Guy Raz & Parker Harris: Building Salesforce

Beautiful. It is a beautiful day. Yeah, it is a beautiful day.

Well, I will start by saying one of the luckiest moments in my life was when I met Marc Benioff. Myself and a couple of co-founders had been at a small startup for many years. We were doing Salesforce automation. And it was Salesforce automation on a laptop. You know, it was early, early days. It was not cloud computing. And we left that company and we did some early cloud computing work as consultants. And we met Marc Benioff through one of those jobs that we were doing. And Marc had taken a sabbatical from Oracle Corporation where he was an early executive and very, very successful there. And he had this idea. He's watching like Amazon.com as a bookseller and at Oracle they were doing the network computer. This was if you remember client-server went to thin client and Netscape Navigator was out. And so there's two ideas in his head when he was on sabbatical. He had this idea, you know, what if selling software was as easy as buying a book on Amazon.com? You didn't have to buy the computers. You didn't have to install the software. We had the background in Salesforce automation. We had the background and early like what does it mean to build something for the cloud and it was just this magical meeting where the CEO of this company we're working at introduced us. Marc was an investor. We were consultants. It was down in Burlingame. And we had lunch and Marc said, 'I want to start this company' and we were looking for our next thing and we just hit the ground running and we built the prototype very quickly. This was in the dot-com, remember the dot-com? And here in the city every company was growing and they were, you know, and these dot-coms were our first customers and they were totally willing to take a chance on this startup. They're like, 'Great, you know, love to use you.' And working in the cloud, awesome. And you know, and the bigger customers were not ready. They were like, 'Oh, you want me to put my customer list on the internet? I don't think I'm going to do that.' And so we started with these little customers, boom, and just grew from there.

Well, I think the first was, you know, every corporation was saying, 'Look, I can run this better than you.' Like the IT departments were about security. They were about performance. You know, when we launched the company one of our metrics was fast at 56k. That was a modem. People were still like using software on the internet with dial-up modems and we had to think about well what is performance, you know, at that kind of low baud rate. And so performance, security, you know, and trust. You know, like look, we understand you're giving us a lot of control and we're going to run it for you, but what we're going to give you in return is we're going to give you that trust. That's why when we talk about our corporate values, trust has always been our top value because when we started, people were still not sure they could trust their credit cards on the internet. Right. They were like, 'I don't know if I want to put my credit.' Now we're like, credit card here, credit card there. Buy this service. It's like everywhere, much less your customer. And you know, what if you lost your customer list? What if there was a hack? What if it's not available and I need to sell? But actually one of those startups in the dot-com, it was a great moment for us. Well, not for them because they had their company broken into and all their computers were stolen. Every and these were desktops. These were not like laptops and big screens. Everything was stolen. And they were very upset. But then they were like, 'Well, it's actually not terrible because Salesforce still is running my service for me. They still have my contacts and they still have my deals and I can just get a different computer and use that.' And I know that's kind of obvious today, but back in the day that was revolutionary.

How much do you want me to be a marketer versus reality right now? Just kidding. You know, one of the things we did is we really started with the word co-pilot and there's a company that we compete with that's very focused on co-pilots and we wanted to really differentiate ourselves. So, let's just start by like we definitely need to lead with, you know, how we position ourselves and so we really went forward and talked about agents and agentic AI. And if you really take the literal definition of that, agentic AI means it's an agent that can do things on your behalf. It's agentic. You can say, 'Please go off and research this company for me or go off and create a briefing for my next meeting or prep me for this talk today.' And it goes off and does that work for you. I think that is the future. And I think there's a lot of examples we have. Salesforce has something called a sales development rep. It's a human being we hire and they dial for dollars into our accounts. And now we have a digital one and it'll automatically, autonomously go and email customers that leads we weren't sure we're going to close and if the customer responds that agent will have an agentic response. But I still think we're early days. I still think that yes, there is autonomy. There's great examples of autonomy, but we need to be successful with AI right now. And some of it is just like, let's augment ourselves. Let's just become more productive and, you know, let me have this expert right there with me. I'm in a call center. It's helping me understand who this customer is and I can work with it. Or, you know, I'm a developer. You're seeing, I mean development is an amazing place of productivity and I have AI there to say like don't check that in because there's a bug there. And how amazing is that?

Well, there's actually a company that's just above me, a company called Adecco and they created a separate company called Our Potential and I just had a brief talk with their CEO and they provide human capital. They provide, you know, you need employees, they will source those employees for you or those consultants, contractors, and they'll do their work for you. And they had this big idea, well, what if you wanted to hire an agent, an AI? And so they're building AI employees basically that you can go hire and that will do different jobs for you. And so that's kind of the far out like, wow, that's a big, big idea and they have like a potential agent that's going to go and look for opportunities in your enterprise. But then you just look at like Williams Sonoma and you just like I really believe in simplicity and let's just start simple and they have Chloe who is, you know, and you can go into the Agentforce and check out a lot of these examples. But you know, my sous chef and help me think about like a recipe and you saw it in the keynote. And so we're starting out simple where, you know, Chloe maybe doesn't know me yet, but it does know recipes. It does know the product line of Williams Sonoma. It will help me know like, okay, I have this pot. What's the right recipe? But then we can go from that. We can be like, oh, well, now it knows me. Not only does it have an expert in cooking and an expert in the product line of Williams Sonoma, but it also has deep knowledge of me as a customer and then it can say, well, I know everything you've bought and, you know, here's some new things that we have or new recipes based upon all the things that you have in your kitchen. And I think that's where every customer is going is, you know, they're starting simple kind of expert knowledge agents through to highly personalized interactions that are leveraging all the data across the enterprise about the customer or the consumer and giving those incredible interactions.

Well, I think, you know, customer service is the first place where I think every company has gone. Whether, you know, it's examples on your website, if you go to help.salesforce.com, we're trying to be the first to use all of this AI. You can see we went from using a chatbot to using an agent built on Agentforce and we're showing you the statistics. And so, you know, that's a good starting point. We're now using them in the call center as well. And so I think you need to look at customer service as the first place of disruption because we need less humans to do some of those, you know, basic interactions and it's creating a lot more productivity in the enterprise. Now, a corporation that's thinking about, you know, expenses could just say, 'Oh, that's great. I'm just going to cut my workforce entirely.' But you can't cut yourself to growth. So you need to think about okay well that's an opportunity so we can take that capacity and we can retrain them and they can do higher level, you know, complicated support tasks, they could be in the field with your customers. We're taking a lot of people and retraining them and they're in with our customers as forward-deployed engineers. And so I think you need to think about where are the first opportunities, what is it going to do to your workforce, how do you bring your workforce forward and transform them because we all don't have enough workers, human and digital, and we need to move our humans forward and their potential and retrain them and then we need to add that infinite, almost infinite capacity of AI. And the final thing I'll say is keep it simple and iterate and test and know that you're successful. And I think you can get carried away like we have amazing marketing here at Dreamforce and, you know, kudos to our entire team for the show that they put on. We want to show you a vision of where we can go and where you can go with us. But where are you going to start? And where are you going to know that that starting point was successful? And what's the next thing and the next thing and the next thing? And so you need to hold that grand vision in your head and inspire your employees and your customers. This is where I'm going to transform my company. But then how do you break it down into things that break that risk down that let you get there? And it's not a straight line to get there, but how are you going to get there iteratively and successfully?

Well, I think you said two things. When will they be strategic? When will they make decisions? And then when will they be autonomous? And I think it's two different things because I think we can start, you know, Marc as an example, and I think we need to change how we work by the way, you know, how many of you are still, you know, just searching for answers versus talking to AI to get those answers and it's a very different mindset. I've watched Marc really lean in hard on, you know, how can he transform and he, I mean to a fault, you know, we're uploading or he's uploading to AI our business plans, our strategy, you know, the earnings script, you know, of the next, you know, these highly confidential things, and he's talking to his friend, his friend is AI and he's asking, you know, he's building the business plan and he's got a small team of like two people and an AI working with him and that AI is making decisions. It is, you know, but he's translating that. He's taking that and thinking about do I agree with it or do I not just as he would any top employee who is advising him. It's just another input. And I think we all need to start there. Change how we're working. Trust the AI with, you know, a lot of this information, and use platforms like Agentforce that have these trust layers that secure the information and go outside your comfort zone. You know, are you asking AI the same thing you're asking the top expert in your company? And ask them both and have them work together and start to transform how you're working in the enterprise. And I do think AI will help us make the decision. It's not going to make the decision for us, but it's going to say, 'I think you should do this. I think this is the right answer. I think this is the right strategy.' We still need CEOs. We still need to make decisions. We do not want to like cut the cord and go like, 'Yeah, run my company for me. I'm going to go play golf.' You know, have the AI CEO, but imagine if you have an expert next to you who's helping you as one of your additional advisers and then maybe in some cases you do want to say, 'Okay, yeah, go do this autonomously.' But I think it's two levels of complexity. Yes, I trust AI to have an email conversation with a potential customer and go back and forth and then when it looks like it's a deal that would close that SDR agent will hand it off to one of our best salespeople and they'll close it. But for our top customers here, I want to talk to them. I don't want to send them over to the AI and go talk to the AI and the AI will summarize and tell me. You know, I believe in the human potential and I think we're all rising because of AI. And so I think the AI can help us make decisions. It can also be autonomous, but it's two different things at this moment in time.

Well, I think one of the first challenges is, you know, you go to ChatGPT, you go to Anthropic, you go to Perplexity, they've trained these models on all of this public data. Maybe not all public, but they've trained it on a lot of data and you can ask it things and they will answer. But for our customers, it doesn't have your data. It doesn't know your customers. It doesn't really know deeply your products. And so the first problem is have you unlocked all the potential of all the data assets you have in your corporation? And many people when I say that will think about all your structured data, all your databases like my contact lists and the activities of my interactions and my invoices, the payments, that's all structured data in CRM and ERP. But what I've discovered is the power of unstructured data. This conversation, the text on your iPad, the voice conversations, the Slack messages, the emails, all of the documents that are being written, there's a wealth of information there that we couldn't tap before and now with LLMs and generative AI technology we can unlock that and it's almost like there's more value in that data because it has all of these human interaction. It has much deeper knowledge and we can now tap that. And so the first problem is have you unlocked all your data and have you unlocked it in a secure way. It doesn't mean take it all and throw it in an open database because, you know, I have access to data at Salesforce that I don't think every employee should have access to. I'm on the board of directors. I report to Marc Benioff. I need access to a lot more information. So I need a whole security model around it. And then beyond data, does the AI have access to the APIs, to the skills, to the tools? Can it go and create a briefing document for me? So does it have an API for that? Can it check on the shipping for a customer? Can it see if the customer paid their bills? These are all different systems, so it's reaching in through APIs to do work on your behalf. And then are you testing it? Do you have the data and analytics every day that are knowing that the AI is working?

Well, if it's a repetitive task by the nature of it being repetitive, it can be automated with AI. You know, I'd like to say 100%, but a large majority of repetitive tasks will just be automated. So like for example, in a company onboarding somebody, we're already doing it. We have agents that will onboard our employees. We have agents that will train our salespeople. Our salespeople will do video pitches to an agent and the agent will take the transcription of that video based upon a rubric, judge that salesperson of how well they did in pitching our products and give them constructive feedback. And, you know, that kind of thing is amazing because I'd be kind of embarrassed like I don't want you judging me about my pitch as a salesperson, especially if I'm a new employee. I want to impress you. But if I'm working with AI, it's like my private little tutor and it's working with me. So onboarding is a great use case for productivity.

So I think jobs are going to change dramatically. And I think the model providers have to pitch that vision that their models are going to keep getting smarter which will do the work of some white-collar jobs, some percentage of them, and I think that's a wonderful vision and I think that will happen more and more that, you know, just as I said, some of the functions of a salesperson or some of the functions of a customer service person can be done with AI. But that just means like I think there's going to be new jobs that are created. You know, when the industrial revolution happened or when the personal computer came out, we didn't have less jobs, just the jobs changed. We were more productive, we did new things. You know, you're going to have a paralegal or you're going to join, you know, the young people who join in financial services and investment banks and they're research analysts, AI is going to be able to do a lot of that and so does that mean we don't need anyone entering the workplace in a law firm or anyone entering the workplace in a financial services firm? No, but maybe when they come in they better be able to understand how to use this new tool to do the research so that, you know, they're not spending the weekend doing it 24/7. It's just happening with them. They're doing deep research or, you know, they're having the AI help them understand the legal brief and help write the legal briefs. So, they're going to have higher orders of work. And so, that means a radical transformation in training and enablement in the corporations. Jobs are going to change, new jobs are going to appear. And we need to help. It's not just a technology shift. It's a huge human shift of human capital.

Yeah, that's the problem is, you know, this AI is a whole new world. It's not predictable. You know, so we came out with something called Agentforce here at the conference and it's nothing more than workflow or a state machine. It's not AI. It's in the reasoning loop of AI because when we had our forward-deployed engineers working with our customers, they would tell the AI, do this, do this, and then do this. And everyone's like, yeah, and the AI is brilliant and it'll just follow those steps. But it didn't always follow those steps, you know, and you do the demonstration, you see like, oh, that's great. Deploy it. And then but then when it doesn't follow those steps that can be catastrophic for a company that, you know, is it a workflow to return, you know, at Williams Sonoma that pot, you want to go through a series of steps. And so we had to mix it together and most companies are doing that where we're definitely leveraging AI but we're also trying to add some predictability into how it's working and this is a completely new technology for how are we building it, how are we deploying it and then how are we testing it. So the other thing is, you know, we're not a model provider. We use the best models in the world from the companies like OpenAI, from Anthropic. But what we do do is we're building a platform around them that can give us the data or give our customers the data of like okay this model and this AI and these prompts and this topic on Agentforce is working great but you know what when users were asking this question it wasn't working or we have a whole testing harness, the AI will create, we have AI that will create a bunch of questions, prompts to send to the agent and then it knows when the answers are right or wrong and so you can, how do you test it? How do you do A/B testing? So Joe Inzerello is going to speak later. He's our new chief digital officer and he thinks about rolling something out but I'm going to give you this new version and maybe only 5% of the users are going to get that and the old version is this version. And then we start to see, okay, that one's working much better because we added this data and we tweaked this prompt and so we have to think differently about how to deploy and how to be successful and that's why we have these forward-deployed engineers. Since last Dreamforce, we had to transform our workforce. And we want our people right now in there with you and iterating with you and showing you how can you iterate to success. And we'll take that and we'll perfect it and we'll build more into our platform. We'll train more people because we need partners. We need more trailblazers out there who are in our customers. But we're going to do it first so that we get it right. But we need a whole workforce out there that's in there helping us. And we want to train our customers too. Ultimately, you need new roles in your corporations that understand this new technology, understand where it's amazing, where it's not, how do I make it work, how do I make it predictable, how do I make it trustworthy, on brand, etc., etc.

Well, I think you need to rephrase and explain to me what you mean by value judgment. But, you know, I've talked a lot with Dario Amodei at Anthropic. And when they broke off and created Anthropic, they created what they call constitutional AI. And it really started with values. Here's what, you know, the values of what I want you as an AI to how I want you to operate. And I do think, you know, Salesforce operates by a set of values. I do think we need to figure out how do we have values imbued in the AI because you're representing my company, you're representing my brand and how do you bake that in? And I think it's great to have companies like Anthropic leading that charge and then behind that, how do we in our platform give a set of instructions that will do that and we have guardrails, you know, so in Agentforce you can create so here's the values I expect you to imbue here, but we can also do the opposite and say and by the way here's guardrails I don't want you to do. And, you know, it's kind of like with an employee. If we hire a bunch of new people out of school, which we do all the time, and put them in our workforce, we have to train them. We have to explain to them what are the values of Salesforce. We have to put guardrails to say like there are ways that you could violate our values or our security. And when they do, we have to have repercussions from that. I think it's no different now with AI. And I don't want to over-personify AI. We can go too far on this and think about them as complete humans but I think we can learn a lot by the idea of bringing new humans into the workforce. There's certain things that we do. How do we think about bringing AI into the workplace? And because it can be brilliant but it can also do the wrong thing just like a human. How do we think about bringing them in, training them, setting the guardrails and the values and then checking that and holding them accountable. And, you know, you might bring an AI agent into the workforce and put them on a low value but highly repetitive set of tasks. And then you may see like, huh, they're doing that really well and I'm going to actually give them a higher order and, you know, I'm going to train them for that. Give them the data, give them the skills. That's the training. And I'm going to unleash them onto something else. But just with humans, we're not going to just let that happen willy-nilly and just, you know, we have to manage it with care.

Well, I think, you know, if you just think of it as a tool, you're not interacting with it. You know, the magic of AI is that collaboration. And I think we've all done it is, you know, the first answer is not always the perfect answer. It's not always the best. But as we have a conversation and that back and forth, you're learning from each other. You're collaborating and it gets better and better and better. And I think it's both the quality of the AI, but also the quality of, you know, how are we as humans adjusting to leverage that not just as a tool. If I could use it as a tool, I could say, 'Tell me, you know, what are some restaurants around this neighborhood that you know are really great?' And it goes off and here's a list of restaurants. And that's a tool. It's a better search engine. That's highly valuable. But then I could have a conversation, say, 'Well, I'm actually interested in this type of food, and I have someone who's gluten-free and blah blah blah, and, you know, oh, what's available?' You know, oh, can you bring up a review and interact with it more? That's more collaboration. It'd be like talking to you and saying, you know, let's have a conversation about it versus, you know, I don't use you as a tool and just pull something out of your brain. But I do think we have to change human behavior a little bit to get used to this.

I'm really trying to figure that out. I think it's a fascinating topic. You know, I have to go to a board meeting this weekend at my college and it's a liberal arts college and I really think about the human qualities of critical thinking and creativity. You know, some people believe that AI is highly creative. Maybe it is or maybe, you know, it's like Picasso, great artists steal and AI is just stealing the past and remixing it. And so I think we do need to explore what is uniquely human and uniquely valuable. You know, we have empathy. You know, I think AI unfortunately does not have empathy. It's not really truly an empath and thinking. Critical thinking, like maybe AI is, you know, how many times have you talked to AI and it gives you the wrong answer but you know it because you're thinking critically, you're like that's a really interesting idea but I think you're off here and then it says oh you're totally right, so sorry, and then it gives you a better answer. But, you know, are you thinking critically? Like they teach you in school. You can read history, you know, from books and you can learn, you know, the great people have written about what's happened in the past, but is it all true? You know, who wrote that book? What was the reason that they're putting forth that idea? And there is truth in there, but it's also biased and everything you're receiving has some source of bias. And are you thinking about that? And so just as we're, you know, we won't be able to leverage AI better if we don't have that critical thinking. And so, but I really think about how do we teach young people that are going to come into the workplace, what do we need to teach them, and it's not just, oh, we want to teach you how to do a prompt and use ChatGPT or whatever, learn Agentforce. That's all tactical. What do they really need to learn to come out into the world to be productive and to take these new jobs that exist or to create them even?

It would be ironic if AI makes us more human. It would be very ironic. You know, back to my alma mater, they're known for languages, and one of the things we were talking about is, well, what does it mean to teach foreign languages, to learn French, to learn Russian, Spanish, Chinese, and, you know, with AI, the new Apple AirPods, you know, oh, it'll with your phone, it's going to translate as we're talking. So, who needs to learn a foreign language? I'm just going to speak mine, you'll speak yours. But when you think about language, there's nuance and what are they really saying, and it's not just translating the words into English, or whatever language you speak, but what are they saying? Like why did they choose those words and understand that nuance is something you need to learn and so again there's that higher level thinking. Basic translation of just tell me the words, fine, but what is the meaning behind it and help me understand that and that's something that, you know, I do think because you need to understand the human too, you know, and what's the motivation, what's the thinking, with the body language, everything behind it. It's a higher level of thinking that the AI could help us with the basics. That's great. But maybe, as it's doing that, we're thinking at a higher level.

Absolutely. And, well, just think about computer science. I loved computers as a kid. So, you know, I taught myself how to program and then yeah, I did do English literature. But think about computer science as a degree and people are saying do you need to learn computer science? Oh, the computer is going to write code for me and just take that as a basic thing. We do need people who understand how computers work. How do computer languages work? And, you know, and yet we're seeing huge productivity lift in the enterprise, where people that, you know, there's a lot of IP out there that's public that was trained these models trained on that you can now vibe code and that's amazing. But if you don't really understand and think critically about that, what we found inside of Salesforce and actually in the Slack business unit is these junior programmers are using AI to vibe code and to go a little further than, you know, and that's a little bit probably outside their skill set because they haven't learned enough. But we still have humans review code. That's very important. Like they want to before it goes, you know, put into the main codebase, we want someone else to look at it. These senior engineers are looking at it and then they're going, 'You're wasting my time. Like what is this?' And so, you know, it kind of worked on the surface, but that junior programmer didn't have enough knowledge of like well the right way to do things. The senior one did through experience. And so I think and it actually was reduced productivity in this case. So, you know, how are we going to like I still think we need to teach computer science, but, you know, it's a higher level of thinking. We do need like take English literature, learn how to speak, learn how to debate. You know, I think think about like YouTube, social media, all these technologies are wonderful, but much of them are reducing our human interaction. You know, people are talking to AI and not talking to each other and that's a good thing and a bad thing like talking to AI as a fellow collaborator but we also need to amplify the human.

I think it's going to happen faster than a decade. But, you know, what if you had yourself but smarter than yourself but it's kind of like your digital twin and it's there as your advisor but it knows you. It knows your voice, your brand, your knowledge. And I think we're all going to have these AI twins that are out there helping us. So, I'm an optimist and I believe it's going to raise human potential, and I don't think, I do think we need to have impact in the world and we need to go out and do things that make the world a better place to create more value in the world. And so, I don't think we're going to have AI, you know, everyone's going to get the universal basic income. Maybe we'll get to that. But, you know, if we all stop doing anything, and we're all just kind of relaxing and AI running the world and creating better energy and food's abundant. I think that would be a horrible world to be in because I don't think it's going to happen. I don't think we're going to be in this world where we're doing less and I think it's going to increase human potential, human productivity. But we have to have conversations like this. We have to think about well what are the risks? There are the downsides of it. But in 10 years my hope is that we're all better human beings. We are, you know, maybe AI will help us be a little less partisan, a little less broken apart where we are right now. And the world is a better place.

Thanks for having me.

Thanks everyone.