Marc Benioff On The Future Of AI

If you’re like a lot of CEOs these days, you’ve come through three psychological phases since generative AI first joined us in late 2022. Phase one: Wow, this is incredible! We have to get going with this! Phase two, circa mid-2023: Wow, this is unreliable… and risky. We have to be careful with how we use this. And now, in 2025: Will this ever actually make me money?

The answer, if you ask Marc Benioff, founder and CEO of Salesforce, is yes, absolutely. But only if you do it right. And right may not be—is probably not—by engineering what he calls a “DIY project” where your own team plus some experts, plus a big LLM model like those from Google or OpenAI are forged together into a homegrown product.

For Benioff, the answer is more likely to be through a third party with technical chops and experience who will be deploying tools you can plug into and use, rather than create on your own. You don’t care who builds your word processor, right? You don’t even call it a word processor anymore. You just type.

Is he talking his book a bit here? Of course. Benioff, our 2022 CEO of the Year, is not only an in-his-bones technology geek who came of age writing programs on a Commodore 64, but also one of the most talented technology salespeople—reared inside of Oracle back in its heyday of dominance—American business has yet to produce. With Salesforce, he built a $35 billion (2024 revs) juggernaut by wooing companies to take their first leap into cloud-based software at a time when most of us could still hear CDs whirring inside our desktop computers.

He was ahead of his time then, and he thinks he is once again with Agentforce, an agentic layer (in the parlance of Silicon Valley) that has AI “agents” augmenting the efforts of human workers, fed by the terrabytes of data you’re already feeding Salesforce in the very likely likelihood that you’re already his customer.

The key to AI, he says, is above all else just getting going. If you want to win, you’ve got to play. Get it up and running and doing things for you, rather than just tweaking tech. What, in his vision, is it doing? Expanding your business capabilities without expanding your human workforce. He’s got client case studies: homebuilder Lennar goes 24/7 with homeowner services as a new line of business; Disney keeps its “cast members” current and able to sell one of the broadest suites of products in all of media and entertainment; retailer VF’s revitalized website that gets you exactly what you want, without people, sooner.

Finally, he has his own shop, Salesforce, that includes 9,000 customer support workers now supported by the company’s own AI products, making Benioff one of the most experienced CEOs in the world when it comes to deploying—at scale— what he calls “digital labor.” And that’s why Chief Executive reached out to him for help as we all pioneer this new frontier together.

“I’m probably the last CEO that you’ll meet who only managed humans,” says Benioff. “All of us going forward are going to manage humans and agents together.” What follows was edited for length and clarity:

You talk to a lot of CEOs. When they ask you about AI these days, what are they asking? And what do you tell them?

A lot of CEOs are embarrassed that they don’t know more. Maybe they’re thinking they should have done more. Maybe they have a failed deployment now on their hands from some DIY project where they tried to do it all themselves. We hear a lot of stories like that. What we’re trying to do is give these CEOs wins, one win at a time. Like this morning when I got up, I went to vans.com. I saw that they deployed Agentforce. You can see it for yourself on the app. The icon in the middle is now Agentforce so you can order shoes, you can ask questions, figure out how to interact with the company to get your support issues done. It’s a whole new way to navigate their site.

Their CEO, Bracken Darrell, he’s amazing. He was the CEO of Logitech before this, now he’s the CEO of VF Corp. They have brands like Vans and others. He sent me a note saying, “Here’s a little taste of the reaction to our new site. Direct customer feedback: ‘I hit the new website. Everything changed. It confused me at first, and then it hit me, this new website is awesome.’” He had to send that to me this morning. That’s where I was like, “Whoa. This is, like, amazing.” That’s where I’m like, “All right. How do I continue to help these CEOs kind of get those kinds of wins?” That’s my main goal, to get that to happen. I build the relationships with these CEOs, and then what I try to do is motivate them to get this stuff turned on.

What you have to remember is what we’re trying to be the catalyst of is the biggest, most exciting transformation in the history of technology, which is the digital labor revolution.

This was the phrase you used when I heard you speaking at an event. You said, in essence, “there is the potential for the first time to have growth without additional headcount, without additional labor, and it’s through a new form of labor, which is digital labor.” What is the promise, broadly speaking?

It will touch every company in the world. We all have the ability now to extend our human workforce with a digital workforce. There’s no question that AI has this capability.

At Salesforce, we see it ourselves. We’ve deployed a site called help.salesforce.com. I have 9,000 support agents. So, six months ago, those 9,000 support agents were using our service cloud app, and customers would come to our website. We tried to deflect them where possible. Then, a lot of them would move into that support mechanism, and we would manage their cases and escalations and all those things, right? But now those human workers are working alongside digital workers. There are thousands of agents at Salesforce that are working hand-in-hand with human workers.

Do I still need 9,000? Probably not. Probably need about half as many. I can redeploy because those 9,000 are now being complemented by thousands of other workers. So I need to rebalance my human and digital workforce.

The one thing I’d love to get to with you is leading the culture change. You’re a living lab for this. What are you learning about what it takes to lead the AI transformation that you’re seeing and wanting other people to do?

It’s that scene where Indiana Jones has to take the leap of faith. The bridge is there, but we have to realize we can really do it. This is nothing that any of us have had training on. This is another bridge that CEOs will have to cross. It’s a leap of faith. It’s complicated because you are doing something in business that has never been done before. When we talk about a $12 trillion digital labor opportunity, this will impact every business in every geography around the world.

I live in a small rural town. In my small rural town, I think all the time about, “How is this really going to impact us?” But I’ll tell you, if you go across the street to the hospital, we don’t have all the physicians and experts that we need. But with AI, we can have a lot more capability. If you go down to our local schools, which are only a couple of blocks away, you know what, our teachers and tutors and mentors are all going to be augmented with this capability. So, in the case of schools and in the case of hospitals, things are going to just get better. In some places it’s probably going to remain the same for a long time, like construction. Tourism, it’s probably mostly going to be the same. It will be a long time before an autonomous plane is landing.

When you talk about augmentation, what do you mean? What’s your sense of the augmented worker a few years down the line here?

In the case of Gucci, we put our call center product in their Florence call center, and we thought, “Oh, now it will be more efficient. They can reduce the number of call center workers.” Instead, revenues went up more than 30 percent. They found that their call center became a sales and marketing center and also that they didn’t have the training before. RBC Bank is finding the same thing. They have thousands of call center and customer support agents who can now sell mortgages. So, that is augmentation—the ability for humans to do something they could not do.

I’m an investor in an amazing company called Artera. Artera is the first AI platform for urologists dealing with prostate cancer. What you can do if you’re using Artera is basically have the same capabilities as Peter Carroll, who is perhaps the top prostate cancer doctor in the world at UCSF. He has the intuition and the datasets and the capabilities. Maybe a lot of urologists, especially the ones in our little town, don’t have that—but now they do. You put your labs, your data and your scans into Artera, and it says the standard of care is this. So, that augments that urologist or augments that call center worker or augments whoever it is to make them a little bit smarter, a little bit better than they might’ve been.

The core of all of this, though, like all things about culture when people interact with technology or with each other, comes down to trust. Give us some thoughts, since you’re in the game already with this, about the role of trust and what leaders need to do to develop the trust for this to really be what you see it could be.

If you go back and look at our last interview together [in 2022], you’ll see that trust was a major theme because trust is our highest value at Salesforce—it always has been. Having that trusted relationship with all of our stakeholders. When we deployed our AI platform, Agentforce, the first thing we built was our trust layer because we didn’t want all your corporate data going to one of these LLM companies and them trading on it. So, we built a trust layer.

Trust can mean a lot of different things. It can mean that your data is not our product, that we’re going to protect your data. Trust can also mean that this technology is accurate, but LLMs are not 100 percent accurate. Even in our own use of our Agentforce platform, it’s only 84 percent accurate, and that’s at scale with the best deployment.

Some of our best customers—our best media company and our best bank—are in the low 90s, but it’s not 100 percent because the LLM is this weird architecture where you’re putting a word, a word, a word, a word, a word together, and those words do not create perfect accuracy because they have these kind of spiral or spider-type networks off of the words, and then all the words are connected together into this network, and then that’s how you get this answer.

But the answer may or may not be accurate. That’s the nature of a large language model. It’s just not the nature of AI and how it works. At some point, maybe there’s 100 percent accurate AGI, but that is not where we are now. So, the models are only so accurate.

That means there is a role for humans. When you’re on help.salesforce.com getting your issue resolved on our website and things are not going exactly right or the LLM is confused, boom, you can hit a moment, and you can just go right over to our humans. Because in our architecture, and this is the key part of it, it’s not just about AI. The AI and the apps are together. It’s about, “Oh, we’ve built it so that humans and AI are working together.”

The first layer in architecture is our apps, all those apps. The second layer is all the data, the amalgamated data and the federated data. The third layer is the agentic layer. All of it is just one piece of code. But when you’re on help.salesforce.com and you’re like, “This is bullshit, get me out of this,” boom, the human then gets presented with a screen with all the conversations you had with the agent and can see your customer history, everything is right there. And the humans and the AI are working together around one dataset. This is the critical part.

Where do you see this going from here? And what do CEOs need to do now to prepare themselves, not just to lead their organizations but also not to get left behind?

Well, technology is constantly getting lower cost and easier to use and it’s constantly marching forward. So, since Salesforce kind of came along 25 years ago, we did the cloud, we did social, we did mobile, and we did AI, but we did predictive machine intelligence, machine learning. Then, we got into generative. Now, we’re going to do agents. Soon we’ll be into robots. These are different waves of technology that are kind of coming at us. Even today at Trail DX, which is our conference in San Francisco, there’s a robot standing next to a human on the check-in line checking customers in and answering questions. We can do that because that robot is just a manifestation of Agentforce. Pretty interesting, right? It’s the next level. But you know that robot’s not giving me 100 percent and so there’s a human standing right by it. It’s a metaphor, but it’s also a physical reality.

I would say there’s three big changes happening in an immediate future that we can say are definitely true. One is that these LLMs are commodities, that it’s not about OpenAI, Anthropic, xAI, U.com or Perplexity, or any of these. It’s really having a platform that you can mix and match and bring in the model you need when you need it. And that these models are mostly a commodity.

I’m ready to delete ChatGPT off my phone. I’m not really using it anymore. I’ve moved to Grok completely. That’s the most interesting app for me, the most accurate, the best user interface, the best voice, all those things. Six months ago, that wasn’t true. Six months from now, I don’t know what is true, but I do know that the model is a commodity and more models are coming. The LLM is maturing as a concept, and there’ll be another model that will replace or extend the LLM in a different way. The LLM is like a part of the core infrastructure. The way chips or disk drives or memory are kind of a core part of infrastructure and technology now, the LLM is that way.

Two, companies are investing hundreds of billions of dollars in data centers. For us at Salesforce, we just have a much lower-cost way to deploy. This will be incredible for us. We’re taking advantage of these low prices to deploy on more substrates.

Three, there is definitely some super accurate model down the road. People call it AGI. Maybe there’s some other kind of model, a different description of it coming. AI is only going to get better. So, commodity models today, B, low-cost infrastructure, C, some incredible future state that we’ve seen in movies. This is an opportunity to transform our businesses.

What do CEOs need to do to prepare?

I’m just trying to get them to get a lot of base hits right now. Get on base, have your pilot, your lighthouse app, show your employees and your customers what you can do.

There’s this great company in Miami, Lennar. They’ve got these co-CEOs, Jon Jaffe and Stuart Miller. They’re amazing. They’re one of the nation’s largest home builders. Their people came to Dreamforce. Their tech team came back. They called an emergency meeting with the CEOs. They said, “We saw something we think can radically extend our business. We’re going to become a lot more profitable and grow our revenues way more than we thought.” They said, “Prove it.” So, they had a hackathon. The hackathon went so well, they had five huge use cases. The CEOs got on the phone with me and they said, “Is this real? Are you going to back us?” I said, “Absolutely. I’ll do whatever you need.”

So, then I let them rip at it. Now they’re ready to deploy this technology. They’ll be able to service homes 24/7. They couldn’t do that before. They didn’t have the workforce. They’ll be able to sell products, heterogeneous products like mortgages, insurances to existing customers that they couldn’t sell before. They didn’t have the salespeople. They have all kinds of new ways to augment their company and to make it more profitable. So that’s amazing.

Another company right next door to them, a big one, Disney, is doing this incredible work where they’re augmenting their employees because they have these super complex products. They’re also taking some of their marquee products like Disney Plus or their hospitality products like their hotels and putting the agent in as a direct connection to the customer. This has never been done before.

Anything else to guide folks? They’ve heard a lot of hype. They’ve heard a lot of “you have tos.”

Put your failures aside because a lot of them now have these big failures. You didn’t get the huge AI magical boost that you thought you were going to get. The agent revolution is what every enterprise needs. This is the brass tacks of AI, the human work, the digital workforce, the digital labor revolution. You are either going to be part of the digital labor revolution or you are not. So, you’ve got to embrace it.

But none of us have any training on this, so you’ll have to take the leap of faith. You’re going to make mistakes just like you did in every other kind of technology, but this is how you’ll move your organization forward. Your company will become a lot better. You’ll also be able to rebalance your workforce, put more people into areas that will help you grow and take other digital workers and put them into areas where maybe you had gaps, and you’re going to do things that you just could not do before.

The Lennar example is a great example because I love that they’re able to take their existing customers, service them better, sell them new products, and that they weren’t able to do it because they just didn’t have the humans to do it. Now they have digital workers. That’s a cool thought.

Not necessarily eliminating people, but just being able to extend what the people can do to do things that their business just could not do before.

Exactly. So you can think about having a connection, connect with your customer in a whole new way. And this is probably one of the greatest opportunities for CEOs in the history of business. It’s definitely the most transformational. Properly executed, businesses will just become simply better.