and we did machine intelligence, machine learning, deep learning, and now generative, and the thing that I learned when I did Einstein was that this AI needs to kind of be baked in. That this is too technical for customers, you’ve got to just make everything better. Another great example is Google Search. There’s a lot of AI when you’re doing Google Search, it just gets better.

But in this case, however, we can see that there’s another leap that can now be made with AI with agents, because most companies today do not have an agentic layer on top of them. In fact, when I’ll say to most CEOs, everybody should be architecting now their agentic layer, most CEOs will say to me, “What is an agentic layer?”. I’ll be like, this is about humans with agents driving success together, and I would say for 25 years we’ve been focused on building and automating all these customer touch points for companies.

I’m wearing this hat from Under Armour, that’s a huge Salesforce customer or Disney, I’m wearing their shirt, I only wear clothes from my customers. I think that it’s fascinating that this idea that we’re able to automate all the customer touch points. In the last few years, what I was mostly excited about was we’ve built this incredible data cloud on top of all those customer touch points that federates, which means it can connect into any other data repository, lock it in and bring the data together. But then I just laid this agentic layer on top of the data cloud and boom, customers are having this unbelievable experience and this experience is their costs are going down, their employees are highly augmented. They’re having better customer relationships and their fundamental KPIs of their business are better and that’s where I’m saying, “Finally”, this is what I’ve always wanted to get done in my career and it’s working.

The Force AI Platform

Is there a bit where one of the big things when you built the SaaS model and then you built the past model of on top, Platform as a Service and companies could come in, you recognize we can’t meet every use case, “We built an application layer for developers to come in and solve that”. Is that an area that is potentially obsolete in the long run? Does AI just solve those problems or is it just that developers need to approach those problems in a different way via this sort of agentic model?

MB: I think this is another area where there’s broad confusion, and I’m so glad you asked that question because I think it really gets to the heart of our core strategy in our company. Now, here we are 25 years later, now we have about 230 petabytes of data in our system managing our customers. We have many apps, not only our apps, but all the apps that our customers have made with our platform in our system. We have our data cloud, it’s all one piece of code, right? It’s like one system. We’ve even tried to rewrite now all the applications that we had bought and companies we had bought. We’ve done about 60 acquisitions over 25 years, and to create one piece of code, we call it more core, just a solid version of our core with this huge canonical data model and data lake over at Data Cloud, and then we’re adding this agentic layer.

Before you add on that, why did you do that? Why did you build that way? You can look backwards now and say, “Oh, that’s actually a big asset to have rewritten all this stuff to be in a common place”, but if you didn’t necessarily know or did you know what made those architectural decisions? That’s obviously the back and forth to microservices and monolithic is sort of a thing that’s been a debate in tech for ages. Why did you go that direction?

MB: This is another great question, and I can’t give you absolute clarity because we have written and rewritten our product many, many times over two and a half decades. But our fundamental, what I would call our market texture, has not changed. That is we knew we wanted to have a metadata system, we knew we wanted to have a robust sharing model, we knew that we wanted to have deep data integrity, the security that our customers needed, we knew we needed to have robust applications, we knew we needed to have all these capabilities. Because we took a platform approach, which is what fundamentally was different from everybody else, that our stuff was all built as a platform. In fact, always more than 50% of our engineers have been platform engineering, so half of them were working on the apps, but half were always working on our platform.

Now, let me tell you why this is important, and I’ll even tell you about something that I just saw that amplifies this, because when we built Einstein and put it in the platform, it’s just rolling along, not so unlike in Google, the Google Search, the AI is in there, you’re never going to see it.

But when we added the agentic layer, if you’re going to get these agents to work, you want these agents to work, what do you need? What are the ingredients? Let’s say we’re making a cake. You want an LLM, but that’s just like sugar. That’s going to make it great, okay, you’re going to have to have a pretty advanced RAG [Retrieval-Augmented Generation] technique or an ensemble RAG technique. You’re also going to need data because LLMs by themselves don’t know anything, they’re just an algorithm, you’re also going to want metadata because the data is better if it’s explained to the LLM what the data and metadata are. You’re going to want some kind of hyperscaler approach so that you can scale up and down as you need to go. You need to have some kind of user interface with the customer where your employee is participating because while all this is great, no one’s LLM, no one’s capability, and ours is I think the most accurate, lowest hallucinogenic out there, is 100% accurate yet, it just isn’t. We’ve all had those experiences where we put the query into the LLM and we get something crazy out so you’re going to want humans to be able to mitigate and be part of the process or keep the humans in the loop and that means the apps have to be integrated as part of the process.

If you can do these things, then you can deliver huge value to companies, and I have a lot of stories now in my pocket, and I’ll give you one story like just happened yesterday, which is I’m talking to the CEO of a very large telecom company, and she just wants to say, “Explain to me what this technology is going to mean for me?”. And I said, “Let’s just have a conversation. But first of all, let’s recognize two things. One, have you ever had a conversation in your career about an agentic layer?”, “No, I haven’t”, “Two, have you ever had a conversation in your career about humans with agents working together?”, “No, I haven’t”. Okay, so we are in a new world.

Now, let’s have a fantasy about what does an agent-first telecom company look like? What are companies going to look like that are agent-first that are in telecom? We riffed on healthcare right before, now let’s do telecom. That’s a good example because if you’ve ever actually tried to call a telecom service, it’s a little bit funky, and what would be the ideal thing where you had a high performing agent for customer service, but also high performing agents for your employees to give very technical information, because a lot of these telecom companies, they have very technical implementations all over their customer sites.

Even in my home, I have a rack and it has some very sophisticated equipment that I don’t understand it. Some people come here and they come to work on it and update it and upgrade it. Those people need to be augmented through an agent, it can be much better, and in all these cases, an agent-first telecom company means that, whether you’re working with their employee or you’re working with their systems directly, you’re going to have a more fulfilled, accurate resolved experience where before you might be left hanging.

There’s not going to be any hanging, it’s going to be like, “No, we’re going to reschedule your MRI, it’s done, I just emailed you the information”. “Oh, who are you?”, “I’m Agent So-and-so, I’m an autonomous agent working on your behalf, I’ve already reviewed all your medical records, I also looked at everyone who’s ever achieved this issue with the Achilles before, I looked at all the records of the whole hospital to see what the best practices are, these are the three things you need to do right now”.

So this bit about how you structured Salesforce over time on the platform, this more monolithic approach, is very compelling to me as to an attestation for why you can succeed in this and why you’re differentiated from other folks. Does that mean, though, to the extent that’s correct, this is a winner-take-all game in the context of an organization? So this isn’t like tweaking Microsoft because we want them to partner on good terms so we can get access to all their data, it’s that, “No, we need to go in and, whatever data they have that we don’t, we’re going to convince organizations they need to just shift over to us completely so it can be in this model so we can actually operate on it”?

MB: I have a much simpler idea for you. Do you remember back in the mobile world when mobile was becoming hot and everyone had a mobile app and there were mobile companies and all these hot mobile venture started, and then we’re like, “Whoa, mobile CRM, somebody’s going to take us out”, and then we’re like, “Well, is that really true? Because here we have the cloud CRM, maybe we could just from the platform up and out and manifest on mobile”. Then we did from the platform up and out, it manifested on social and then the AI was inside and then everything was inside and then that’s where, for agents, all the agents are is an extension of what we’ve been doing for 25 years. This is not some separate piece of code that you’re going to implement, this is not a separate system that you’re going to train some new team in your company on.

Data and Disruption

Right. But to the extent there’s data in a company that’s not in Salesforce, it’s in other company’s systems, and Microsoft is pitching, “Oh, you can integrate your third party service into us”, that was their v1 of agents, now they have a v2 which you tweaked them on Twitter that sounds more like yours.

MB: I’m sure it’ll be a successful as version one.

This idea that you’re going to have these disparate services that you can bring into this agentic layer, I’m skeptical of that. My thought is having the data, LLMs are incredible, the extension of this which they can work over bad data, but if you actually want accuracy, good data and good organization is important. Can you extend, can you partner with other companies or do you need to get everything into Salesforce for it to really be useful?

MB: No, no, because I told you our data cloud is federating.

For sure.

MB: So the data cloud federates to other data sources, and that’s very powerful. It can go off to mainframes, it can go off to other data lakes like Snowflake’s, Databricks, yes, even the Microsoft one, the Google one, the Amazon one, and your proprietary ones as well. We bring it all in and then we build a canonical model.

Now, that idea that there’s many things to create a successful agent, but data is one of those things that actually is important, getting the data right and here’s the thing, one of Microsoft’s liabilities here is that that’s not what they do for customers, they don’t hold all these customers’ data. What is Microsoft’s most successful enterprise app that we’re talking about that holds very important customer or employee data?

SharePoint.

MB: Okay, the prosecution rests.

(laughing) I served that up for you on a platter, but you invited me to do so, so I had to do it.

MB: But you’re very smart, you know this whole industry. Am I off base here?

No. The basis of their enterprise—

MB: If you don’t have the data, you’re not going to have the intelligence, it’s not just the LLM. This, again, I have to just come back, I think that there are people who are doing a disservice to the industry on podcasts, priests and priestesses, other vendors when they say, “Listen, now that we have this, whatever it is, AI, something or another, LLM, but now all of our problems have been resolved”, and the reality is that is not true.

The reality is that to create intelligence, it needs many components and then it’s the integration of these components in a proper platform by he way that We talked about, sharing models. For example, we have a lot of banks as customers, every bank, my account manager looking at my account balance cannot see your account balance. Did you know that? There’s actually laws and regulations country by country over this? So you can’t just amalgamate all the data and then say, “Oh, all the data, then we’re going to train the model on the current set of data with all everyone’s account balance”, that’s not exactly how it’s going to work.

It needs to be a sharing model, this person can see this thing, this person cannot see this other thing, that’s why you saw in that Gartner Report that got published, Computer World wrote about it that there was this idea of spilling data. I’m characterizing it, but I think it was very clear that they found some very serious anomalies and then they said, “Things were not set up right, there wasn’t a sharing model put in place, there wasn’t a security model put in place, and customers ended up with this data spill”.

Well, this is a problem you had to solve with Salesforce at the very beginning because people were skeptical of multi-tenancy and having everything on one server. So this permission structure being at the root of things, that is an advantage you had from the very beginning.

MB: And for sales organizations, which is where we started and managing these sales hierarchies, and you have this big sales organization all over the world and you’ve got 10,000 salespeople and now you have all these sales managers, one sales manager cannot see what every other sales manager is doing. You can have closed and open sharing models, you can have discretion, you can have all these different pieces, but sales reps should not just be able to see whatever they want. They should be focused, put their blinders on, “Do this”, “Get this done”.

Well, we have gone a little bit over, I could easily go for another hour here, but I am a big Clayton Christensen fan as well, and when you reflect on this and this opportunity, do you think you cracked the code for avoiding disruption, or does it turn out that actually Salesforce was just really well suited for this opportunity and you have the chance to be the disruptor all over again?

MB: I think that, in many ways, I guess I would say there’s two things. One I could say is happenstance. I didn’t really expect to be on your podcast today talking to companies that they can totally transform themselves and think about how they’re going to become agent-first companies, whether they’re healthcare companies or banks or tech companies or communications companies or governments. On the other side, I could say we always felt that the beginning with our Platform as a Service approach, which is a term I actually came up with back in the late ’90s, that the platform would always carry us through that, the fundamental truth of our industry over many years is it’s not about the app because the apps tend to be very temporal, but really it’s about the platform and the data over time.

Here’s the Oracle executive, he’s still here!

MB: Yeah! And having platform and data over time and workflow and metadata that this is extremely important. Now that we’re in the age of AI, we truly are, and in the age of agents, we truly are, and as we start to define our agent-first companies and as we say to our ChatGPTs, tell us, “Describe my company if it was agent-first”, “What would be the first five things where I could offer more value to my customers?”. It’s the data and the metadata and it’s going to be the RAG and it’s going to be the LLM and it’s going to be all of these things combined, and it’s going to turn into this incredible agent that’s going to really take you and your employees and your customers to new heights.

It’s incredible what it is. I mean, you could start to ask these questions, “What if my workforce had no limits?”, “What if I could start a company and I could start selling and servicing at scale?”. We’ve worked with customers for years to build their sales forces, their service forces, marketing forces, now we’re building their agent forces, and I think that for many CEOs of large companies like myself, they have never had a conversation about an agentic layer. They may have never used the words or they don’t understand when we say agents and then some customers are getting it so fast, they realize that problems that they have had now for a generation of their leadership can be resolved rapidly by implementing this model.

Zero marginal employees.

MB: Customers expect more, they are going to demand this to happen and you do need, like you’ve been saying, a trusted and secure platform that’s scalable, that’s accurate, that’s easy to customize, that’s got the AI built in, that has the apps built in too. The metadata platform, the ecosystem, the community of trailblazers, all of these things are important, and that’s Salesforce.

I’ll give you one last story, one of the first customers we gave this video to was Wiley, the textbook company, and this has been a customer for a long time at Salesforce, and they’re always a very much a first mover on a lot of these.

They’ve appeared at many Dreamforces over the years.

MB: Many Dreamforces, incredible. Everybody likes them because they don’t have propellers on their heads at Wiley, these are textbook people, and our job is to make the textbook company better.

At this time, all the kids are going back to school and they’re ordering their textbooks, it’s like Wiley Super Bowl, they need to order more textbooks, some service, some scale up and add gig workers. No, they don’t, because now their service teams, who might’ve been facing these increased back to school demand, well, now Agentforce is handling all that for them, and that is a moment for our industry, that it’s not some big hype cycle. This is code that is shipping now that’s available for companies of any size to deploy to be more productive, augment their employees, improve their customer relationship, increase their KPIs. Right now, it’s not a fantasy.

Also beware of the false prophets, that’s why I feel like I have an obligation to just let people know how bad products like Copilot are, but also some of these narratives that are coming out of these other vehicles where some of the stuff is just not true. That’s why I really like Yann LeCun’s tweets and this kind of stuff, and some others who I’ve seen including Linus [Torvalds] who invented Linux going, “In five years maybe or ten years with a different thing, not an LLM, but maybe a multi-sensory model or a world model”, but we have to say what we can really do now and what really works and what does not, I think that we have an obligation to say that today.

Marc Benioff, you said a lot, I enjoyed it greatly, and I look forward to talking to you again soon.

MB: Oh, it’s great being with you. Thanks so much, I really appreciate it, I hope you’ll invite me back.

I hope your Achilles feels better.

MB: Thank you.

---

This Daily Update Interview is also available as a podcast. To receive it in your podcast player, visit Stratechery.

The Daily Update is intended for a single recipient, but occasional forwarding is totally fine! If you would like to order multiple subscriptions for your team with a group discount (minimum 5), please contact me directly.

Thanks for being a supporter, and have a great day!