Hey guys, welcome back to another edition of AI in Action. Eric with Fast Slow Motion joined again by Zack Terry, Fast Slow Motion’s Director of AI. Zack, what’s up man, what’s new?

Eric, good to see you. A lot of updates to Agentforce to talk about, so looking forward to diving in.

Things seem to be moving at a bit of warp speed, which is not surprising. The AI universe in general, seems like daily there’s just acquisitions, development, new models. I mean, it’s a little tough to keep up with.

Yeah, it absolutely is. And it’s not just with Salesforce. It’s the entire ecosystem is moving so quickly. It is quite difficult to keep up with the latest updates. So a lot of news coming in all the time, but hey, that keeps you on your toes.

Yeah, I feel like if we were even doing this live by the time we got to the end of the episode, maybe some of the information’s already outdated. So, yeah, for sure. But let’s stay focused on Salesforce because I think obviously that’s our area of expertise. That’s something that we’ve invested heavily in. Talk about, you know, last time we talked was about a month ago, May, 2025 is June, 2025. Now over the last month, what is Salesforce kind of continued to roll out and integrate and enhance on the Agentforce platform?

yeah, we probably would miss a press release already.

Yeah, we’re seeing a lot of continued updates to Agentforce just in general and both in terms of new agents that are coming out, new prebuilt capabilities that Salesforce is offering as well as just enhancements to the platform itself. So the platform itself, we now have the ability to do things like send in context variables to allow the agent to derive context from a conversation. So for example, you can deploy an agent to an experience cloud site. Someone can be logged into that site.

and the agent can derive the context of, who is this person that’s logged in? Which gives you a lot of benefits. One, it allows them to keep a record of that conversation over time, but B, they don’t have to actually tell the agent their identity. They don’t have to provide an email address because they’re already authenticated. We already know who they are. So it makes it a much more seamless user experience for that particular use case. We also have some other things like some dynamic rich cards that are coming.

to the Agentforce Service agent interface. now we can do things like build carousels and image sliders and cards that can populate just a much more rich experience as opposed to solely text based back and forth. So that’s coming out with this new summer release that is already available for some orgs and for others will be available by the end of next week. So June 13th would be that sort of release window. And then yeah, we’re seeing a lot of updates to other agents as well.

SDR agent if you’ve been experimenting with that it’s been limited just to leads for well really since it was introduced but now it’s also gonna have support for contacts in person accounts so the implication there is if you’re an organization that doesn’t use leads in Salesforce or you want to outreach to people that are maybe a little bit farther in the sales cycle as opposed to right at the top of the funnel you know have the ability to expand who you’re able to automatically reach out to using that SDR agent

There’s some additional testing capabilities with that SDR agent as well to allow you to preview the emails that are getting generated and sent out to make sure that they’re adhering to your business policies and that the emails have the tone that you would like and things like that before you actually activate that agent. So those are just a few of the updates that are coming out. And of course, this is being updated every day. So we could hit on several other updates if you want to. But those are some of the biggest ones.

Yeah, I feel like it is almost on a daily basis. We’re learning more. We’re seeing more. Salesforce is continuing to publish a lot of educational content to teach people how to think about agents, how to explore use cases for agents. Let’s just, let’s rewind a bit and maybe focus there on, you know, kind of how Salesforce defines an agent. What are the attributes of an agent? If we’re thinking about, you know, enhancing this or integrating this in somebody’s Salesforce org, kind of what is that framework that we walk through there?

Sure, they have a framework called the five attributes of an agent and it’s very simple. It’s not complicated at all, but it really just helps distill your thinking to the five most important aspects of what you need to evaluate when you’re building an agent. Those five aspects are the role, the data, actions, guardrails, and the channel. And I’ll kind of dive into each one of those. The role is essentially the job that should be done. So what is this agent supposed to be doing? What is their job?

That could be as simple as a customer service agent to help answer questions. If it’s an SDR, it’s to reach out to leads and try to get them to book a meeting, right? Basically just determine, hey, what’s the purpose of this agent? And then the data piece, which is, okay, in order for them to do that job, what knowledge do they need access to? I’m using the word knowledge loosely. That doesn’t necessarily refer to knowledge articles in Salesforce, although it absolutely can, depending on the context, but that could be any data that you have in your CRM.

or any data that you have connected from data cloud from external systems or anything like that. And another update that was just brought in is you can now use external objects as data sources as well. So the capabilities for what data you can give the agent access to continues to expand. But the main thing is just to say, hey, if my job is going to be a customer service agent, what data do I need to give the agent access to? Probably things like contacts, accounts, cases.

maybe knowledge articles to be able to answer questions, but just thinking through, what is that agent need to do its job? And then it’s the actions. So actions are what the agent will actually be doing. So if that job is to help customers answer questions and the data we’re giving the agent is knowledge articles to be able to reference, let’s say FAQs or business process documentation, the action the agent needs to take is to actually go and retrieve information from that knowledge base.

And so that’s going to essentially be the answer questions with knowledge action, which is a standard action that’s pre-built in Salesforce that you can give agents access to. But this could also be an entire set of custom actions that can call flows that you’ve built, Apex that you’ve coded, sort of anything that you really want it to do. So there’s a whole list of standard capabilities as well as custom capabilities that you can build.

But ultimately it just comes down to, what is that agent going to need to be able to do to accomplish its job? And then sort of the opposite of that would be the guardrails, which is what should this agent absolutely not do? And that’s just as important because we don’t want to expand the capabilities of that agent beyond what we intend for that agent to be able to access. And so some of that is making sure that you’re restricting access to data that the agent doesn’t need access to, making sure that you don’t add

extraneous actions that are not needed. So if all that agent is supposed to do is answer questions, we don’t want to give that agent the ability to update a record in the CRM if that’s outside the scope of what that agent can do. So guardrails are a combination of restricting access to data, restricting access by permissions, and also explicitly providing instructions that can tell the agent, not do this, always do this, things like that. And then finally is channel.

And this is where is the agent going to be doing this work? And I think the easiest one to understand is a chat bot on a website. So in that case, the channel is your website or for Salesforce, it would be messaging for in-app and web, which is the chat component that you would place on a website. But it can also be SMS, Facebook messenger, WhatsApp, email, sort of any of these communication channels that you might have, you know, human to human. Think about that as agent to human.

or even agent to agent if you want to build a complex workflow with multiple agents kind of working behind the scenes. But ultimately it’s, hey, where are they going to be doing this work? So to recap, that’s the role. What job is the agent going to do? The data, what knowledge does it need to be able to access to do its job? The actions, what capabilities should that agent have? Guardrails, what should the agent not be able to do? And then you’ve got the channel, which is where is that agent going to be working?

And underscoring that entire framework, of course, is the trust layer. So, you know, the fact that, hey, CRM data can be a little sensitive, right? We don’t want it to just kind of be out there. So how does Salesforce accomplish that protection of data? And what’s that concept of a trust layer?

This is one of the more compelling features of Agentforce in my opinion, because I mean, if you’ve worked with LLMs outside of Salesforce, Chad GPT, Gemini, Claude, whatever that may be, and you’re doing that within a work context, you have to be careful what information you actually provide to these LLMs because unless you have some kind of agreement at an enterprise level, they’re going to use whatever you upload to continue to train those models, right? So there’s a lot of privacy concerns there.

So Salesforce built what they call the trust layer to address those concerns. And there’s a lot to it. I would say for anyone interested in diving into it, you can really explore the specifics of it by going to Salesforce’s website, AgentForce. There’s a whole sort of list of documentation available for the trust layer. But at a high level, it’s doing a few things. One is they have a zero retention policy with OpenAI. So OpenAI is what essentially does the reasoning for the Atlas reasoning engine.

and there’s a zero retention policy there. So that means whatever data is being sent to OpenAI, they are not allowed to retain any of that information, which is great on its own. Then on top of that, the trust layer is actually going to dynamically identify things like personally identifiable information that may be going out of Salesforce and into the LLM. So if in my prompt, I have information about a customer, email address, a phone number, social security number,

which probably shouldn’t be going in there anyway, maybe it’s credit card information, whatever that is, that information is going to be dynamically masked so that by the time it even gets to OpenAI, it’s not able to actually identify any of that information. It’ll go and take it and process the prompt and come back with a result. And then when that comes back, it gets demasked so that it makes sense to the end user, it makes sense to your CRM and the system. So those are some of the primary features I would say.

And really, think that’s one of the more compelling reasons why you would want to, let’s say, use Agentforce over just building a custom integration out to ChatGPT. It’s because you have the ability to mask that data, encrypt that data, keep that zero retention policy with OpenAI. So it gives you just a much, I think, firmer footing in terms of using AI at an enterprise level, because a lot of the security concerns

have been built for you so that you don’t have to worry about that or build that yourself.

Yeah, and we’ve kind of seen this where, at least with our clients, there’s almost been this hesitancy of, man, how much do we want to dip our toes into AI? How much do we want to explore this? The ones that have kind of taken that leap forward.

have really seen some results. We’ve got a lot of great use cases we could talk about. But for the CEOs out there, for the operators out there, the marketers out there, the success managers out there that are wanting to maybe learn more, what’s the best resource for them to dive into Agileforce?

Yeah, absolutely. Well, first of all, I would say if you have concerns, those are not unfounded, right? This is a very new technology. So I think everyone realistically has concerns. But I think the takeaway is maybe it’s a little less scary than you would think initially. And part of that is because we do have tools like Agentforce that do have some built-in trust. I would say if you’re interested in sort of dipping your toes into Agentforce specifically.

There’s a really easy way to get started through Salesforce foundations, which is a product that Salesforce offers for free to enterprise customers and above. And it gives you a bunch of features, things like access to marketing tools, sales tools, service tools. But as a part of it, you also get free access to a limited number of conversations or credits to use agent for. So there is a new pricing model with Agentforce, which is flex credits. And so now.

If you wanted to get on board with foundations, first of all, just talk to your AE. They can provision that for you really quickly. It’ll give you access to data cloud. That’ll give you access to Agentforce. but when you do that, you can have a conversation with your AE around, Hey, should I go with a thousand free conversations or should I go with a hundred thousand free flex credits? And the difference between the two is conversations flat billing, right? It’s $2 per conversation. So you get a thousand of those for free. You know, that’s the value that you’re getting with flex credits is a little bit different.

It’s not a flat rate per conversation. It’s actually variable based on how many actions that agent takes. So again, in that framework where we’re talking about the five attributes of an agent, when the agent actually goes to take action, it’s calling a flow, it’s calling some apex, it’s retrieving some information using a retriever. Every action that gets taken is billed at 20 credits, which is the equivalent of about 10 cents. So if you have quick interactions where people are not, you know, spending

minutes to hours chatting with an agent, you’re likely going to get a great result by using flex credits because you won’t be paying that flat $2 per conversation. It’ll be more incremental. I take two or three actions, you 20 or 30 cents. Whereas if you have interactions that are much more ongoing, long form, people are sitting there having long conversations with that agent, that $2 per conversation rate might be a better.

Yeah, and of course our team’s here to help them think through those things as well. So we’ll drop a link in how you can reach out to us in the comments of the video.

so Zack, talk to us about the use case framework that Salesforce has provided.

Absolutely, Salesforce has a pretty concise framework that you can summarize as the when I want so that framework and this is just to help identify use cases for Agentforce. So the idea is when that’s the moment in time that you want to trigger an agent’s involvement. So that could be just a conversation or it could be asking a specific question, things like that. I want is what do you want that agent to do the desired action for the agent to take given that triggering moment?

And then the so that would be the intended outcome or the business objective of that use case. So an example is, hey, when a customer asks an Agentforce agent a question, it could be, want that agent to dynamically retrieve that information from my knowledge base so that I can cut down on the amount of time that my human agent spin answering simple questions and they can focus more time on answering the higher level questions and the more important questions, right?

very rudimentary, very simple use case, but you can apply that sort of across the board or any use cases you’re considering. And really, I just think it helps hone in on this is exactly not only what I want the agent to do, but why I want it to do that. And I think it’s important to tie that to business goals as well. So if the goal is to reduce time for my human agents on basic questions, I would frame that up as I want to see a 40 % reduction in the amount of time.

that my agents are spending answering these simple questions so that they can spend more time focusing on key accounts or on more difficult questions.

Yeah, thinking through those key performance indicators so that you can kind of get to some idea of ROI in the implementation. Makes sense. Zack’s been fun as always, man. Looking forward to catching up with you next time and just kind of seeing how the rocket ship’s going between now and then. It’s definitely a lot to keep up with.

Absolutely.

yeah, I’m sure we’ll have

a whole new set of use cases and features to talk about next time.

Looking forward to it, my man.

Thanks, Eric.