Today on the Salesforce Admins Podcast, we talk to Gary Brandeleer, Senior Director of Product Management, Emerging Tech and Products at Salesforce.
Join us for a roundtable discussion of everything Copilot: what it can do, how you can customize it, and what you need to do to get your org ready to get the most out of it.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Gary Brandeleer and Josh Birk.
What is Einstein Copilot?
So there’s a big announcement at TDX today about a cool new AI tool for Salesforce called Copliot. We thought it might be nice to hear all about it from the PM in charge, so we brought Gary Brandeleer on the pod to find out all about it. We’re also joined by Josh Birk, who has spent a lot of working with Copilot.
Simply put, Copilot is an AI assistant that will help you get things done in Salesforce with natural language prompts. So you might ask it to give you a list of all your opportunities in the last month, or to summarize your most recent opportunity, and it will give you an answer in natural language. But we’re only scratching the surface of what it can do, and Gary was excited to tell us more about it.
Customizing Copilot to get more done
Salesforce has been working with AI for a long time, and you’ve probably seen it integrated into things like lead scoring and analytics. So what’s so special about Copilot?
For one thing, natural language processing makes everything much more user-friendly. You can chain multiple actions into one request. For example, “summarize the most recent case and write an email about it.” If you think about how many clicks that would take to do on your own, it’s easy to see the potential of Copilot for your users.
As for what you sort of actions you can request Copilot to do, there will be several options available out of the box. But because Gary knows how important Flows, Apex, and other customizations are to admins everywhere, you’ll be able leverage those skills to create your own custom actions, too. The possibilities are truly limitless.
How to get your org ready for Copilot
We’ve talked a fair amount on the pod about what you need to do to get ready for the AI tools coming to Salesforce. Data cleanup is more important than ever before.
For custom actions to work well, you need to make sure that you’ve updated the descriptions on all of your flows so that Copilot knows what it’s looking at. In general, you need to take a look at your labeling and organization practices for all of your data.
Finally, it’s important to remember that prompting AI is a skill that you need to practice. Both Josh and Gary recommend spending some time with a tool like ChatGPT seeing what kind of prompts work best. Try to get it to give you a recipe, or tell you a dad joke, and then see what kinds of questions get the results you’re looking for.
There’s a lot more in this episode about how Copilot works, so be sure to take a listen and subscribe so you’ll never miss out.
Podcast swag
Learn more
Admin Trailblazers Group
Social
Full Transcript
Mike Gerholdt:
This week on the Salesforce Admins Podcast, we are talking with Gary Brandeleer about Einstein Copilot. Now, it's March 7th if you're listening to this, the day this podcast drops, which I'm sure you are, you could be at TDX. Or not. So this is why I'm bringing this to you because we're talking about Copilot at Trailblazer DX and wanted to bring you a little bit of a conversation that myself and my fellow evangelist, Josh Birk had with Gary Brandeleer on some of the challenges and features of Copilot, the direction that they're going to go in terms of building it, some of the really cool capabilities of it. It's just a really fun discussion. I appreciate Josh jumping in, helping me out with this podcast. He really had an opportunity to get hands-on with Copilot at this point, so he really helped steer the conversation. I hope you find it intriguing. I did.
Now, of course, if you love what you are listening to, can you do me a favor and just make sure you're following the Salesforce Admins Podcast? So if you're listening to this episode and you like what you hear, listen to a couple more. Hit that follow button on iTunes or in Spotify or iHeartRadio. Because then every time a new episode comes out, it will drop right on your phone. But enough of me, let's get to the conversation we had with Gary and Josh about Copilot.
So Josh and Gary, welcome to the podcast.
Gary Brandeleer:
Thank you.
Josh Birk:
Thanks for having us, man.
Mike Gerholdt:
Good. Well, wanted to have a little bit of a round table discussion because Copilot is such a very cool product that we're launching actually today because today is the first day of TDX if you're listening to this when the podcast comes out. And I know you are because you downloaded it on your phone right away, just like I said in the intro to do. But we've got Gary on, and I brought Josh on, a familiar voice from those that listen to the Admin Podcast because Josh has actually been a little bit more hands on with Copilot than I have. So Gary, let's kick off with you. How did you get started at Salesforce and what do you do?
Gary Brandeleer:
So amazingly enough, I started on the Salesforce field service product as a solution engineer, and then I moved to the US. So obviously I'm from Belgium. I cannot get rid of this accent.
Mike Gerholdt:
I thought it was a southern accent. It sounds like Tennessee to me.
Gary Brandeleer:
Exactly. It's directly from there.
Mike Gerholdt:
Is it? Okay, the mountains.
Gary Brandeleer:
And so yeah, in San Francisco now it's been six years or a bit more. And so I worked as a product manager on Salesforce Field Service. Then I moved to the emerging tech team where we worked on blockchain and Web3 related aspects. And then of course when GPT happened, we got asked to work very fast on that technology. And so that's what I've been doing now for a couple of months, if not already a year.
Mike Gerholdt:
Great. And Josh, we know your history. You created Trailhead, you've done a lot of stuff. You've been hands-on with Copilot. So I'll let you kick off the discussion with Gary.
Josh Birk:
Yeah. Now, first of all, I'm shy about the whole created Trailhead thing.
Mike Gerholdt:
I'm not.
Josh Birk:
I know. A lot of people aren't. I had a lot to do with the prototype and getting involved in the first year. But Trailhead took a village for sure. But anyway, moving on from that. Gary, it's good to talk to you again.
Gary Brandeleer:
Good to talk to you.
Josh Birk:
Well, let's start at the very basics, and I'm going to ask you a very basic question, but I want to get an answer that's pointed to those nouns you just used like blockchain. Pretend I don't know anything about that. So slow walk me through what exactly is Copilot?
Gary Brandeleer:
So Copilot is really an AI assistant that's going to help you to do your tasks inside Salesforce. And I think that the easiest way to understand this is simply to tell you what you can ask it essentially. And so something you could ask to your Copilot is, "Can you give me list of the opportunities I have and what's the total amount?" And you use natural language and suddenly you have a Copilot AI assistant that is helping you at getting the answer. It's giving you the answer in the form of text or other forms. And that's pretty much it. It's really helping you to be more proactive using Salesforce data. And of course it can use also external data through data cloud and so on.
Josh Birk:
Right. I want to follow up a little bit on that conversational model, but kind of a historical question because as you mentioned, GPT happened, when did Salesforce first started working with technologies like this? And then what has the last six to eight months been like for you?
Gary Brandeleer:
I think it's a little bit tricky to answer because we started to work on AI very long ago. So GPT are just one of the many, many technologies out there that we can use under the banner of artificial intelligence. So I would say Salesforce started long ago on artificial intelligence in general. And you can see that, for example, from in Sales Cloud, you have lead scoring, we have also analytics that can use different algorithm and so on. So we had a lot of already of AI intelligence, I would say, in our product.
But then what happened really is that when we saw the power of what LLM could do, especially around analyzing texts, giving you answers in the natural language or using natural language, we were thinking of, okay, now how can we use that on Salesforce? And I think really we started a bit earlier than when it became very public that LLM we are going to change everything. If you look at our Salesforce AI research team, they have been working on LLMs for quite a while and they had already a lot of patents and white paper about it.
But I think it's accelerated once the public have seen how much value could LLM provide. And so that accelerated, I would say, starting January of last year. And since then it's been very intense, I would say. And the reason why is first of all, shipping a product very fast is not easy. Shipping an AI product very fast is even harder. And the story is even getting harder, as you look at the AI space right now, it's evolving like crazy. Every week I have something that is blowing my mind. I'm reading an article and I'm like, "Wow. That is feasible now? That's mind blowing." So keeping the rhythm, keeping yourself informed about what's feasible and then making sure that we can deliver as much value as possible to our customers using the latest and greatest is really, really a big challenge. But it has been super fun so far.
Josh Birk:
Yeah, it is. And I can sympathize with you because I have made statements to the public about AI, which were then disproven about three weeks later. It can be so hard to say, "This is exactly what the feature set it's going to be like." It has been a fascinating journey kind of interacting with them. Pretend I'm somebody who has heard about ChatGPT, is kind of familiar with the idea of a bot, but probably kind of in a more traditional sense of a bot. And when I hear we can do things like ask Copilot for the last three leads that I worked on or something like that, I might also think, well, that sounds like a dashboard or a report or a list view or aspects of the UI we're already familiar with. What novelty, what innovation is Copilot bringing to the user interface that's giving this power? What is the LLM and IN adding to it?
Gary Brandeleer:
I think there is two answers to that topic. One is if you look at LLMs, generally speaking, they're very good at managing text, summarizing, generating content, and so on. The second part, and that's more Copilot related, is that Copilot is able to chain what we call Copilot actions, which is really basically stuff it can do. I would say another way to position this is to say that Copilot will have a brain, and we call that the planner. That's the technical term so far.
But basically that brain will select different actions based on what you are asking the Copilot. And so what would happen is that you could say, yeah, if I want to find the latest scales, I can do that by going on a list view, for example, so on and indeed you could ask Copilot, "Hey, find the latest case," it will find it for you. And then you could ask this follow-up question of, "Hey, summarize it." And so it's going to summarize it for you.
What's much better is to say, "Hey, summarize the latest case." And in that scenario, the Copilot will combine different actions. It'll find dynamically which actions it need to combine to answer the requests. And so then you unlock a lot of value and a lot of different use cases simply because now the Copilot is able to chain the different actions together and give you an output that will be relevant for your request. And so I think more and more as it evolved and as we get user feedback, you will see that people will say, "Oh, wait a minute, I can do that with clicks, but now I could have done this with 10 clicks or I can just ask one single sentence to Copilot and the 10 clicks [inaudible 00:09:52] for me."
Josh Birk:
Right. Yeah. And I want to dig into actions a little bit more, but let me give you a theoretical based on what you just said. With Copilot, I could ask one question, which is, "Provide the three most recent open case leads I have." And then I could say, "Summarize those based on the amount of active cases that they have." And then I could say, "Okay. For that lead that has the most active cases, could you give me an email version of the summary that I could send to my manager?" And that's three prompts and I would get that actionable piece of content, right?
Gary Brandeleer:
That's exactly correct. And so you could even go as far as, I would simplify a little bit, but I would say you could go as far as saying, "Hey, summarize this lead or summarize this case and write an email about it." And at that stage you will not see the summary first. You'll basically get as an output directly the email, even though the Copilot has executed two or three actions to get to that output.
Josh Birk:
Got it. Now, first of all, I absolutely love the definition of an action as stuff that it can do because I feel like that boils it down so wonderfully. But let's bring that up another level. What is powering an action? What's the technology behind it and what is Salesforce providing out of the box with that?
Gary Brandeleer:
That's extremely important to flag it. And there is differences of course between what we ship. So as Salesforce, we will ship standard Copilot action or Copilot standard actions. And an example of that would be query CRM, draft and revise emails, summarize records. And these are really standard actions that are coming with Copilot. But then what we know is that many, many of our customers still love to configure, customize Salesforce. We also know that a lot of Salesforce admin tailor flows, apex and so on. So we are like, "Okay, wait a minute, because we need to be sure that the Copilot can be configurable, so how can we do that?"
And so we introduced the concept of Copilot custom action, and you can then create these custom actions and select either invocable actions, either auto launch through so far, either prompt template. And that's unlocking a lot of value because you can then cover a lot of use cases. On top of this, I would say, it's introducing one aspect that people will have to learn, which is you might already have an auto launch tool, or you might have already an invocable action that you are thinking of, "Hey, wait a minute, if I set up that in Copilot, this is going to be a super cool use case that Copilot will be able to do for my user."
But what is very important is to describe very well what the action is doing. And that's, I think, a new pattern that is popping up is that we are not very good at describing. Every time I'm creating a flow, I'm like, "Hey, I'm creating the flow. I'm naming it." And then the description, I'm just skipping it because I was like, "Nobody's going to read the description of the flow anyway." And that's just the way I'm doing it. Maybe some people out there are more disciplined than me.
But now it's extremely important because there is actually something that's going to read that description and it's going to be the LLM. So the LLM will only know if it needs to pick up this action or not based on the description you have for that custom action. And so to put that a bit more in context, it would be, I have a flow that is allowing me to, let's say, create a case. Then I would've to create the custom action, select that flow, and then I would've to describe, okay, this action is allowing you to create a case which is a Salesforce object used in the context of call center. And now the LLM will be, "Okay, if there is a call center agent asking me about creating a case, I will use this action. That's something I can do that has been well described to me."
Josh Birk:
Yeah. First of all, I love that acknowledgement of human behavior. My father is a surgeon and he got flagged by an administrator because none of his notes had his signature on it. And his response was, "They're my notes. They're for me." I know I wrote them, they're my notes. Why are you bugging me about my signature? And I think a lot of people think, well, if I put the label in and it's human friendly enough and most of the people are going to be seeing it are the people who are using it, the description is just sort of an add-on. But first of all, I want to hang a lantern on something. We've said LLM as an abbreviation a few times. It's large language model.
Gary Brandeleer:
That's correct.
Josh Birk:
Which is basically... I've heard it described, I mean, I kind of have it in my head as it's the very specific kind of data that the AI is looking at, right?
Gary Brandeleer:
I would put it another way, I like to simplify stuff a lot. And this is an oversimplification.
Josh Birk:
Okay.
Gary Brandeleer:
And it's probably a very, very, very simplified version. But for me it's more like what the LLM can do, at least in our context right now. And for math, I was going to use a calculator and I'm going to do one plus one equals two, which I can prove now that I'm very good at math. But the second thing here was for text, I could not really use anything. And now I have these LLMs that are able to ingest a lot of texts, generate a lot of content and so on. And that's what I think is important. How it's built, we can go very technical. But basically neural network and so on and so on. So I mean, we could create a full podcast just on that if you want, but it's more important to know what the use of it and the use of it is. Now everything you are doing with text can be much more automated or I would say much more augmented in some ways.
Josh Birk:
It's very good that this is a podcast format because it means we don't even have the idea of adding in the formulas that make this thing work that made my eyes bleed the first time I saw them. So I think that's an excellent description. And I think it also, thinking as terms of a calculator of AI and hallucination, some of the ways we phrase these things kind of makes it sound like they're almost a thinking sort of thing, but they're really more of a calculating kind of thing. And I wanted to say that to kind of emphasize your importance on, well, why do you need a really well fleshed out description? Because you're talking to a calculator that needs as much, it needs all of the numbers you would put into the calculator in the first place.
Gary Brandeleer:
That's correct. And on top of that, I would say the basics behind is that basically when you ask a request to a large language model or LLM, each word is basically a statistic. Meaning it's going to think of, "Hey, I'm going to speak about the cat, and the two words are going to come based on statistics." So it looks to us like it's complete magic and you have nearly someone speaking to you. At the end of the day, it's just statistic behind the scene that are popping up the right word and that's important to keep in mind, essentially.
Josh Birk:
Got it. So I can leverage my existing flow skills. I can even, to a certain extent... When we talk about invocable flows and headless flows, are you seeing flows that are kind of like, if I make my descriptions really good, they can pretty much serve as actions or what's your recommendation there that I kind of take an existing flow tailored for Copilot and then make sure the descriptions are a nice hefty paragraph?
Gary Brandeleer:
My recommendation there would be think really about the different strengths of Copilot and of LLMs in general. So if you have an existing flow that is already working and you would think of this would be worked for the Copilot to be able to call it whenever a user is requesting. Then, indeed, the only thing you need to do really is to create your custom action, describe it very well, describe the input, describe the output, and that's pretty much it.
Now that I think is a little bit too much of a dream, it'll not work as easy as that. What I mean by this is that now you might have, let's say 10, 15 actions that you have assigned to your Copilot. It might be that you add one more action and now your description was pretty bad. So this time for some reason, it was late on a Friday, you wanted to close your computer fast and you created very bad description.
If your description is very bad, it might be that suddenly, even though your first actions were all working fine, that now when you ask some requests to the Copilot, the Copilot will always select that action with the bad description simply because it's a description that is so bad that it's kind of overrule all the others. So I think what's important really is that testing of utterances, which is another word for simply a request from users. And so every time you create a new action is think about the value. Is there a real value to add that in a Copilot, yes or no? Is that using the power of LLMs? So content generation, summarization. In some ways text analysis or I would go for data analysis even though it's not exactly right. But think about that, the strength of the LLMs themselves, and then think about, wait, is that going to be overlapping with other actions I have already?
And the last piece is that going to be used and chained with other actions? And that I think is a very important point is back to that question of, if you ask it to summarize the data scales, will it use different actions, chain them? And the answer is yes. If you create a new custom action, is there something super cool you could do by thinking of, hey, wait a minute now I could create this flow. It's going to retrieve some data, for example, and then it's going to use the existing summarize action to summarize something, a customer record or something like this. So really that chaining of action can be also a good reason for why or why not taking an existing flow and creating that as a custom action or using that as a custom action.
Josh Birk:
Got it. Now Copilot's going to arrive out of the box. It's going to have some standard actions. We've talked about actions. What else does an admin need to know day one if Copilot's enabled? What other kind of setup tips and steps should they know they should be working towards to get it up and running in an org?
Gary Brandeleer:
A couple of things. One, there is one Copilot for your employee in that Salesforce org. So generally speaking, you will have one single Copilot so far for all your employees in your Salesforce org. Second, you can, of course, control the access and the access to the Copilot is done via permission sets. The data access, you don't have to change much. It will respect, of course, the system we have for years now. Which basically if you have a profile and you don't have CRED access on some objects or whatever else, we will not of course certainly overrule that because you are asking to Copilot to update an opportunity where you don't have access, for example.
Josh Birk:
Gotcha.
Gary Brandeleer:
So the beauty there is really activate the Copilot, give the perm sets to the right users so that they can use Copilot. And that's pretty much it. And that's using the out of the box standard actions. After that, once you see and get feedback from your users, I think then think about what kind of custom actions you could add, what other use case you could cover, but don't go too fast. Like, little by little. New technology, people need to get used to it. I think there is a big, big part of expectation as well. People expect a lot from AI. The reality is that at first it'll do a few things pretty good for you and your users will tell you, "Well, I keep asking to the Copilot, 'What is the weather in San Francisco?' And it doesn't reply." And I'm like, "Okay, that might be a custom action. I'm not sure you should really do that in Salesforce." But that, I think, is important is look at what kind of requests your users are requesting, and from there you will find cool use case that you can customize and configure.
Josh Birk:
Now-
Mike Gerholdt:
Josh, I have a little bit of a specific follow-up question before you go.
Josh Birk:
Oh, go go. Yeah.
Mike Gerholdt:
Because Gary, a lot of the podcasts that we've been doing and a lot of articles around getting ready for AI have really focused on data cleanup, and we actually had quite a few customers at Dreamforce talk about data cleanup in terms of prepping for AI, which is a best practice for a Salesforce admin anyway. It should be in their essential habits. What I'm also hearing, and this makes my heart sing, is that you now have a reason to have a systematic plan to go through and either describe or clean up your descriptions on key things like flows, custom objects. I'm hearing this is a priority, right?
Gary Brandeleer:
I think it is, at least for the flows you want the Copilot to use once you set up your custom action. But yeah, it's becoming much more important. And data quality has always been super important. I think what's new is, I would nearly say metadata quality is also super important now. So not only if you look at an opportunity is the opportunity description well described is the amount at the right level and things like this. So the data quality itself, but then did you describe really well what that flow on the opportunity object was doing? Keeping in mind that if you, for example, go in one of your existing flow and you add descriptions there or edit the descriptions there, once you will select that flow as a Copilot custom action, we will of course copy over all the description from your flow. So I would say best practices would be, hey, get the source cleaned up so that whenever you are using it in Copilot, you don't even have to change the description. You just take the one that were set in your flow essentially.
Mike Gerholdt:
Gotcha.
Josh Birk:
How does things like duplicate fields and records and other aspects of unclean data, can that impact how well... Is it on the same level as what we usually see with that kind of turn? Or can Copilot be even more affected by that kind of thing?
Gary Brandeleer:
I would take the example of summarization. If you try to summarize an opportunity where many of the fields are completely blank, nothing has been really changed and there is not even a good naming convention of your opportunity or whatever else. You can do whatever you want, but there will be no magic. The summary will be looking pretty bad simply because it's trying to summarize data that is pretty bad. So that's one. When it comes to identifying records, I would think more about it as a search mechanism where, generally speaking, if you have opportunity names that are a bit weird, as long as you search for the right name, the Copilot will be able to find them the same way you would be able to search them.
But of course, if you have, for example, let's go a bit more to a practical example. If you say, okay, I have a deal called Acme, but now you have three deals actually called Acme. If you search for it, you'll find three records. If you ask a Copilot to update Acme and you have three opportunities named exactly the same way, then at that stage, Copilot will try to ask you more information to find the right record. So it'll probably ask you, "Which record are you talking about? Because now there is three of them that have exactly the same name, so which one do you really want me to update?" So that's what you can expect. So I think a search mechanism would be more like we show you the result and then you figure it out. With the Copilot, it's more like, hey, we get some results, and then we ask you more questions to know that we are acting on the right record.
Josh Birk:
Which speaks right back to the power of that conversational model. Which is really hard, I feel, to know until you experience it. One final question for me, where can people learn more both at TDX since the magic of time travel we'll be talking to people sharing TDX when this launches. But also beyond TDX, what are some of the best resources?
Gary Brandeleer:
So we are working on a couple of trails that will be published by TDX. We are also, of course, updating our websites with a lot of data there. And then I would say release notes are still your best friends. We will make sure that they're as clear as possible. But I think these are a couple of places where to find this. And then we have, I think, multiple communities as well that are set up for AI and we should reuse just these to have our discussions about Copilot. So that's what we'll have by that time.
Josh Birk:
Nice.
Mike Gerholdt:
I have a feeling this is going to be on everybody's lips when they're at TDX and all of these groups thereafter.
Gary Brandeleer:
I have a feeling as well that this will be very much discussed.
Josh Birk:
Yeah. And the hand things right back off to you, Mike, I have a feelings will probably not be your last AI centric episode.
Mike Gerholdt:
No, there is no such thing. We're just getting started. I am excited to do an entire presentation on best practices for updating your description fields, though.
Gary Brandeleer:
Yep.
Mike Gerholdt:
I know it sounds insanely boring talking about how white rice is, but man, let me tell you, it sounds like that is going to be key.
Gary Brandeleer:
It's going to be key. I can tell you [inaudible 00:29:20].
Mike Gerholdt:
It's all the little thousand paper cuts.
Gary Brandeleer:
Yes, yes. And we even ourself, we struggle with it honestly by building this product. We are like, every time we build a standard action, we also need to describe it correctly. And there was a lot of back and forth on how to name it, how to describe it, and so on. So that has been quite a challenge, I would say.
Mike Gerholdt:
Well, to echo back to an episode that I published in February with Marissa Scalercio, who is a customer that was on the podcast talking about her pilot use of Prompt Builder. One thing I asked her, and I think it reigns true for this episode, regardless of what we're talking about in terms of AI for Salesforce, her advice was, "I can't tell you how much I wish my past self would tell my future self to start using AI now. Just any AI. Because asking it questions, doing things."
Josh, you probably follow me on Instagram. I'm spitting out a whole bunch of AI generated images just because I find it's interesting what you ask AI and what it comes back and then getting better at learning how to ask questions and learning how to, not train the model, but think through, oh, this is literally what I asked for, but in my head it was something different. And I will echo back to her advice because I think it reigns true. Even getting ready for Copilot, you're going to have to get better at asking questions and get better at cleaning your data up.
Gary Brandeleer:
That's correct.
Josh Birk:
And there's no reason this magic of a conversational UI, which is so hard to describe in any way other than just saying, "Do it." There's no reason to wait. You don't have to wait for Copilot access or anything like that. Go to Bard and ask it to give you dad jokes. Jump in and just get that conversation flowing just to feel the fact that, oh, like Gary was saying, I thought it'd be three prompts. It actually could be two if you know how to ask the right questions.
Gary Brandeleer:
That's very involved, I would say-
Mike Gerholdt:
Gary, thanks so much for coming on the podcast, and I'm sure people... I have a feeling you're going to be a little bit popular at Trailblazer DX.
Gary Brandeleer:
It might be. I will be there hopefully being able to answer as many questions as possible from the customers. And if not, we'll try to find a way to get maybe a Copilot answering the questions. So we'll see.
Mike Gerholdt:
There you go. Oh, look at that.
Gary Brandeleer:
We will see.
Mike Gerholdt:
Very meta answer of you.
Josh Birk:
One final question, Gary. Are you a coffee person?
Gary Brandeleer:
Actually, so amazingly enough, yes, but I stopped because I was not sleeping well because I was drinking too much coffee. So now I'm a tea person, which is much more boring.
Mike Gerholdt:
That's the best part of coffee.
Gary Brandeleer:
Yeah, kind of.
Mike Gerholdt:
You quit because of the best part.
Gary Brandeleer:
Kind of in the morning, yes. But in the evening when you're in your bed and trying to get asleep and you're still thinking about all the stuff you could do and so on, and you can't just go to sleep, it's just a little annoying. So after a while I was like, "Okay, let's switch to tea." But I must say I'm really missing a good coffee cup, especially a bold espresso. That would be the best. But for now, this month is coffee free for me.
Mike Gerholdt:
So it was a fun conversation. Boy, there's a lot to pick up. A lot of really cool features coming, I feel, in the next few years with all of the AI possibilities and some of the stuff that's going on with the ability to automate things and ask conversations, and talk with our data. Isn't that something we've been talking about for a while?
So if you enjoyed this episode at the beginning, I asked you to hit follow. Hey, maybe share it with somebody. You got a fellow person on your team that's looking to expand their admin skills or learn more about Einstein Copilot or the plethora of information that we cover on this podcast. All you got to do is just really tap the three dots and click share episode. You can post it social, you can text it to a friend. I appreciate you doing that. And of course, if you've got more great resources, your one stop for everything is admin.salesforce.com, including a transcript of the show. So that way, if there's a part of it that doesn't make sense, you can go through, read the transcript and get some information that way.
But be sure to join in the Trailblazer Group community discussion. A lot of great questions there and people sharing the podcast, which I appreciate. Also, if you have feedback, hey, I'm on Twitter and Threads and TikTok, and I think I'm on everything at this point. But send me your feedback. I'd love to hear it. I'd love to know what you think. I do enjoy reading all of the comments and hearing about the podcast because it's something I enjoy creating for you. So until next week, we'll see you in the cloud.
