Salesforce CEO Marc Benioff Has Thoughts on AI Agents, Automation, And The Future of Your Job

There’s a narrative that AI agents will be personal bots that take action on our behalf, but Salesforce CEO Marc Benioff has a different idea.

Instead of our own assistants, Benioff thinks all companies will build their own AI agents that will act on our behalf when we check in with customer service or need regular attention. And he's angling Salesforce as the data provider that will make it all work.

I spoke with Benioff for this week’s Big Technology Podcast, where he shared his vision on why his system of record should make AI agents feasible. When agents have customer data and company policies on hand, they can take effective action on our behalf. And since Salesforce has all that information, Benioff wants to help companies build this experience.

Benioff and I go into depth on all this and plenty more, discussing the future of business software, along with Twitter, Elon, and Time Magazine (which he owns) at the end.

You can listen to the full podcast today on Apple Podcasts, Spotify, or your app of choice.

Here’s our full conversation, edited lightly for clarity and length.

Alex Kantrowitz: You’re helping customers build agents that can perform customer service and sales tasks typically done by humans. And through these technology decisions, I believe you’re making a statement on labor. You’re saying that people are pretty inefficient, and rather than replace them, technology and AI-based automation can help increase their productivity. Is that right?

Marc Benioff: You’re touching on the third rail of what's going on — which is that we’re building software that just isn't helping companies manage and share information, which is what we've been doing for the last three or four decades. But we're going to provide software that is very much an equivalent of labor.

I want to get your thoughts on productivity though. You're building this set of automation tools with Salesforce, and you must feel like there's an opportunity to make labor better, more productive, more efficient.

Think about it like this: I'm wearing a boot here on my right foot because I ripped my achilles a couple of months ago. Healthcare is the largest industry in the United States, and we're very much our own advocate for our health. I'm going in to have my achilles scanned, and they’ll say to me, “We're going to use this contrast, and you're going to want to flush it out of your body, so drink eight glass of water over the next day or two. And we're giving you these meds, and you're going to take those and you'll finish them.”

Then you really don't hear from anybody for a month or two. And so whatever the rehab instructions they're giving you, or the medication instructions they're giving you, or drinking water instructions that they're giving you, you're on your own. You're your own advocate, and no one is going to call you the next day and say, Did you drink the water? Did you take the drugs? Are you feeling okay? Do you need to come back for another scan? Do you need an appointment with your doctor? Do you need your labs redone? Because there aren't enough people to do that.

You go and talk to any major hospital or medical institution there just aren't enough people to do the jobs. These are jobs where patients need very specific, detailed coaching and information and ideas, and how are you going to provide that? And the answer is with technology.

Software has gotten to the point where it can deliver another level of the workforce and ‘agents’, like we're calling it, or an ‘agentic’ layer on top of hospital systems like UCSF, which I work with in San Francisco. So as I'm interacting with UCSF, I have to wait for somebody to call me back to actually get an appointment with a doctor. But if I can interact with that agentic layer, it'll be able to mitigate all the medical information, and all the USCF services, all the scheduling, and all the things that I really want to get done with UCSF.

The idea that even in our largest industry, we don't have enough for people to really achieve health care, to guide you and help you become healthier, that's a big thought.

When you apply it to other industries like financial services and how you're managing your wealth or education, there's going to be an opportunity to do that through this agentic layer. It’s through agents and this idea that you're going to have a limitless workforce, that you're going to have a level of abundance, not only as a consumer, but on the flip side, as a CEO. My company can now do things that it couldn't do before, where it was trapped by labor.

AI enables workers — like those caregivers at the hospital — to be able to do the things they couldn’t due to constraints on time. As opposed to job elimination, that seems like the future we're looking at.

That's exactly right — this is about humans and agents working together. In 2024, we haven't had a labor force expansion in the United States. It's stagnant. But we’ve had a productivity expansion in the third quarter, and economists are attributing it to the growth and expansion of artificial intelligence.

So we're able to expand productivity — which is linked to GDP — and we're able to expand our labor force without hiring more people. This has never been done before in the history of business. So we are definitely at a threshold moment. We are able to look at what's happening in a new way and come up with new ideas.

In the past several weeks, as we've been rolling this out to our customers, we're talking to so many customers, but some of the folks that I'm talking to are bank CEOs. Bank CEOs are interesting, because when you talk to them, in many cases, they've expanded their business with lots of new products, but they haven't expanded geographically. Maybe you see a Canadian bank expanding the United States, but rarely do these banks attempt to go global. And the reason why is because of the labor issues associated with going into new markets.

I think we can start to reconceptualize business itself, the products that we're in, the geographies that we're in, the size of our workforce, that we're working with humans and agents together. Just like you said, this is a big thought. This is not science fiction. This is not the future. This is now. This is the present.

Whenever I get optimistic, I always have to bring myself back to reality because there's a lot of ruthlessly bottom-line-driven companies that might use this technology to cut workforces in a big way. Are you worried about that?

For sure, people could rebalance their workforce. For example, we're the second largest software company in the world and the largest provider of CRM. Of our 75,000 people, one of the things we do is we interoperate with our customers — there’s 135,000 companies who ask us about 2 million questions a year, and they do that online and on the phone.

So all of us have this mix of phone calls and emails and we've tried our best to use deflection systems to reduce them, but it tends to be a very time-constraining thing, and can start to take up a lot of your employees’ energy and focus, and it's not the best use of their time.

Now that we've deployed Agentforce internally, some number of our employees now are going to get freed up to be able to do other things. I can rebalance my workforce and put those people who have those very good skills working with customers to do other things, so that we can grow, we can get more market share, we can innovate. We're obviously fighting the number one software company in the world— Microsoft — so we want to be able to do more and expand and go faster.

Let's get deeper into the ‘Agentforce’ rollout. I initially thought agents were going to be assistants operated by individuals, as opposed to Salesforce’s vision: something that interacts on a company's behalf to better handle my needs. So I'm curious, do you agree that your vision of agents is smaller than the broader, more common vision of personal agents? If so, why did you decide to go that route?

Salesforce helps companies connect with their customers in new ways, that's our fundamental mission. We do that, first of all, through automating every customer touch point, and that could be managing sales, customer service, and call centers or marketing and emails. It could be their commerce on their website. It could be their analytics. It could be Slack, Tableau. These are all connection points between companies and their customers. For years, we've been automating their sales forces and service forces and marketing forces, and their commerce forces. And now, we're about to automate their agent forces.

A great example is Disney. if you go to store.disney.com, that's Salesforce. And if you go to the call center at Disney+, that’s Salesforce. The folks that are managing the sales for the real estate and the cruise ships and all the different customer interactions are all Salesforce, including the Disney guides and the parks. They're all using tools from Salesforce to manage the customer information.

The second part of that story is not only automating all those customer touch points but building the Data Cloud. We're helping put all that data together in an amalgamated way. As we get into this next level of AI, it needs the data to be intelligent. It can't just run off of a large language model (LLM), which is an intelligence model. It has to be what we call ‘grounded’ or connected into the customer data itself, and not just the customer data, but the metadata.

Now the LLM can put it all together and be smarter. You have the agents that are able to reach out, and you also have the database, and you have all the automated touch points. All of this for Salesforce is one piece of code. It's our core platform.

Back to Disney; We're in the parks having a great time with our kids on our way to StarWars Land, and the Rise of the Resistance ride that we love. It's super high tech, but the ride breaks, and all of a sudden, the agent contacts our guide and says:

“Hey, we know you're on the way to Rise of the Resistance with Marc, but it’s broken. And a couple things you need to know: I just looked at flow control across the entire park, and Marc’s ride history, and put these two things together. The ride that Marc needs to go on next is Toontown. We just opened it up, it's the same technology as Rise of the Resistance and he’s going to love it. Take a right turn. Walk five minutes. You're going to be there. There's no line. You're going to go right inside.”

The idea that an AI agent could do something that a guide couldn’t do — to look across flow control and my ride history, and instantaneously make a recommendation — that's everything we're talking about. That’s agents and humans working together to drive customer success.

Share

Do you anticipate that people are going to have their own bots, and they're going to interact with company bots, and it's just going to be a circle of AI dialogue forever.

You can already start to see that on your phones. I have about a dozen of these AIs on my phone from all these companies, they're kind of commodities at this point. You're just talking to basically the same data set through a very similar algorithm, and it's a commodity product.

At some point you're gonna be able to set up preferences and do things with those tools which you can't do today, like asking them to run projects and coming back the next day, and it says, “Here, I researched this. I looked at it, I laid this out. Here's the idea for your Christmas vacation.”

That's one idea. But now let's say I'm one of those vendors and you're going to interoperate with us and our company, and you might do that through an agent right now or you might just be talking directly to us, because that's the most trusted path.

We're in a new world where the technology that you're using on your phone is going to be smarter, and the technology at these companies is going to be smarter, and eventually they're going to interconnect and can and collaborate and share and help you achieve your goals and achieve that level of abundance that we're talking about. You're going to be able to have it all done with the technology and the agent.

It'll be all fun and games until my bot and some company’s bot start to fall in love and don't do what we need.

One of the writers involved in Minority Report (2002) and War Games (1983) is actually our futurist at Salesforce. We've seen that idea play out a lot in science fiction, we can see where that could potentially go. But we could also see where we are right now. This is just an incredible moment that we're living a new part of the future.

Okay. Disneyland question. When you're on the rides, hand on the rails or hands in the air?

For me, I'm six, five and 300 pounds, so I'm just holding on and hopefully not get thrown out or anything that.

Your agents are $2 per conversation and the standard AI chatbots and copilots are $20 - $30 per month. Explain how the math is going to work for companies that have lots of AI conversations with customers, it seems like it's going to get pretty expensive pretty quickly. How is this going to work?

Our customers have traditionally paid a per-user-fee of $1,000 a year to use our Sales Cloud or our Service Cloud. When we moved into new products like our Commerce Cloud, or our ‘sandboxes’— where you can build your products, or even our Data Cloud, we started to do consumption pricing where you pay how you're interacting.

For our customers now using our agent platform, they're paying between 50 cents per conversation to $2 per conversation, depending on the volume of conversations that they're doing. This is attractive to them because the comparison might be a customer interaction today that's costing them $7 or $10 or $20. I talked to a customer where the customer interaction was costing them $700. So this is a very different level of interactivity, and our job is to provide value as well as the trust, safety and security, privacy associated with our technology, and that's the next generation of the pricing model as well.

So it's definitely a new technology model. It's a new business model, and we couple it with our new philanthropic model. 25 years ago, we put 1% of our equity, profit, and time into a 501(c)(3) foundation. That was easy because we had no equity, profit, or time, or people, or anything.

But today, we've done 10 million hours of volunteerism. We run 100,000 nonprofits for free on our service, we've given away about a billion dollars. And we also run a lot of these nonprofits and NGOs, and we meet these incredible folks, and one of them is called College Possible.

It's amazing, they help get the college advisor going to high schools to work with your kid, ask, “Hey, what are you interested in? And we're building a profile of your kid. And here's the profile of the college, and the counselor is there, and boom, all of a sudden you get a college recommendation.” How many college counselors do you think there are in California per 100 kids?

Maybe one?

There's one per 500, so there's not too many, and it's like what we were talking about in with labor. There just aren't enough college counselors, and so you're wondering if your kid is going to get that counseling in high school. All of a sudden, College Possible is using Agentforce, and they're building an agent and then an agentic layer on College Possible. They already had the data and the metadata, and they'd already automated all the customer touch points. So they were able to do this in a day, and they're able to take their human workers and extend them.

So you’re basically saying that it’s worth it to spend that $2 per conversation because you’re getting so much more on productivity.

I think you’re getting something you could not get through any other vehicle

I want to ask you about the future of business software. There’s an idea that, instead of using a program like Excel, you could just put all your numbers in a chatbot, and you talk to your bot instead of Excel. And so, some people think you’re being so upfront about AI agents because, if this vision wins, the future of business software may end up looking very different from Salesforce. Is that right?

Let's talk about where we are right now, and what could happen. We manage our data on disk drives, there's an operating system, and then there's a database. And because all the data has to be logically stored in the database and secure, there also has to be a sharing model. So we know who can see what data and what data cannot be seen.

We have another vendor who's delivered some of this technology — Microsoft. I just saw a story today where the wrong data got in the wrong hands, and it was because there was no sharing model, and the data was not locked up correctly. That's something that you don't want to have happen, especially in a regulated industry, like healthcare or financial services.

For example, my banker can't see your balance, and your banker can't see my balance. And that's how financial services is set up, and that's a security and sharing model. Then we have the apps, and the agentic layer on top of all that. You put a big bow around it, and that's how it should work now.

If you're in the US right now, we're running a test where if you go on your phone and go to salesforce.com, you'll see at the bottom right of the screen a talking bubble, and if you touch it, we're running a test with Agentforce. It’ll come right up and front end our website.

Instead of looking around our website and our products, you can ask, “What do you think about Agentforce? How does it work in financial services? How does it compare to that Microsoft Copilot?”

You're not authenticated, so it doesn't know who you are. But once you log in as a customer, we're now interoperating with you in a really smart and creative way, and you're talking to these agents. So these are real examples that you can get your hands on today to understand what the agent's future is going to be like.

Is it 100% perfect? No. Is it everything that is in Minority Report? Is it Her? No. But it’s pretty good, I think that the idea that customers can have a better relationship with your company, because you're going to have an agentic layer in your company, is a real idea.

You're not really afraid of, let's say, the Klarnas that are saying, instead of using business software, we're just going to-

I read that. They're a big customer of ours. They use Slack, they use…

Really, they said they were cancelling?

They said they were canceling but I didn't get their cancellation notice. All I saw was a lot of provocative statements. At the same time, they were also writing on LinkedIn about how they're loving Slack and rebuilding it.

We have many tools at Salesforce. But look, I have 75,000 employees. Let's take that as an example. Those employees, I have an employment record on, I hire them, I pay them a salary. They have stock options, they've got benefits. They have 401k plans. I have different information about them based on how they're doing. Whether they're doing well or not — their performance record.

Where do you want me to store all that information? That's all I want to know. If you have some magical new way to store it. Because I'm using disk drives right now. So if you have a new way to store data besides disk drives, I'd like to know about it, because it's probably an incredible investment opportunity. But right now, I'm using disk drives. I have an app on there called Workday. I have Salesforce. I have Slack. If there's something else other than that, I need to know.

There's a lot of people making provocative statements — they understand the future and they saw a movie, therefore this is what it's going to look like. Great, I saw the same movie, and it was a great movie, but just show me exactly what you're doing. Because I didn't get the alien technology from the UFO federation. All I got was the disk drives that I have today. So tell me what you're using.

Well, I messaged Sebastian Siemiatkowski from Klarna earlier who’s been on the show. No response yet, so we'll have to save that for our next conversation.

A great executive. He's going public and it’s so amazing.

We had a fascinating conversation here about how he's automating with AI.

He's a smart person. He's done a lot of great things, and maybe they have a new way to store financial information because they’re going public. They're going to have to be able to report information on their employees including all of their employment details. Maybe they have a new way to manage employment information reports and they're also gonna need to know all their customer information. So if they have a better way to do all of that, we want to know exactly what the technology specifically is.

But if you make a provocative statement and say, “Listen, the future of software is not what you're using now. It's this great new thing,” you have to at least tell us what the great new thing is. Otherwise, how are we going to figure it out? We're not psychics.

You’re a loud and prominent detractor of Xopilot by Microsoft. Your argument boils down to Copilot is basically just ChatGPT. I'm going to counter by saying they put it in everything—Outlook, Excel, PowerPoint. There are some cases that are just better defined and work better than necessarily the Copilot in Outlook, like talking to your data in Excel or being able to prompt a PowerPoint presentation. Are you saying that all of it is useless, or just that one part?

You have to remember Microsoft is our competitor, and they're the number one software company in the world. We're number two, so my job is to compare and contrast our product to theirs.

In a recent Business Insider report, they were testing Copilot and found that information was going into Copilot and while queries were being made, consumers were seeing information that they were not supposed to be seeing. In one case, somebody saw a CEO's emails. Not only was this data getting spilled and shared, the security model was getting misplaced. Customers were reporting they just weren't getting value from copilot.

The reason for this was because Microsoft just repackaged OpenAI's product and didn't integrate it very well, and didn't think from the ground up what it was going to be in its best situation to offer value to the customers. There's no finish line when it comes to security or reliability or availability.

I've yet to meet a reporter or even a customer who has defined a huge, valuable proposition for Copilot, where it's offered value to their company. Maybe Microsoft will pivot, like they usually do, and follow us into this new agentic world, because I think that's really where all the action is going to be.

I have a story about Claude from Anthropic.

I’m an investor in the company, they’re great

I have a running conversation with Claude. We talk about what I'm eating, the calories, whether it adheres to my whole foods guidance, and then it gives me a grade each day and we weigh in and talk about it the next day. Do you want to call it a Copilot, a chat instance? It's pretty useful.

It’s a great example. I’ve used these things a little bit as a therapist, where I’ll have a psychological issue and I’ll ask it a crazy question, and have gotten great answers. There's so many things you can do with large language models, but let's get clear what large language models are and what they aren’t.

They start with these next generation algorithms that are quite good, built based on deep learning and these artificial intelligence techniques. A lot of them came out of Stanford in the early 2010s and had a lot of breakthroughs. Prompt engineering was invented at Salesforce, which we're very proud of. Then what happened was we started to expand deep learning into bigger and bigger data sets, bigger and bigger compute, and all of a sudden we had generative AI. And really, that's the story.

This idea that we now have generative AI made possible by so many interesting companies, like Anthropic, but there are others and they're all very similar right now. You can go to the app store and download five or six of them and put them on your phone; They can draw pictures, answer questions, and some of them can talk to you. Google's Gemini is amazing. You can talk to it with almost zero latency. It's really cool. That’s the current phenomenon where we’re starting to have cool generative AI experiences like this.

The next step is, how do we actually connect it into meaningful data sets? Because wouldn't it be cool if it was connected to all your medical information in real time, and not only would you be connected into your medical information and labs and scans and diet and family history, but now your doctor's conversations too, and to end up in your doctor's office where it's right there in front of you in a big screen. That would be incredible, and we're close to that happening.

As we come to an end, I have two questions on different topics: a media question and an Elon question. Media question first: Are you going to sell Time Magazine?

Well, I love Time. It's been a lot of fun and I really enjoy it. I work on it every day. It’s connected to the Elon question; We just published our fifth cover on Elon. He just tweeted and texted me, “Oh, I don't want any more covers,” but he does love the covers, he wants ‘Person of the Year’. He's a great guy. He's doing incredible things. He's brilliant. He's running like 2,000 companies and the U.S. government all at the same time. We've never seen anybody like this in our careers, right? It's like this person can't be human. How is this even possible? But it is. And whatever he's doing, you know, whatever he's drinking, whatever he's eating, whatever drug, I want some of that because I've never seen anything like this.

On the Time question?

Oh, I love Time, and it’s amazing. We bought it five years ago, we're doing great with it, and we love it.

There's been a bit of a question about the billionaire owned media company moment that we’re experiencing. You bought Time, Bezos bought the Washington Post, Patrick Soon-Shiong bought the LA Times. I would say you're probably doing the best of the three. Do you think this was a mistaken era?

When it comes to our journalists, they tend to represent their beliefs and ideas and visions and aspirations for their own lives. A lot of it comes out of certain geographies where they're based. Because those GEOS have certain kinds of characteristics, even political or cultural, they permeate into the journalism itself, so it can be a little bit difficult for a journalist.

I'll give you an example in New York and Bangalore, India. It's two different worlds. It's two different economies. It's two different languages. It's two different philosophies of life. Everything is completely different. Then let’s add in Jakarta, Indonesia. These are different mindsets. I think when journalists are writing, they're writing from their mindset.

I'm actually finishing up an article that I'm going to publish in Time about all the things that we're talking about right now with labor. I'm writing from my perspective as a CEO of a software company. So journalists are writing from their perspectives, and you have to just understand that it's not always going to match exactly your perspective, because it may not be where you live, how you grew up, your religion, your philosophy of life, or your mindset, and that tends to be the conflict, and that's what I see playing out.

We’re fortunate in Time that our journalists, and especially our editor in chief and CEO, understand this, and they’re trying to upscale how they’re thinking about things, and making sure that we don't go too far left or too far right, that we go down the middle.

By the way, that’s not the History of Time magazine at all. If you go back to the history of Time Magazine, which is 100 years, if you just look at the covers or read the magazine, you'd see that there was a political intention. There was even a time when the founder had a very specific political ideology that he was expressing through that. I don't think that's appropriate in the current world.

We're trying to create a neutral, agnostic, balanced view on both sides. Do we always do that? No, because these are human beings writing these stories, and they're going to write it from their perspective, and that's all that's going on.

I don't think anybody should be demonized or criticized for this perspective. I think that we should just have a realistic view. You have a lot of experience in the media industry, how do you see it? Do you think that that's a fair way to look at what's going on, or do you think there's some other reality?

I think it's going to be the essential question of journalism forever. How much of your own viewpoint do you insert into what you're doing? Ultimately, your job is to unearth facts.

Last question. During the Twitter sale, you texted Elon: “Happy to talk about, if it's interesting, Twitter as a conversational OS — The town square for your digital life.” What do you mean by that?

What Elon has done has been amazing with Twitter, and he's really transformed it and done something that I could have never envisioned. When we were talking about buying Twitter, that's not what we were going to do. We were going to build something that was more of an application development and deployment capability where you really had the same Twitter feeds, but you'd have hyper cards in those frames, not just photos.

If you remember hyper cards, which came out of the original work in the Macintosh, I worked at Apple in 1984 so I loved hypercard. That idea that you could have apps and an app store and applications and not only have a creator economy and people building incredible videos and dialogs and conversations, but also apps. That's what I always wanted, for apps to be part of the Twitterverse, if you would. That was a vision that I had which was a little different.

I talked to Elon about it, but he's not interested in that. He's very much on this, “I’m opening it up to everybody. I'm going to turn on everyone who's turned off before. I'm going to let the absolutely information free for all, and no guardrails. And let's just let everyone work it out. It's kind of like The Hunger Games. And I think that that's fine, everybody knows what it is, and he's doing his thing, and there's Section 230 preventing any liability for him, so he can just let that roll.

That's not how I looked at it. I had a whole different vision that will never be realized, because I did not end up going in that direction, mostly because my investors did not want me to go into this crazy world that he is doing a very good job in, actually. And so I'm not in that world.

Share