Understanding the Data Cloud and Its Role in Enhancing AI Agents

Today on the Salesforce Admins Podcast, we talk to Amit Malik, the Content Portfolio Lead for AI within Product Education at Salesforce. Join us as we chat about how we can teach AI effectively to admins and the easiest way to learn Agentforce.

You should subscribe for the full episode, but here are a few takeaways from our conversation with Amit Malik.

The shift from knowledge to value

As the Content Portfolio Lead for AI in Product Education at Salesforce, Amit is the perfect person to talk to about where admins should get started with learning Agentforce. After all, his job is all about planning the courses that are offered globally about Agentforce and Data Cloud.

What Amit emphasizes is that past knowledge matters less than what learners do in the next 12 months. Agentforce’s capabilities are growing with every release, so he recommends focusing on understanding the core concepts of how AI works and building from there.

Malik’s agent framework for admins

When you’re building with AI, the first step is aligning on why an agent is needed in the first place. From there, he recommends asking five questions to guide your process:

1. Is an AI agent the best way to solve this problem? Would it be easier to build a flow? Just because you can solve something with Agentforce doesn’t mean you should.
2. What agent type do you need? Salesforce has several pre-built agent templates for specific use cases, like Service Agent, Employee Agent, or Guided Shopping Agents. Consider those options before trying to build something more complicated.
   
3. What topics do you want to assign to this agent? Define the set of business problems you want your agent to solve. There are standard pre-built topics like FAQ or escalation, but you can make a custom topic if needed.
   
4. How will you provide data to your agent? AI is only as good as the data you provide it, so you need to make sure you have everything you need in Data Cloud and set up access with the Agentforce Data Library.
   
5. What actions do you want the agent to perform? “This is where the magic happens,” Amit says. There are four types of actions: Flow, Apex, API, and Prompt Template.

It’s important to understand that Agentforce is the final layer of the org, the interface. An agent is really an aggregation of the topics you build it to solve. Those topics comprise the specific actions you enable your agent to perform, and those topics, in turn, are possible based on the data you’ve integrated into Data Cloud.

Adopting a learner’s mindset

For Amit, the most important thing you can do if you want to learn Agentforce is to adopt a learner’s mindset. “The art of learning is to become curious,” he explains, “education is not about the instructor, it’s more about the learner.”

Where many businesses go wrong is viewing AI as a solution in search of a problem. But once you start to get curious about the specifics of your business processes, you’ll start to find all sorts of new ways that AI agents can help. And building around these specific, real-world use cases is the easiest way to get started mastering Agentforce.

Amit has a lot to share about learning, teaching, and working with Agentforce, so be sure to tune in for the full episode. And don’t forget to subscribe to the Salesforce Admins Podcast to catch a new episode every Thursday.

Podcast swag

• Salesforce Admins on the Trailhead Store

Learn more

• Salesforce Admins Podcast Episode: Architect Courses for Admins with Amit Malik
• Trailhead: Discover Agentforce
• Trailhead: Review Agentforce and Data Library

Admin Trailblazers Group

• Admin Trailblazers Community Group

Social

• Amit on LinkedIn
• Salesforce Admins on LinkedIn
• Salesforce Admins on X
• Mike on Bluesky social
• Mike on Threads
• Mike on X

Full show transcript

Mike Gerholdt:

This week on the Salesforce Admins Podcast, we’re welcoming back, Amit Malik to talk about a topic that’s very top of mind for every Salesforce admin right now, which is how do we teach AI effectively? Amit’s going to share with us some insights from over a decade of Salesforce instruction diving into why teaching AI is a little bit different now, and how admins can use his five question agent framework to put to better use. Now, whether you’re just getting started with Agent Force or you’re already experimenting with Data Cloud, this conversation’s packed with practical ways to bring clarity to your learning and to your users. So with that, let’s get Amit on the podcast.

All right, Amit, so welcome to the podcast.

Amit Malik:

Thank you.

Mike Gerholdt:

I know a few years back you were on to talk about the architect mindset, so I’ll be sure to link to that episode, but now we’re talking everything Agentforce and Data Cloud and Metadata and Customer 360. But for people that haven’t been around listening to the podcast for three years, and there’s a few of them, could you reacquaint the audience with what you do at Salesforce and your journey to Salesforce?

Amit Malik:

Sure. I joined Salesforce in 2013 and I have been lucky to teach audiences across the globe on Salesforce technologies every year, and I’ve trained on different topics about Salesforce, like whether it’s about teaching administrators or developers or consultants, architects. So I’ve been fortunate to talk about Salesforce aspects to different audiences. In my current role, I’m working on as a cloud content portfolio lead where I’m specializing in Agentforce and Data Cloud. So my role is to plan what kind of courses should we offer to our customers globally so that we can enable our ecosystem on Agentforce and Data Cloud.

Mike Gerholdt:

Yeah, I mean, there’s a lot to learn. I started using the platform back in 2006 and even trying to keep up is a lot, and so I can imagine learning constantly, there’s so much.

Amit Malik:

It’s easy. I would say nobody cares what we know in the past. What matters is what we are working today and for next 12 months.

Mike Gerholdt:

Well, there you go.

Amit Malik:

So I’ve always retrained my mind that what I have done in the past does not matter today. What matter is what I will do in next 12 months.

Mike Gerholdt:

So let’s talk about what we’re going to do in the next 12 months. I would love to know, this is going to be my first hard question, they’re all hard, I think, not to scare you. I don’t think it’s hard. But I am curious because you’re on a different side of the fence than I’m at. What is different about teaching AI than teaching other technologies?

Amit Malik:

I would say the interesting part here is there is knowledge explosion in the current times. And when a learner is learning, he has multiple channels to learn from and there is no direction what is the right way to spend your time. So what we need is knowledge distillation. We need to tell our learners, if you only have 10 hours, what should you learn in your first 10 hours. Or if you’re not time-bound, what are the first 10 words should you know to do your project better in next three months. That my suggestion, that we should stop the noise and try to develop our attention span in these current time of knowledge explosion.

Mike Gerholdt:

I really like that because there is a sense of needing clarity when there is this much noise out in the world. Do you find when individuals come to your classes that they have a preconceived idea of AI that you might have to work past?

Amit Malik:

Yes, because the interesting part is that, for example, let’s take a word AI agent. If someone is learning about AI agent from Google’s perspective or OpenAI perspective or how Claude thinks about it, how we think about it at Salesforce, we may have different explanations, but at the core, AI agent is an AI system. So once we start getting a clarity of thought that where are we heading, we are trying to suggest how do you build a digital labor who will work along with the human at present. Once you know the goal, I think then it does not matter what is your understanding. You just need to understand why are we learning AI agents. Just because people are saying about it? Or do you really believe in the spirit of, “Oh yeah, it makes sense. If I can get my job done autonomously by a AI system, I can be more productive than I am.” I think that’s where, the why has to start first before you start doing what is what.

Mike Gerholdt:

Yeah. Yeah, I often think a lot of times solutions come before people see problems and so they make up problems for the solution when in fact they didn’t need to make up the problem. They already had it.

Amit Malik:

Very well said, very well said. I would say in a different way that for every problem, Agentforce is not the answer. So we need to find out a use case and say, “Oh, this is fit for predictive AI, this is fit for generative AI, this is fit for autonomous agents.” So when you start as an architect, you start questioning to yourself, “How should you think to arrive at the solution that, oh, this is a use case of Agentforce now.”

Mike Gerholdt:

Yeah, absolutely. So as somebody that has instructed admins, architects, probably even developers, I mean, you run the gamut, because I know I’ve signed up for developer courses. If you had to describe Data Cloud to an admin and give its relevancy, what would you say?

Amit Malik:

Very good. So think of Data Cloud as a portion of all your data from multiple data sources. It’s a one repository where we want to bring data from multiple data sources. It could be Snowflake, it could be Databricks, it could be AWS, it could be any data source in your enterprise business. And once we have all this data in Data Cloud, that can act as a knowledge for your agent. As the analogy, think like this. When you join as a employee to a new organization, you need to learn about the system. In the same way we want to give the knowledge about systems to our agent so that they can understand the business and then talk to the customer on our behalf about our business.

Mike Gerholdt:

And I even think beyond just the customer, I mean, even the employee. If you only set up an agent to look internally within your CRM data, that’s almost like giving them a book, whereas if you use Data Cloud, it’s like giving them a library. Is that a kind of a competent analogy? Would you use that?

Amit Malik:

Yes. So see, Data Cloud is in simple way exposing all the data sources and how we attach it to the agent technically with the help of Agentforce data library. So we do not want our learners to get worried about Data Cloud so much because in the future or in the present world, we are trying to encapsulate the plumbing of Data Cloud. So if you’re a [inaudible 00:08:50], we just ask you to upload the PDF file to Agentforce data library, and then we take care of all the plumbing behind the scene. So as the admin, you just need to know how to upload your data and we take care of everything. And that’s the beauty of Salesforce platform that we try to make things simple because our engineers have done a complex job for you.

Mike Gerholdt:

The irony is it’s usually the businesses that make things complex with processes or approvals or reviews.

Amit Malik:

That’s fine. See, we cannot change the business. I think as a technical person, what I’ve learned over the years is never challenge the business, rather adapt to business.

Mike Gerholdt:

That’s very good.

Amit Malik:

Because that’s a very important learning. If you keep on challenging, because as a technical mindset, we always think, “No, business is not right,” but hey, business exist and that where technology exists. So we need to adapt to the business needs and that’s where as you grow as an architect, your fight is not to learn technology because you will have lot many people under you who knows much better than you. But your job is to understand what business value business need to create and how do you bridge that gap with the technology solution. That’s where the real fun is. You tie up the business outcome.

Mike Gerholdt:

And I feel like, because I was just going to ask you, back to our first question, how do you kind of filter out the noise? I feel like your previous answer is kind of that, right? Really focusing on what the business outcome, the business need is, right?

Amit Malik:

Yeah. So I would like to give you a very small framework which I have been using for myself as I’ve been learning Agentforce from the front, and I’m lucky to be part of all the teams here whom I interact with internally.

Mike Gerholdt:

I love frameworks.

Amit Malik:

The first question I would like to give to my learner is ask yourself, do you need AI agent? If the answer comes yes, that yes, you need AI agent, then move forward in Agentforce thinking, because it could be that you can handle something without AI agent and you don’t need it. Once your first question is answered to yes, then you move to the second question, what agent type do you need? For that, you need to know what agent types are offered by Salesforce. Like we offer, say, customer agent, which is implemented through Agentforce Service Agent. We offer employee agent, which is for employees. We offer sales agents like SDR agent for initial outreach and booking meetings. We have coach agent, which is for coaching and mentoring. So once you understand our offerings, then you see, okay, given a use case where customer says, “I want my agent to answer business questions and handle my refunds, or tell about my loyalty balance,” we know that I need something for service, I need something for customer, so I need a customer agent.

So once you start reasoning that I need a agent and I need a customer agent and I can implement customer agent using Service Agent, then comes a third question, what topics do you want to assign to this agent? Now topic is a very interesting vocabulary which we have launched in Salesforce ecosystem, which is how do you define the job to be done? So think of topic as a aggregation of your business process. So if you want your agent to work on, say, loyalty balance check. So that’s all the questions about loyalty balance will be handled by loyalty management topic. So that loyalty management topic is like a grouping of all the actions. Once you know the right topics that do you want to leverage the existing topics like standard topics, like say general FAQ or escalation, or do you want to make a custom topic, your third question is answered.

Once you have answered the third question, the fourth question is, how will you provide data to your agent? And here where your Data Cloud come into play. Because agent is dumb. Agent does not know anything about your business. So to provide the value to your agent, you need to connect your agent with the data library. And now with Agentforce data library, we support web search, we support files, we support your knowledge articles. So once you understand how do you want to educate your agent, your agent is now having a data about your business. Once these four questions are answered, the fifth question is, do you just want agent to answer the questions or do you also want agent to act upon it? Customer says, “No, I want agent to act upon.” And then comes the fifth thing, which is what actions do you want agent to perform?

Now this action is where the magic happens, and this is where all the administrators and developers can start working on leveraging their past knowledge. So if you’re good in flows, you can have a flow agent action. If you are good in Apex, you can have Apex agent actions. If you’re good in API, you can have new soft calls, you can call external services. If you’re good in prompt engineering, you can use Prompt template. So depending on the use cases, we give you four reference action types, which is Flow, Apex, API, and Prompt template. Once you have answered to this five questions, now you know a lot about solutioning. So this framework will help you through how to position the right solution for your customer.

Mike Gerholdt:

I really like that. Thanks for sharing that. And I was thinking through as you were talking through that because we have, I mean, internally we use Salesforce on the admin team to manage our content. And Josh Birk built what we call Agent Goat, which helps us answer questions and create relevant records. We even did something fun, I don’t know if you think this is cool, we think this is cool. We uploaded the release notes to Agent Goat and gave that as a resource.

Amit Malik:

Wonderful.

Mike Gerholdt:

Yeah. Because then we can ask it about new features.

Amit Malik:

And recently we launched our Salesforce documentation action. So we can just plug in that action to our agent and now it can start answering questions based on our Salesforce documentation. We just launched.

Mike Gerholdt:

I mean, there’s so much to do.

Amit Malik:

Yes, that’s the fun thing. Once you understand that agent is nothing but aggregation of topics and topic is nothing but aggregation of actions, and actions can happen on your data. So agent, topic, action, data, that’s a master framework.

Mike Gerholdt:

Right.

Amit Malik:

Once you start thinking on this four words, agent, topic, action, and data, everything will start falling into place.

Mike Gerholdt:

So I’m curious, and I genuinely don’t know the answer to this, but have you been to any of our events and happened to see any of the agents that customers have built that you find really inspiring?

Amit Malik:

Yes, I was lucky to be part of TDX Bengaluru Hackathon.

Mike Gerholdt:

Ah, yes. Tell me more.

Amit Malik:

It was so wonderful to see our customers showcasing the real use cases. Let’s say a job application agent, some student is applying for a job, that how we can have a agent which can read the resume and do the shortlisting and send the further steps for the candidate. Or it could be other scenarios which customers were showing in insurance that how we can process the claims with the claim agent. Or as we know in the Conquer, for example, expenses, how we can have expenses being reviewed by agent and auto approve, saving the time of senior management to approve the expenses. So see, once you start thinking of use cases, then the value is derived. The value is not in technology. The value is in applying the technology to benefit the customer and improving the customer experience.

Mike Gerholdt:

And would you also extend the customer for admins and architects and developers to users of Salesforce, not just external customers of your organization?

Amit Malik:

Yes. So let me define this. So here, when I say customer, so customer is, say, Marriott who is buying our Salesforce license, or any customer who buys our Salesforce license. And then in that customer, we have employees who are using our product, they’re our end users, and if this customer is further implementing our solution for their customers, then they become end user.

Mike Gerholdt:

Yep, that makes sense. Amit, I’m curious, in the instructional side of Salesforce and having done instruction for a long time, if you were to look into a crystal ball a year out from now, how do you see AI influencing instruction at Salesforce?

Amit Malik:

See, it’s very funny. I would say this. Everyone believes that in the future we all will learn through ChatGPT, Claude, and we’ll keep on learning through AI tools. But to me, there’s a need for clarity of thought. Because if you don’t know what to ask, you will never get the answer. So the magic of learning is your curiosity. So as everybody’s moving, we are moving towards AI teaching us step by step, but if you do not know how to leverage AI, then AI will be useless to you. So the art of learning is to become curious, and once you know how much curious you are, then AI or human does not matter. Then you can learn from AI also because you know how to ask right questions. So I think, in my mind, education is not about the instructor. Education is more about the learner. If learner is good, he can extract more from a teacher. If learner is not curious, then teacher or AI does not make a difference.

Mike Gerholdt:

That’s a really powerful point.

Amit Malik:

This is my experience of teaching for last 20 plus years that I’ve seen some interesting audience and it changes the whole class when you interact with the different kind of groups. Just a good example, the way you are asking me questions, that is a prompt to my mind and I’m giving you live answers. So you are getting best out of me because of you, not because of me. So it’s just a very interesting phenomenon to understand that it’s all about how much you extract from my mind. My mind is prepared.

Mike Gerholdt:

Yeah. I never really thought of it that way of when you go in, because as a customer before joining Salesforce, I’ve taken quite a few of our, I believe five day or four day classes, instructor led. One of those was with Wendy Braid.

Amit Malik:

This is amusing.

Mike Gerholdt:

They were all great, but they all, to your point, all of the classes were different. And they’re different, well, one of them just because the instructor is different, but because the people are different and the goal of what the people are trying to get out of the instructor changes, and it also changes based on the level of curiosity, I think you said that, of the learner, right? So if they’re genuinely curious, they’re going to get more out of the learner than if they’re just kind of sitting there expecting to be brought up to speed, right?

Amit Malik:

Exactly. I would like to add one more point to this discussion here.

Mike Gerholdt:

Oh, please.

Amit Malik:

So the role of the instructor is not information sharing. Information is available in abundance through internet channels these days. So the goal of the learner is to help learner connect to their business goals. So when a learner is listening to the instructor, instructor need to understand what is the goal learner trying to achieve in their business goals. And you only get to know that when you’re in the classroom. Before that you don’t know. So how can you plan your speech? And that’s where I believe in dynamic speeches where the real agenda gets opened up when you start engaging with the real audience. Topic remains same.

Mike Gerholdt:

So I just had a question pop to me that I really thought brings this kind of all the way 360. One of the roles that admins, developers, and architects play is your role. They have to be an instructor to end users. When they roll out a new feature, whether it’s a new application, Data Cloud, an Agentforce agent or whatever, part of their role is to teach their end users what the feature is, how to use it, how to be productive in their role. With your vast experience in teaching and educating individuals, what advice would you give to people that kind of have to temporarily step into that role and be productive?

Amit Malik:

I would say listen more, speak less, think more and process your thoughts. Articulate your answer well before you speak to the listener, because sometimes we try to propose what we know and what we want, but that is not the winning edge. The winning edge is trying to understand what is the business problem customer is trying to solve, and then within your mind, you know all the products what we offer, and then find out the best solution and say, “This is what will suit you. Use Agentforce Service Agent, and that will solve you in this use case. Let me show you how.”

Mike Gerholdt:

Yeah, well, that’s powerful advice. You would think it would be the opposite, right? Talk more. But sometimes there’s a reason we have two ears and one mouth.

Amit Malik:

I agree. And I have learned, again, I’m fortunate thanks to God that I’ve learned this by, I’ve been teaching for more than 20 years now, I’ve learned this skill in real world that the best teachers speak less and personalize to diversified audience where every audience should feel that he is speaking for me.

Mike Gerholdt:

Yeah, I like that. Thanks so much for coming on the podcast, Amit. I appreciate these check-ins and really giving us this perspective on what our team is doing to get people up to speed and really also just have admins and developers and architects think about how they can be better educators themselves when they have to demonstrate the platform to their end users. I appreciate this.

Amit Malik:

Thank you. I will just say at the end, as a good practice, make sure you are spending at least 30 minutes doing something on the Salesforce org. There’s nothing which is better than doing something. People are watching YouTube, people are reading blogs, listening to podcast. But 30 minute of doing on Salesforce org is the most powerful experience.

Mike Gerholdt:

Absolutely. I mean, if you follow pro sports or any entertainment, all of those players practice for a reason. And it’s not just to be out on the field, but it’s actually the motion and going through it and keeping themselves sharp and the same holds true for us. Thanks so much for coming on the podcast, Amit.

Amit Malik:

Yes, thank you. Excellent. Thank you for the opportunity.

Mike Gerholdt:

Big thanks to Amit for joining us and sharing such a clear and thoughtful perspective on teaching AI in a way that actually works for admins, for users, hopefully for your organization. I really think his framework is something you’ll want to listen to, take notes, and probably come back to again, so. Hey, if you thought this episode got you thinking about how to introduce AI to your org, share it with somebody that you are working with so they can give it a listen. And until next time, we’ll see you in the cloud.

The post What’s the Best Way To Teach AI to Salesforce Users? appeared first on Salesforce Admins.