Experts of Experience - Digital Labor Is Now: Why 2025 Will Be a Turning Point

Episode Date: July 23, 2025

2025 is the last year of human-only teams. Are you ready to lead both people and AI? Usman Nasir, VP of Agentforce Acceleration at Salesforce, joins us to explore a future that's already here: digital labor working hand-in-hand with humans.   Usman explains why 2025 marks the end of the human-only workforce and dives deep into how AI agents are transforming the workplace — from automating customer support to driving internal productivity. He shares practical advice on starting your agentic journey, debunks common myths, and explains why trust, data quality, and modular agent design are the pillars of successful implementation. Whether you’re leading a Fortune 500 company or bootstrapping a startup, this episode will shift how you think about work, leadership, and the AI-enabled future. Key Moments:   00:00 Introducing Usman Nasir, VP of Agentforce Acceleration at Salesforce03:13 The Future of Human and Digital Labor06:32 Salesforce's Agentforce (Chatbot vs. AI Agents)19:41 Real-World Use Cases and Misconceptions34:16 Exploring AI in Operations Management37:12 Identifying AI-Ready Use Cases41:44 Change Management for AI Adoption45:35 Lessons from Early AI Adoption01:04:29 Future of AI Agents and Predictions –Are your teams facing growing demands? Join CX leaders transforming their AI strategy with Agentforce. Start achieving your ambitious goals. Visit salesforce.com/agentforce Mission.org is a media studio producing content alongside world-class clients. Learn more at mission.org

2025 is the last year that CEOs will be managing human-only workforce, because we're now going to have human and digital employees working side by side. Think of a chatbot as a vending machine. But in today's world, the AI agents would be like a personal chef. Customer service and support continues to be the most prominent and most common use case. The beauty of this whole creative relationship with AI is what I call compound learning. What I've seen it already do is allow humans to be more human. Your AI agents are only as good as the quality and the consistency of the data that you're providing to them. Almost 80% of an organization's data that sits in your

company today is unstructured. So these are things like emails, PDFs, and Slack conversations. And now AI agents can actually sift through all of this data, structured, unstructured, videos, PDFs and Slack conversations. And now AI agents can actually sift through all of this data, structured, unstructured, videos, transcript, et cetera. And then it can basically give you the answer. Is there like an ideal size for a company to make this investment into AI agents? If you're ready to hire employees to do a job, your company is already ready.

Start small and scale strategically. What technology are you most excited about? What are you like betting on? I'm betting my career on this topic, right? So that's what I am 100% focused on. So what advice would you have for people that are maybe looking at starting their agentic journey now? Welcome back to Experts of Experience. I'm your host Lacey Pease. And I'm Rose Shocker. I produce Experts of Experience. I'm your host, Lacey Peace. And I'm Rose Shocker. I produce Experts of Experience.

And we just got off the mic with Usman Nasir, AVP of Agent Force Acceleration at Salesforce. And what a fun episode this was. We did a deep dive into AI agents. We've talked to a lot of guests about how they're using AI and AI agents specifically across the different industries that we've been able to speak with. But for this episode, we dove a lot

into the actual technical logistics of AI agents and how to think about when to implement them, when not to implement them, and what it actually means to have this dynamic of human plus AI labor or human plus digital labor. This is a great episode for the skeptics. This was an episode that was packed full of myth busting. We talked a lot about misconceptions

concerning AI agents. So I challenge the skeptics out there. And if you know one, send this episode to them. Because this was like hearing about physics directly from Einstein. Like we got, this was such a thorough conversation. I would say more thorough. we got really into the weeds. He talked a lot about specific use cases

and troubleshooting in real time. And what his team does is, and correct me if I'm wrong, Lacey, but works directly with their customers to oversee the deployment of AI agents and make sure that it's actually happening at the most productive and progressive level. Something I think is interesting,

whenever we get to sit down with AI agent experts, I've noticed them allude to the magic really being in a team's ability to know when to use AI and when not to use AI. Because similar to what you were just saying, Lacey, we'll see all over LinkedIn, we'll see all over online in general, people saying, oh, I'm going to wipe out this entire team

and replace it with AI agents. And while maybe that's in our future, we're not quite there yet, which I feel like is what Usman was talking about a lot, was you need to know when to use it so you can allow your humans, your human employees to be more human and strengthen the connections they have with their human customers. Yeah, well, and he mentioned a quote

that Mark Benny has shared, and I think we've actually shared this in an episode or two before that Mark believes, and for those who don't know, Mark Benioff is the CEO of Salesforce. Mark believes that there is, like within a year, no manager will have a team that's just human labor, right? Like every workforce, every team will be digital plus human labor. And that doesn't mean only digital labor. It doesn't mean replacing the human labor. It means figuring out how can we train our teams that do

exist and get them really comfortable with the technologies that are out there. So that way when we are facing this future world where there is AI agents involved in everything that we're doing, we can really have this collaboration and not any, not this friction that we hear a lot about like on LinkedIn, like all this friction about AI agents and what the future looks like. I choose to believe and I'm optimistic in the future of this world where it can be human plus AI. I don't think that means that it will be perfectly pretty or beautiful. I think that there will be some things that roll out in difficult ways, but that's with any new technology. So I think this episode specifically is just a great masterclass in what is an AI agent,

when do I implement AI agents, how do I get started, who is this for, how do I get my team bought in, how do I lead a team that includes digital labor and real human labor. So yeah, it's just a full comprehensive perspective on AI agents. And of course, this is a customer experience podcast, so we don't just talk about the actual technical side of this. We also talk about what does this mean is a customer experience podcast. So we don't just talk about the actual technical side of this. We also talk about what does this mean from a customer perspective?

What does this mean when your customers are employees, right? So in some instances, the tools that you're making are for your employees. And yeah, we just go through this whole scheme of what is AI agents, what can we use them for, and what the future really looks like. Yeah, and if you're a Slack user, be sure to stay tuned in through the episode

because he talks about a really cool feature, AI agents being integrated into Slack, which I thought was amazing. And towards the end of the episode, he also talks about how you and your team can start training yourselves on how to build AI agents and learning more about this entire new world because it's here. And I know I did not cover every question we could possibly have about AI agents. So if there's something we did not cover today

that you're so curious about and you're like, I want a follow up. I want to ask more questions about this. Drop it in the comments. If you're listening on YouTube, you can drop it in the comments. If you're listening on Spotify, Spotify has comments now. If you're on Apple Podcasts, I'm sorry. You can't do that, but you can head over to my LinkedIn. You can

DM me or you can comment at any of my posts, whatever burning questions you have about AI agent implementation or the future of AI or just thoughts or disagreements about the perspectives that we're holding. I would absolutely love to hear all of that. So head over to the comments and let me know what questions you have. Yeah, be sure to hit the like button, subscribe, so you can be tuned in on a new episode every Wednesday. And with that, this is Uzman Nasir, a VP of Agent Force Acceleration at Salesforce.

Uzman, welcome to the show. Thank you, Lacey, super excited to be here. Yeah, I am so excited that you're here as well. It's not every day that we get to talk to someone from Salesforce. And as everyone from the show knows, we love Salesforce, of course. So I'm so excited that you're here and we get to chat about Agent Force. Before we dive into AI agents today, would you mind giving a quick background on yourself

and what you're doing at Salesforce? Sure, happy to do that. So I've been with the company for around 13 years in a variety of different roles, ranging from account management to sales engineering, and now for the last almost one decade, been part of the customer success team, again, in a variety of different leadership roles. For this year, I am one of the leaders in our agent force acceleration team,

which I'll talk about in more detail. But really, our team's responsibility is to help our customers become successful on Agent Force, deploy and test and create AI agents. And that's something which we are later focused on. Yeah, so you get to kind of see the hands-on execution of Agent Force. And you've been with that team since Agent Force's launch. Is that correct?

Yes. Awesome. That's amazing. So for those who are maybe new to our show and haven't heard some of our other discussions about Agent Force, could you give a brief overview of what Agent Force actually is? Sure.

And I think before we jump into Agent Force, maybe it's a good idea to talk a little bit about what are AI agents, because there's a lot of different definitions flying around. So it might be a good idea to level set on that. So if you think about lazy AI agents, AI agents are autonomous, proactive applications that can understand, they can reason, they can decide.

And probably most importantly, they can autonomously execute complex tasks with or without human intervention. And this really represents a new world of what we call digital labor, where AI agents effectively augment human workforce by handling a wide range of tasks at speeds and scales that were previously impossible for humans to do those alone.

And agent force is Salesforce's solution and our platform to help our customers create, test, and deploy AI agents with a lot of configuration options, right? I mean, there is minimal coding required in order to create agents. You can actually create these AI agents with natural language prompts. And that is what Agent Force provides to our customers. And if you think about it, Agent Force is way more than just a platform to build agents.

It is what we call a deeply unified platform, which really brings all of your organization's data, your applications, your customer 360 into one place so you can build these agents with configuration and out code. I think that's a really important point because there's a lot of tools out there that you can build agents on now. There's just a new startup popping up here and there every single day. And I think what makes Agent Force so unique and so beneficial is that integration piece, that access piece that you just shouted about.

I really appreciate you sharing that. What makes this, if you could give us a before and after of pre-Agent Force, what maybe a workflow looks like and post-Agent Force, what a workflow looks like just so people can see. Theoretically, I understand it may be giving a little bit more of an example of this before and after of what it can actually do. Yeah, I think that's a great question.

And if you think about the pre and post AI agents world, I think a lot of us are used to chatbots that have been around for a while, and predictive intelligence that has been around for a while. And I think there's a lot of confusion around the differences between AI agents and these predecessor technologies. So let me dive a little bit into what

makes AI agents and agent force unique and different. So before we jump into that, I think it's also important to understand that we believe that there are five key attributes that make up an AI agent. And those attributes are, number one, the role, which really defines the job you want the AI agent to do.

Number two, the knowledge, which is essentially the data that an agent needs in order to be successful. Number three are the actions. So these are the actual things and tasks that you want the agent to perform in order to do its job. And then we have guardrails. So guardrails are really important because these are the boundaries that an agent can operate under.

Essentially, you are telling the agent what it can and cannot do. Finally, the channels, which are essentially the applications and surfaces, where an agent operates and lives. These could be, for example, things like your website or your mobile app. Now, to talk about the before and after, I think there is a huge difference between

the previous AI technologies like chatbots and predictive AI and AI agents. And if you think about it, a chatbot really is something which uses predefined rules, decision trees, and scripted responses in order to interact with the users. They're really limited in answering predefined questions. They cannot reason.

They cannot make decisions outside the hard-coded rules. On the other hand, AI agents can reason. They can make decisions based on the context of the interaction and then generate responses that are grounded in relevant knowledge. So they're not really limited to a specific set of questions. And one of the analogies that I always give that seems to resonate is think of a chat pod as a vending machine, right? It has a fixed inventory of

snacks or food, whatever you are vending. These are like the predetermined responses, right? There's a small keypad for user input so you have to stay within the context and confines of that keypad and you know it will only give you exactly what you asked for and select it. It's simple, it's predictable and it works really well if you want to serve that particular need. But in today's world, the AI agents, the same analogy as I gave for the vending machine

would be like a personal chef that has an impressive list of recipes, which is essentially the chef's knowledge and their ability to understand complex requests coming from their guests. These are natural language prompts that we are giving to the chef and saying, hey, I want this dish made this way. And the chef can not only create it,

but they can also learn new recipes based on our preferences. So I think that is a good analogy in terms of how AI agents differ from previous technologies like chatbots. Yeah, that is a great example. Thank you for sharing that. I will definitely be stealing that and using that in the future. So that's a really good one.

With this description, you gave those, was it five categories for an AI agent? What I found interesting about what you listed is it sounds like a job description to me, right? It's like, this is your role. These are the actions you can perform. This is the knowledge we need you to know. And so when we talk about this being digital labor or a digital employee, you're kind of giving it this job description

like you would a human employee or human labor. So I think it's an interesting shift from like, this is a tool or technology that is very static to now we're treating it kind of like a dynamic human being in certain ways with how we're training it, and we're gonna keep training it, and it's gonna keep getting better over time as we continue to work with it, just like a human employee. So I'm curious about your perspective on how managers can sort of prepare for this world

of digital labor, digital employees, plus human labor, human employees, because it's definitely gonna be a very interesting mix and we're already kind of there. Yeah, you're spot on. And in fact, because as I mentioned, these AI agents can act autonomously with minimal supervision and they can think,

they can reason, they can take action. They're not just tools. Just like you said, they are intelligent digital workers and that is where we believe that this whole notion of digital labor is going to be out there. And my prediction is that every single customer experience will shift due to AI agents. And I think as leaders and as folks in the software industry,

it is absolutely critical for us to be prepared for this revolution and making sure that we are ready to manage this digital labor. So I don't know if you may have heard, you know, our CEO Mark Benioff recently said that 2025 is the last year that CEOs will be managing human only workforce, because we're now going to have human and digital employees working side by side. And I would extend that statement to say that this is the last year that any manager, any people manager is going to be managing people-only workforces.

So I think adopting AI as an intellectual partner is one of the most critical skills that leaders can build in order to manage both human and digital labor together. So a few specific areas that come to mind in terms of the influence as well as where we can better prepare are, number one, decision-making and speed of innovation.

Because as leaders and managers, we make decisions every day. And there's vast amount of data available to us for decision making. And while that's a great thing, it can actually also become a blocker because analyzing all of that data, especially structured and unstructured data, can really slow things down. So one of the areas where we are seeing a lot of, you know, momentum is leveraging AI agents as digital employees to analyze all of the structured and unstructured data, to identify patterns, blind spots, and recommendations, and even for managers and leaders to simulate scenarios

to test their decisions can really help leaders make faster decisions that are more objective and data-driven. And I think, you know, another area would be really for managers and leaders to scale themselves in multiple domains, right? To be effective leaders, we have to have a wide horizontal view as opposed to like specializing in just one area. For example, we need to think about strategic planning, operational excellence, etc. Having AI agents working with us in our teams will really help us go deep into those areas.

Because I think trying to do everything ourselves, number one, is very, very hard. Number two, it's inefficient at best. Having AI help us will really help with our cognitive amplification. Finally, I would say maybe there's one more area that comes to mind, which is probably one of the most important ones, is innovation. Because I think inviting AI agents as part of your team, as your digital employees, into the creative and ideation process

will really generate and uncover new ideas and strategies that we may not have thought about. And I think this is kind of like having a creative, whiteboarding session anytime you want without having to bring a whole team together. And the beauty of this whole creative relationship with AI is what I call compound learning, meaning we as you know, humans are learning new experiences, but at the same

time, the AI is also getting smarter based on our input. So I think the leaders who embrace these concepts sooner will really be the winners in the future. I completely agree with that. Yeah, I think what I find really intriguing about digital labor and digital employees and these AI agents is that efficiency piece that you mentioned. Of course, the creative brainstorming that it allows us to do.

But I think what I've seen it already do is allow humans to be more humans because we can focus less on this data crunching repetitive we can focus less on this like data crunching, repetitive tasks and focus more on, oh, how can I actually provide like a great customer experience? How can I spend some more time thinking about this or meeting my customers or my employees wherever they are? So I think this like human plus AI future is not just one of like, oh great, we get to be more efficient and do more work and work harder,

but also one of I get to focus on the things that give me more joy and play in my work. So that's what I'm most excited about for this. 100 percent. I think as we think about, in fact, I spent a lot of time working with customers in the field, and there are some misconceptions out there as well. One of the misconceptions is really around the fact that, hey, the AI agents are going to completely replace human labor. And the reality is it's far from the truth.

That is just not the case. And what you mentioned about AI agents taking on a lot of these rudimentary tasks that are repetitive in nature, that are not really interesting and exciting in nature, I think really gives us as humans a lot more time to really focus on things that are, and especially from a customer experience perspective, our customers also need that human touch

and that conversation with the human, depending on the situation and the context. And I think now we will have a lot more time. It's interesting when I talk to a lot of my team members, they want to do a lot of that work. They want to spend a lot more time. It's interesting when I talk to a lot of my team members, they want to do a lot of that work. They want to spend a lot more time in those capacities, in those conversations. But the reality is that in order for them to do their job

and hit their KPIs, there's a lot of that work which isn't necessarily exciting work that they have to do. And that's just the nature of business today, right? So I think AI agents will make a huge difference. So yeah, I mean, I completely agree with your assessment. Are there any other misconceptions that you think people have around AI agents or just maybe AI adoption in general

when it comes to business? Definitely, I think as we have kind of like, you know, started working with customers, there is a bunch of different things where I would say there are some misconceptions that people are thinking maybe too deeply into and those are just not true. So as I mentioned, the first one is really around AI agents completely replacing human beings, right? So that's definitely something which we don't believe is going to be

the case. The second one I would say would be around some misconceptions about the accuracy of and you may have heard the word hallucinations, which is essentially the AI agents or AI making stuff up or answering the questions. And I think a big part of that really comes down to the data that you're feeding to your AI agents, right? So at the end of the day, your AI agents are only as good as the quality and the consistency of the data that you're providing to them.

So yes, if the data is not accurate, the answers might be inaccurate. So there are work rounds to it, which is clean up your data, make it consistent. And that's something which at Salesforce, we help our customers do every single day. So once you can fix the underlying problem, then a lot of these issues are on hallucinations

and inaccurate things go away. I think there's a misconception around trust. And as an example, at Salesforce, as part of our agent force platform, trust is just embedded into the platform. This isn't something that you have to build separately. That's why I talked about this notion of Godrails, because one of the big misconceptions customers have

is that AI agents can just go rogue and start working autonomously without any kind of human intervention. Again, that's a misconception which isn't just true because as an organization building agents using a platform like Agent Force, specifically Agent Force, you can put those guardrails around and you can actually tell the agent what it can and cannot do. And this isn't like too different than, for example, when you have a new employee come into your organization, right? The first day your employee comes in, you don't really open up the entire organization

and access to them. You give them access to what their job requires and then you kind of like, you know, start growing from there. So I think that's another area which definitely is a misconception. Maybe one more I will share is this notion of the fact that AI agents and the AI large language models operate in a black box, meaning you don't really know what's going on from a reasoning and thinking perspective, and you can't really control it. And I'll give you a perfect example

that if you're using Agent Force and the way we help our customers create and build and test agents is that when you are testing an agent, and in fact, testing is a very, very important component of deploying AI agents, when you are testing an agent, and in fact, testing is a very, very important component of deploying agents, when you're testing an agent using Agent Force, you can actually see the step-by-step thinking and the reasoning that the agent is doing. So you know exactly how the agent is

getting to a particular decision. And you have the opportunity to go and tweak it, give it different instructions, or give it different type of data. So I think that's a big misconception that the customers are struggling with these days. I mean, from like using, personally using ChatGPT, I can't see that reasoning, right?

There's no under the hood for me whenever I'm using a tool like that. So having access to something that does let you see, oh, here's where I went wrong. It's almost like whenever you're debugging code, but now I can actually read line by line what the reasoning method was and how I might be able to change that. So I think it's a really important component of all of this. You sort of shared now a lot of how you guys are

thinking about implementing Agent Force. And I would like to hear more about what your team is actually doing when it comes to implementation. So for the Agent Force Acceleration team, what does your day-to-day job really look like? I'm happy to talk about that topic. That's one of my favorite topics, you know, and before I jump into like my specific team and what we do with our customers, I think it's important to understand, you know,

how our team came about and how we came about so quickly. So number one, you know, customer success is one of our top values. Salesforce is a values driven company. That's one of the reasons why I've been around for 13 years myself. And it really is core to our top values. Salesforce is a values-driven company. That's one of the reasons why I've been around for 13 years myself. And it really is core to our company's mission and our team's mission,

which is making our customers successful. So our team's mission is very simple. Make our customers widely successful in their agentic journey using Agent Force, right? So we launched this team. We in fact launched Agent Force at Dreamforce last year. And I'm very grateful to be part of this team, and we launched this team shortly thereafter.

You know, our team is a team of highly skilled and technical AgentForce engineers, and we are laser focused on helping our customers create and deploy AI agents to achieve critical business outcomes. Now, the reality is that every company has more jobs to be done than the resources that are available to them to do those jobs or people available to them, right? And Agent Force really helps companies build agents that work together with humans to drive customer success.

And we say if you can describe it, Agent Force can do it. I love that. In a natural language, in a natural language description fashion. So my team's responsibility and my responsibility is to work with our customers closely in a hands-on capacity, help them ideate on what use cases they need to be thinking about, help them create, test, deploy agents in real life, in real time, and really help them on their journey.

So I think we are 100% focused on agent force. That is the one thing that we are focused on. So anything which is around helping our customers build and deploy these agents and any kind of technical blockers that come along with it, those are the things that we help our customers remove and get live with agents as quickly as possible so they can get to their business outcomes as quickly as possible. I think the timeline of all this is actually really interesting because for those who don't know, Dreamforce was September last year, right? Or was it October 2024?

Yeah. So it's been less than a year that you've started to go through this whole journey of implementation. So we're still like really early stages from a technology adoption standpoint and actually seeing how does this work in my company and what are the results? So from your perspective, again, acknowledging that this has only been seven or eight months now, what use cases have you seen be the best ones so far for

companies? And what are you seeing people really dig into and love with what they're doing with Agent Force? Yeah, that's a great question because I think, as I mentioned, in terms of what we do with our customers, we're not only helping them with the technical aspects of Agent Force, we are actually helping them with identifying the right use cases because I think that is, in fact, one of the blockers is starting with the wrong... That's not one of the hardest parts, for sure. Yeah, figuring out what to do with this new technology. Because it's like, oh, I could do anything I want. How do I narrow that down?

So I think that's probably one of the key portions that you guys get to help with. 100%. So narrowing it down and really focusing on the right use cases is the absolute important thing, right? So before I jump into the use cases, I think I'll give you a little bit more perspective in terms of the areas where we specifically

help our customers because that will actually help identifying how we help them with the use cases, right? So as I already mentioned that, you know, AI agents set up configuration development, testing all of the things that we help our customers with. But then if you think about agent force and AI agents working, there's a lot of other components that are required in order for the AI agents to be successful. We already talked a

little bit about data, but our team really helps our customers with technical architecture guidance, whether it comes to data cloud, you may have heard terms like RAG, which stands for retrieval augmented generation, or large language models. So a lot of the technical architecture guidance is needed in order to have agents to be effective. Sometimes there are technical blockers around integration and other systems where the data might be residing. So all of those things are things that our team provides those recommendations.

And in some cases, there are even some non-technical things that can be a blocker around change management. So these are the type of things that from a day-to-day perspective, you know, my team engages with the customers. Now, going back to the question about the use cases, I think what's interesting is that there are a couple of different lenses that I apply when I think about use cases. You know, there are use cases that are specific to industries like manufacturing or retail or financial services. And then there are use cases that are specific to the organizational persona. Like if you are part of the customer service organization or if you're part of sales or marketing, there are different things that you could be using the

AI agents for. But eventually, I think the best way to think about the different agentic use cases is to think about the different agentic use cases is to think about the jobs that your organization needs to do. And then we break it down into two broad categories, external customer-facing use cases where your AI agents are interfacing and interacting with your customers, and then internal employee-facing use cases where the AI agents are helping your company employees. So from a customer-facing perspective, I think customer service and support continues to be the most prominent and most common use case.

I think this is where there is, you know, most amount of customer interaction. So these are things like when customers are reaching out to your organization for different inquiries, they have different questions, there are different issues that need to be resolved. So that is where the agents can really help customers answer those questions.

But then you can really start taking it next to the next level by taking actions. For example, the agent might answer a question, but if the question doesn't resolve the customer's issue, the agent can actually now go ahead and create a case. The agent can escalate the issue to a human. They can schedule appointments. They can authenticate users.

So there's a lot of those actions that you can start building on top of the basic FAQ or answering question kind of capability. Then we see use cases around order management. That's a really big one. One of the industries that I focus on is retail and consumer goods, which is, by the way, a fantastic industry.

There's a lot of really interesting logos I work with. So things like if I placed an order and the order is hasn't arrived, where is my order? That's kind of like one of the most common use cases, processing returns, changing my delivery address or processing refunds. So these are all the things that fall under the bucket of that order management use case.

And then as I mentioned, there are industry-specific use cases, like sticking with the retail industry. One of the most exciting use cases we are seeing is this notion of a personal shopper. So the reality is that recommendations aren't new. Different retailers, if you go to their website, have been recommending products for a very, very long time. But that has been very, very predictive,

and that's very limited. So now what we believe is that the AI agents that we are helping our customers build today, they not only know your purchase history, but they actually know your style. They know maybe the event that you're trying to attend. So they are going to make personalized recommendations that aren't just based on what you're browsing,

but rather based on who you are as a person. And then finally, the second component of these use cases is really around what I mentioned as employee use cases and the possibilities there are, Lacy, completely limitless. I mean, anything that you can think of within your organization to help your employees save time, coach them, those are the type of use cases. So a really big use case is sales coaching, where you can now have AI agents that can provide real-time context-specific coaching to your sales reps or your any kind of employee, right? We also see a lot of HR and employee self-service

use cases. In fact, at Salesforce, we use these use cases ourselves. We're a big user of Slack. And Slack, by the way, is a fantastic, is a fantastic collaboration platform. So within Slack, we have our own AI agents. We are, if I have a question about my benefits or a particular company policy, or if I want to know how to manage my time off or expenses, et cetera,

now I don't need to pick up the phone. I don't need to log a ticket. I can literally just have a Slack conversation with an AI agent. It will give me the answers and depending on the situation, if it can't solve it, it might create a ticket and then route it to the right person. So I think as we think about all of these things, they are different use cases that we are seeing commonly applied at customers. And again, as I said, that we at Salesforce

like to drink our own champagne. That's something which we are doing ourselves. And one more use case I'll throw out there, which I've been using personally myself very extensively, is this whole notion of knowledge management and knowledge answers, right? So if you think about it, almost 80% of an organization's data

that sits in your company today is unstructured. So these are things like emails, PDFs, and Slack conversations, et cetera. When you want to find an answer, in order to find the answer, going through all of this unstructured data can take hours and hours. So now AI agents can actually sift through all of this data, structured, unstructured, videos, transcripts, et cetera,

and then it can basically give you the answer. So you ask the question, you get the answer. You don't have to go through tons and tons of those data elements. So I think those are a few very exciting and common use cases that we are seeing across the board. That last one I love. We use the enterprise version, or the business version,

of ChatGPT in our company and we're able to hook it up to the Google Drive and it can search all of our documents from years past and find, oh, here's the blurb about this. Here's this thing. It doesn't go as deep as Slack or email, unfortunately, but I find that that ability to just be able to search our knowledge super quickly is just so, so, so helpful. So I love that function. I'm curious, are you seeing percentage-wise more internal application of AI agents versus external, or is it roughly the same?

Yeah, that's a great question. And what I would say is that part of that also depends on the industry that the customers are in. For example, you know, if you think about some of the regulated customers that sit in regulated industries, that there's a lot of different compliance and policies. So in order to test out agentic use cases, those customers try to do it internally. So there is minimal risk as opposed to, for example, if you are a retailer and you are simply answering questions about the size of the attire or the clothing that you're selling, I think that's like a low risk. Even if there's a little bit of a mistake made there, it's not that much exposure.

So I would say it's a pretty good balance because we work with customers across all different industries. But I would categorize customer service and support as really the top one. Order management is kind of like really the next one. And then now we're starting to see a lot of employee and internal facing, which is again, going to be applicable to every industry, every company. But we're seeing more of that happening

in the regulated sector more. So we talked a little bit about different industries, but I'm curious, we haven't talked about company size. So I'm curious, is there like an ideal size for a company to make this investment into AI agents? Or is this something that you think no matter what industry, no matter what size, no matter how old or young your company is, this is something you should be looking at?

I would say that if your company today requires people and human employees to do the jobs you want to do, you're ready for AI agents. I mean, that's kind of like the litmus test that I would put. So it doesn't really matter if you are a startup and you are only a few people. And it also doesn't matter if you are a big company like Salesforce with almost 75,000 employees. I think the value prop is absolutely there.

So, and in fact, I would say that for smaller companies, there might actually be a bigger value prop because for smaller companies, they are constrained by the number of people that exist there physically, right? So you want to do a lot more for your customers and everybody's wearing different hats, right? I mean, I actually worked in the startup

for a few years myself. And it's interesting that even though I had a specific title, I had a specific team, but I was actually playing a lot of different roles because it might really be all of them, right? So I think for startups and small companies, this is going to be a phenomenal opportunity to do a lot more with less resources.

And I would actually throw in one very interesting use case out there, which is, as I mentioned, that we use our own AI agents internally. And one of the use cases that I found very inspiring is coaching, right? And certifications, right? Where we, within Salesforce, we do our own corporate certification and a bunch of different things that are required. And instead of like presenting to my manager, I present it to an AI agent,

just like I would in real life. And the AI agent is able to listen to my pitch. The AI agent is able to give me real-time feedback on what worked well, what didn't work well. And as you apply this lens to company sizes, think about it. Smaller companies, in many cases, don't have the budget or the resources and access

to leadership training that, for example, a big company like Salesforce might have. Now with AI agents, because these AI agents are trained on the same data set as large companies use, we'll see a lot of democratization of training and knowledge management where it doesn't really matter how big or small your company is. You can basically get the same type of training

and leadership development as large organizations. Yeah, I love that. I love that. As you were talking, I was thinking more about these different use cases. And we talked about how this initial discernment of this is a good use case for agents and this one maybe isn't, or maybe we should put this further down the roadmap. What advice would you give to a company that

is looking at different use cases and they're trying to figure out which one would be, quote unquote, agent ready? I think for companies, because I hear this from customers all the time, that hey, is my company ready? Is my department ready? And I tell them the time is now.

As I said, if you're ready to hire employees to do a job, your company is already ready. But I think from a use case perspective, that's a really great question, because as you mentioned earlier, that's a really great question because as you as you mentioned earlier, that there are hundreds of things that AI agents can do. So the question then becomes what is the starting point? And where do we start? How do we prioritize? So I think, if you look at all the different jobs that your organization needs to do, generally speaking,

we recommend that hey, if you have use cases where there is clearly defined and repeatable process, there are clear inputs and outputs and decision points, those are the type of use cases that are easy to build, test, and deploy. So you can really test the new AI agents, and then you can start showing the value. We always recommend to start with low or moderate risk profile.

For example, I mentioned Verasmai Order or FAQs. These are, I mentioned, where is my order or FAQs. These are generally speaking, you know, low risk situations because this allows you to build your agentic foundation. And once that foundation is built, for example if you have built an AI agent that allows your customers to track their orders, gives them frequently asked answers to frequently asked questions, then you can start adding more and more actions on top of it,

where you can add an action that says now create a case, process and return. I would recommend don't start with a use case where you're refunding customers because if errors happen there, you've got financial exposure. Start with low to moderate risk profile. I think a really big component of starting use cases is having access to high quality and relevant data because a big part of Agent Force is, and one of the value propositions and

differentiators of Agent Force is our ability to ground the agent's work and actions in your company's data because that is where all the context is sitting, right? Because you need to know who this customer is, what is their purchase history, what kind of opportunities exist. So all of that is what we call grounding the agent into the relevant data, and Agent Force is able to do it. So having access to clean data in order to make decisions, take actions, is very, very important.

And it's interesting that every company has sets of clean data, like if you think about your knowledge base or your billing data, generally speaking, those are clean sets of data. So start there. And maybe a couple of other things I would say would be start with high volume, high frequency use cases that are taking too much of human time. If you have those type of use cases, you can actually show quick ROI to your organization.

And the last thing I would say is being able to measure the outcomes is very important. So having a clear value prop in terms of, okay, with these AI agents, who is benefiting? How are they benefiting and by how much? For example, if we can say that, hey, you deployed AI agents into your call center and your average handle time has gone down by 20%.

That's a very tangible number that the organization can understand and work with. I think what's interesting too about this idea of starting with something that is measurable or highly repeatable versus the there's some risk with doing something with financial information, for example, or doing offering refunds is that it gives your team an opportunity to get used to working with AI agents as well. Because there is a huge component of this that's not just technological. It's like the people, how do we get our teams

ready? How do we get our teams thinking about this and as part of their workflow daily? How do we get them bought in and not let them fall into a place of fear that they might be replaced? So I think I like that you would start there because it gives an opportunity in so many different ways of not just being able to use the technology but also work with the teams. So from that perspective, is there any advice you might have for people

who are thinking about that change management piece that's really like people focused? Yes, I think that is a great question because there's definitely a mindset shift that we all need to make, especially as leaders, because our teams follow the managers and follow the leaders in many cases. So it's very important to understand why are we making this change.

So change management in any technological shift is very important. But I think in this particular transformation, which is probably the largest, the biggest transformation any one of us has ever seen, it is going to be absolutely paramount. So so from a change management perspective, I would say a few things, right? The first thing is, it is very important for the leaders to start adopting AI themselves, because it is very difficult to champion something if you authentically have not leveraged that before or you haven't experienced it yourself, right?

So I would say maybe pick one or two areas in your daily life, apply AI in that, and AI agents into that flow of work so your teams can see that, right? Number two, it's very important for everybody in the organization to understand what's in it for them, right, so because again, if I'm a call center agent, and I am, or maybe we've talked about call centers a lot,

let me talk a little bit about software engineering. So you may have heard of new companies coming up like Cursor and companies like that that actually write code, or software engineers, in fact, Salesforce, we were one of the first companies to build an AI agent called Code Genie that wrote a lot of code and still writes a lot of code for our engineers. So if I'm a software engineer and there is like, and I want to build an amazing product, you know, there's a lot of repetitive

work, which may not be super value added, but I kind of like have to do it in order to build those building blocks. If I can have an AI agent, which is working side by side with me as an assistant, take on all of that work and spend, you know, now I have magically 20, 30, 50% of my time open up so I can really focus on creative problem solving, I think that's something that will drastically boost productivity and it will really give people that ability to work on interesting creative problems. So I think making sure that folks understand that, hey, we're not deploying this coding agent in order to replace you. Rather, we're deploying this coding agent

so our overall engineering productivity goes up. That is very, very critical. So I think having the mindset shift, explaining the why, explaining the what's in it for me, for every individual who is going to be working with these AI agents is very, very important. And I would maybe say one more thing, which is maybe two more things. One is around this culture of experimentation and innovation. I think if you think about generative AI, you know, it's a new field. Obviously, there's a lot of, you know, improvements

that are happening every single day. Like the AI models that existed two years ago today, they are completely different. They are their capacity, their ability, their intelligence. And it's not about having the right answers on the first go. It's about asking the right questions. And in order to create that environment of asking the right questions, we need to create a culture of experimentation and innovation that even if we don't get it right the first time, we shouldn't just shut it down. We should like keep working. We should keep giving feedback. So these models and these AI agents become better and better over time.

So I think those are a few things that I can think of from a change management perspective. And this is an area that is evolving so rapidly. What was true three months ago or six months ago is not true today. But these are some fundamental change management things that I would say are very, very critical. And maybe the last thing is adoption. Just because you have an agentic technology available to you or an AI agent available doesn't mean

that people are going to start magically using it. So making sure that we are focusing on user adoption and adoption of these technologies is very important. So I love all these lessons and takeaways from change management that you've shared with us, but what other lessons and challenges have you guys faced in the early phase? I mean, again, it's only been like seven or eight months, so I'm sure you guys have just been noticing one thing after another thing after another thing and the lessons probably haven't stopped yet And they probably won't since the technology is improving so dramatically all the time

So what advice would you have for people that are maybe looking at starting their agentic journey now? That like you wish you would have known a couple months ago. Yes, and I'm smiling because you know, yes the the lessons We're learning every single day and what's interesting is that Agent Force adoption has just been mind-blowing. We have signed up thousands of customers. And what's really exciting is that customers are ready and they're hungry to get on this journey. And then that's what we're doing, helping our customers. So the speed at which customers have adopted and embraced Agent Force is just unprecedented.

And the speed with which these AI agents are being created is just unbelievable. So yes, I mean, there are some lessons that we have learned, like some things that were more hypotheses at the beginning and now we've actually seen it in action and it's kind of like been proven that yes, these are the things that are absolutely necessary. I would say the first one, which I've highlighted a little bit is really data is the key to success. So AI agents are only as good and effective as the knowledge that they can access. So having outdated or incomplete knowledge bases

leads to poor answers and hallucinations. So I think data is the key. The other really big piece we have seen is how you have set up and created your knowledge base. So there's a big difference between, so for example, a lot of times customers would have, hey, we have these thousand articles, so why can't the agent use it? So there's a difference between a well-written

and a poorly-written article. So kind of like, you know, taking it to one level deeper is very, very important. Now, one of the things you may have heard probably two years ago or a year and a half ago is this notion of prompt design and prompt engineering. And I think everybody knows that, hey, the question

that you give to your AI agent is the prompt. But the importance of prompt design and engineering has just been unbelievable in terms of what we see with the customer. So I think it's all about asking better questions. Because the better question you ask, the better instructions you give to your AI agent, you will get better and better generative results.

So giving clear instructions, well-defined instructions, avoiding broad categories, like, again, I'll give the analogy of a human employee, when you hire somebody new in your organization, you don't really just give them all the manuals and all the policies, right? You start with what's part of their job. So you kind of break it down into modular design. And the same thing applies with AI agents, right? So that's what we recommend to our customers,

that break it down into modular design and then build from there. And focusing on guardrails is very important, right? A lot of times, we see that customers are focusing on what the agent should be doing. I think what's equally important is what the agent should not be doing and what they cannot do and what are the places where they need to really escalate to humans.

And that is where this notion of principle of least privilege comes in, which is when you build an agent, start with the least amount of access and then start building more and more access as it is needed. Another big lesson we have learned, which again is a common theme in software development anyways, but I think in the agentic world, it is absolutely critical, which is

this notion of continuous iteration and testing in order to make your agents better and better. So what's interesting is that in generative AI, which obviously AI agents use generative AI, nothing is really hard-coded, right? Result one that you get from the same question might be slightly different than result two. And this isn't like too different than a human conversation either. For example, we're having this conversation today. And unless I'm using a script, if you ask me the same question tomorrow or next week, the two