Loading...
technologybusinesscareers
How Do I Understand a Complex Salesforce Formula Quickly?
About this Episode
Today on the Salesforce Admins Podcast, we talk to Gaurang Mathur, Senior Product Manager on the AI App Development team at Salesforce.
Join us as we chat about how Setup with Agentforce is changing the way admins create, modify, and understand formulas.
You should subscribe for the full episode, but here are a few takeaways from our conversation with Gaurang Mathur.
The massive scale of Salesforce Formulas
If you're like me, you've spent more time than you'd like combing through your formulas for missing parentheses and commas. That's where this week's guest, Gaurang Mathur, comes in. As a Senior Product Manger on the AI App Development team at Salesforce, he's trying to make formulas a little easier for admins with Setup with Agentforce.
According to Gaurang, there are 5 billion formulas executed on Salesforce each month. They're created, managed, and modified by 7 million admins and developers. The scale of formulas is truly massive, and they're a fundamental building block of Salesforce. However, that means they're difficult to work with, and that's where the Setup with Agentforce beta comes in.
How Setup with Agentforce simplifies formulas
We know that AI is really good at summarizing, debugging, and processing large amounts of information very quickly. Setup with Agentforce lets admins harness those agentic powers to make formulas easier to work with. You can diagnose problems, change an automation, or add something new.
Gaurang sees Setup with Agentforce as a game-changer for admins working with inherited orgs. If something isn't working or you need to make changes, you're usually spending hours wading through formulas to sort out what's essential and what's just buried tech debt.
With the beta, you'll be able to get a helping hand from an AI agent that can validate, describe, and even fix your formulas. Instead of having to figure out what does what, you can focus on making decisions and building something that works faster than ever before.
What's next for AI in Salesforce
As far as what's coming next, Gaurang thinks that AI will continue to amplify what we can build in Salesforce. As he points out, we've actually been using AI for almost a decade in things like navigation apps, recommendations, and even autocorrect. There's no reason to be intimidated.
Looking forward, his team is looking at ways to use AI to collect feedback and make adjustments that make it even easier to work with and more powerful for users.
Make sure to listen to the full episode for more from Gaurang about formulas and Setup with Agentforce. And don't forget to subscribe to the Salesforce Admins Podcast so you never miss an episode.
Podcast swag
Learn more
-
Salesforce Admins Podcast Episode: Setup with Agentforce Makes Salesforce Admin Tasks Easier
-
Salesforce Admins Podcast Episode: How Agentforce Vibes Is Changing How Salesforce Admins Build Apps
-
Salesforce Admins Podcast Episode: Exploring Agentforce Vibes Through Real-World Admin Use Cases
-
Trailblazer Community Group: Setup With Agentforce (Beta)
-
Salesforce Admins Blog: Introducing Setup with Agentforce (Beta): Your New Admin Superpower
-
Dreamforce '25 on Salesforce+: The Future of Setup Powered by Agentforce
-
Dreamforce '25 on Salesforce+: Setup Now with Agentforce: Simplifying the Complex
-
Dreamforce '25 on Salesforce+: Setup with Agentforce: New Superpower for Salesforce Admins
Admin Trailblazers Group
Social
Full show transcript
Mike:
Welcome to The Salesforce Admins Podcast. Today on the podcast, I'm joined by Gaurang Mathur, product manager on the AI App Dev team at Salesforce, and we're going to talk about something every admin has wrestled with at some point, which is formulas. And we're also going to hit on how setup with Agentforce is changing the way we create, modify, and even understand them. Hooray. I can always use some formula help.
So Gaurang is going to share with us how agents can validate, fix, and describe formulas right inside setup, and why that matters, especially if you've inherited an org or you need to do something super fast. We also touch on Agentforce Vibes, which you heard in last week's episode with Tiaan and some developer workflows and kind of what it means to build smarter with tools using AI. So if you're like me and you've ever stared at missing parentheses and wondered why your formula won't save, this episode's for you. And with that, let's get Gaurang on the podcast. So, Gaurang, welcome to the podcast.
Gaurang Mathur:
Hey, Mike. Thank you. Thanks for having me.
Mike:
Absolutely. Well, after we talked with Cheryl, it was important that I get a lot of the people that she works with on the podcast, because you guys are doing some really exciting things around set up with Agentforce, I think I said that correctly, and just some of the things with agents. So let's get started with tell everybody what you do at Salesforce and how you got there.
Gaurang Mathur:
Yeah, sure. Thanks. So my journey has been pretty interesting to where I am today. I started off as a PM for the caching group, which is again, a platform product. Platform Cache is something that's widely used, and that queue is helping a lot of ISVs to accelerate and perform, make their applications better perform.
And over the past five years, I've also managed products in the Experience Cloud Search space, and now I'm currently managing two Agentic products, which is again, hardcore platform products meant for our Salesforce admins and developers, the superheroes in the whole ecosystem of Salesforce.
Mike:
Yeah.
Gaurang Mathur:
And as part of that, as you mentioned, I closely work with Cheryl on the Agentforce setup, or setup with Agentforce, and over there, I am managing the space for formulas. I think that's where I'm going to talk mostly about in today's podcast. And the other product that I'm managing is Agentforce Vibes, which is again, zero to one product that we have all built over the last few years. There, I'm creating an MCP tool, which is basically making that agent much more smarter and getting a lot of ecosystem teams within Salesforce to contribute that, and then make it better in generating whatever metadata that it generates. So yeah, that's my portfolio here in Salesforce.
We are part of this whole group called as AIF Dev. That's where we've got Tiaan, I think one of the other guests in the previous podcast, Cheryl, and Manish. These are the folks that I'm working with, and excited to be in this whole space in the current year.
Mike:
Yeah, wow. Two products. So you sleep only a couple hours a night, I'm guessing?
Gaurang Mathur:
Yeah. I mean, these days, agents have been keeping us alive. I mean ... And so yeah, I'm just, as I sometimes joke with my family, I'm making agents smarter. That's what we work on, so yeah.
Mike:
Right, right. Well, let's talk about formulas because I think I go back to, boy, 10, 15 years ago, a staple in admin track was always a best practices on formulas and validation rules. And I always felt like formulas as an admin was as close as some people felt to getting to code, because we could carry over a lot of the stuff that we'd learned in different spreadsheets into formulas. So tell me a little bit about what you're doing with formulas and setup and how that works.
Gaurang Mathur:
Yeah, thanks. I mean, I think you touched on a very fundamental point. And before I get into the actuals of what we are trying to do, but I'd like to double click on what you just mentioned, Mike, which is formulas have been a very legacy product within Salesforce. They have been integral in the whole platform since, I mean, a very long time, and I'm quite excited to carry that legendary thing. I'll just share some numbers with you in terms of how big and the scale of impact that formula has in the whole ecosystem.
So almost on a monthly basis, there are five billion formulas that get executed across Salesforce orgs, and seven million admins and developers manage, create new formulas, or modify formulas every month. So that scale is fairly huge, and these formula engines and formulas are supported across almost 40 odd Salesforce services, which includes flows, business rule engines, SLAs, fields, validation rules and whatnot. So like I said, the scale is fairly huge, and we've seen admins using it, developers using it, building formulas quite often, but there has been a constant ask and problems that we've seen that admins figure out as is creating them. Somebody is coming up in a new org, and then just trying to modify those formulas. So those scenarios have been quite error-prone and time-consuming processes, and sometimes challenging for admins.
So with agents coming in, we asked admins, "How do you see agents helping you, and what is that agents should do for you?" And among the top 10 asks, managing formulas was right number three and four. So that led us all to make a dedicated decision and team that we should be improving and working towards creating a better formula management experience for our admins with the set up with Agentforce that's coming in. And we did a few releases. I mean, it was one of the few things that we did for formulas, I mean, almost year back.
And now with the beta coming in for setup with Agentforce, it's right up front and center. So back to your question then, "What exactly we are doing?" So we are, as part of this beta that's coming up in TDX, we are giving options to create a formula through a set up with Agentforce agent, modify it, fix it, and let it also describe what this formula is. So yeah.
Mike:
You end very casually, but that's kind of really big. I know, if you were to look back in my search history pre-AI, 90% of it as an admin was, "Why doesn't this formula work?" And I think when AI first came out, that was one of the ... I remember I was at a user group, and that was one of the prominent use cases for using an AI, was to check your formulas, because formulas in Salesforce are much like they are in various spreadsheet programs, so you can kind of understand it. And I remember thinking to myself, "Oh, man, the hours this would've saved me if I could have just copied and pasted a formula and had an AI read it and be like, 'Oh, you forgot a parens, or you missed a comma again,'" you know?
Gaurang Mathur:
Correct. Exactly. I mean, and those mistakes makes us human, and this is where agents can help and chime in to just call out, "Hey, this is the error, this is the fix," and that's exactly the first principles that we are considering as we build these agents, as we are coming up with this whole agent experience for our admins. So let me just share you one example for that, I mean, what we've just recently demoed to some of the customers and which was very exciting overall. So imagine you are in a company that's managing volunteer experiences and other things.
And out there, you have a formula, which is helping you calculate the number of attendees that are there, the volunteers that are going to be participating in that event. And you have all that formula baked in and all that, but suddenly, the admin who was walking on that leaves, or there is change, and now you are on your own and you just have to just kind of get things going, because your event is coming in, or there is some business requirement that comes in. So it gets difficult for anybody to just jump in and get going right off the bat. In this case, we are coming up with modifying a formula, wherein when you see this formula there and you open the agent through the setup page, the formula page, out there, the first thing that the agent does is it's just simply validating the formula and helping you understand what this formula does. Once you get that, now it basically helps you understand and comprehend, "Oh, this is what this formula is, these are the conditions, this is what it'll do."
Now, when you have to make some modifications, you can now just simply prompt it, and then the agent does the rest of the job for you in a very quick and easy way. So this is something that's really profound, and while we were sharing it with our early users, that was a welcome, and I'm hoping the community has to see, and they'll listen, try it out, and then figure out what it does.
Mike:
Yeah, I mean, thinking through, there's two parts to really what you described, which was it can do it, but then, the second part was it can describe, and that's just such a big part of everything that we do. Plus I always feel like, at least for math and formulas felt like math, if I could see a finished product, then I was more apt to kind of understand and work through it as opposed to kind of come to that solution on my own. But you mentioned Agentforce Vibes, and I think that's also, I don't know if you could say companion product, but it's another way to develop on the platform. How have you kind of found some crossover in what you're working on for Agentforce for setup with Agentforce Vibes?
Gaurang Mathur:
Oh, so you can call them as different facades for different personas and end users. So let me help you with what we are trying to do with Agentforce Vibes. Agentforce Vibes is quite focused on from our developer friends in the Salesforce ecosystem as it's a plug-in that you download in Visual Studio Code, and then you can start vibing and create new schemas, the metadata that you want to, and everything and anything that's happens within Salesforce, whereas setup with Agentforce is more focused from the setup UI that you had, and how do you manage them, I mean and managing formulas, permission sets, and different other things that we are going to. So it's a bit of a persona difference that we've had here, but we're trying to have almost the same engine behind the hood, which is what one of my teams is working on, and surface it in different facades, like I said, or the agents so to say.
Mike:
Yeah. Now, that's kind of how Tiaan explained it too. I did a podcast with Daryl Moon, who was actually super interested in Agentforce Vibes just because it really kind of jump-starts you in terms of app creation and some of the tediousness that can be started, so you can really focus on some of that really fun stuff that you like to do, maybe like formulas, I don't know.
Gaurang Mathur:
Yeah. I mean, a lot of it. I mean, so considering in this day and age, wherein we've got different LLMs and frontier models that are trying to do, this is something that maintains the Salesforce trust and the ecosystem intact. It's built on those foundations, so yeah. I mean, I'm excited to just, as I said, just you can prompt it, and then make some utterances, and then play it out with some apps that gets generated by the agent there with Agentforce Vibes.
Mike:
I know you said you worked in search for a little bit. Has any of the things that you've had to learn about search carried over to what you've had to learn about agents or vice versa, or are they just two different worlds?
Gaurang Mathur:
Yeah. That's an interesting one, and thanks for asking it. So it goes back when we started off, or when the agents were just getting launched, I think a couple of years back. It seems eternity away, but yeah, that's how the Agentic universe works, right? We were given a task, I mean, we were just basically figuring out in terms of what best we could do and leverage the agents that we have in place, and create an optimal search experience for the end users, and basically enable our admins do that and leverage the AI power that we were getting in.
So what we did was we were doing something that you even find today in Google search, when you put in a search query, you get a snippet at the top itself, which basically summarizes or creates an answer for a query. Ultimately, the answer is basically based on the search results that are floated, and then you get that summary. So that was the first time that I learned about RAGs, and RAG is Regressive Automated Generation, and all the AI technologies that are there. So I use those, and now I'm kind of applying them again in terms of creating the vibes, metadata experts that we are doing, the MCPs that I talked about. Again, we're trying to apply the same principles in terms of there is an utterances there, "What do you best, you should be focusing on, such that the metadata that it's generated is relevant, is relevant to the prompt, and it's of best quality?"
Mike:
Yeah.
Gaurang Mathur:
So that's where that crossover happened between those two text, but yeah, there's a lot since that time has evolved overall in the ecosystem and agents that we're developing.
Mike:
Okay. So listening to your answer, I came up with another question. I come up with lots of questions.
Gaurang Mathur:
Yeah. No, excited to have that.
Mike:
Do you think, based on the progression of where technology was going, because I heard in your answer about search that there was some form of RAG involved, was the invention of AI just inevitable?
Gaurang Mathur:
I mean, so that's a very fundamental. It has been there for a while. I think we all knowingly, unknowingly have been using AI in some form and factor, I think more than a decade or two. I mean, simple considering with Google Maps that we had or the YouTube recommendations that we had, in some form and factor, knowingly, unknowingly, we all had been influenced by AI for a while now. But in current day and age, AI has taken a form, which is much more tangible and which is much more right in the face with the GPT models and the LLMs that we have.
I think this is where it is picked up as a mainstream for one and all, and yeah. I mean, with that, there are multiple applications that have come into, which is as we as a platform, teams are focusing on, or helping in terms of coding and generation of different codes, and configuration and all that happens. AI applications, again, coming into, I see the latest LLM models generating it, and then helping in terms of the legal services, and helping in terms of the healthcare services, helping in terms of the technology services, and where else there have been multiple applications. So you can just vibe, you can just ask the agents, and then you get appropriate collaterals developed and things that can be done.
Mike:
Yeah. No, for sure. I mean, you're essentially creating the future. Do you envision getting to a point with AI where ... I mean, this feels very meta, but where an agent is almost helping you create a better Agentforce Vibes or a better setup with Agentforce?
Gaurang Mathur:
Yeah. I mean, so we are, in a way, trying to do that. I think with Agentforce itself, we have this learning curves wherein we are trying to feed the system about the feedback that we receive, and then based on that, putting it into agents to just improve that. So yeah, that's something that's very nascent. I mean, it's not mainstream right now.
All humans are involved in that, but I think that's inevitable. It's bound to happen, and then I see a lot of front-end model company's doing that today.
Mike:
Yeah. It's crazy to think that we have a tool that could help make a better tool. Growing up in the generation I did, where I remember having a cordless phone in the house was a big leap of technology.
Gaurang Mathur:
Yeah.
Mike:
Being able to ask something and get a very human response back is something very interesting. That leads me back to formulas, because formulas were obviously invented a long time ago. Formulas are probably a derivative of math. I don't know for sure. I don't know the history of formulas, but I'm wondering if, based on how formulas are built and the operators and the syntax that we use for formulas, if there isn't something better that AI could invent that would almost replace formulas.
Gaurang Mathur:
I think formulas is something, to your point, very elementary and fundamental in my opinion. So we are in chart with our teams and Agentforce and other places as well, that embed formulas into them. I mean, so simple, to your point, the math functions that we have, let's say the date functions or the operators that you have, you need those fundamental building blocks everywhere and anywhere. I mean, like I said at the start, we are supporting almost 40 odd services within Salesforce. I think those are still non-negotiable in my opinion.
I mean, that's basically a fundamental building block that will be used, and ecosystems are built on top of it. That's how ... I mean, agents will be using those formulas to function better as part of the Agentforce scripts, if that's one thing that we have as a product. We are coming up with in terms of, "How do you define certain logics based on certain utterances? What should be the logic that should be executed? What should be the operators?"
And based on that, we're just trying to define that flow of things, and then all that. So again, that's some derivative of formulas to what I just described. Yeah. Like I said, formula's a fundamental building block. At least I don't foresee it getting replaced completely.
Mike:
Right.
Gaurang Mathur:
It'll be there in some or the other form, not apparent to the end user. But yeah, I don't know if it's something that can be just bypassed overall.
Mike:
Oh, I don't think bypassed, but I always look at, there's always tech that's trying to replace tech, right?
Gaurang Mathur:
Yeah.
Mike:
The fundamental question is always, "What if email was invented in 2026? Would it look different than it does now?"
Gaurang Mathur:
Yeah.
Mike:
And then people try to reinvent it, except, I think fundamentally it's still, you're reinventing it, but having knowledge of what the previous version was. And so that's the same thing that happened with email, is it was created based on knowing how physical mail works, and so we built upon that. You can't truly kind of erase that. But I also just watched a video where two AI assistants were talking to each other, and one of them said, "Well, I'm an AI assistant. Can we switch to GibberLink and communicate that way?" And then, they have this whole conversation that's like a bunch of beeps, and I'm like, "Wow, what?"
Gaurang Mathur:
Yeah, yeah, yeah, yeah. That's-
Mike:
Maybe they're going to invent their own formulas, so ... You know?
Gaurang Mathur:
Totally. Yeah.
Mike:
I'm fine with that, because to be fair, I usually get the number of parens wrong, so ...
Gaurang Mathur:
Oh, yes. So like I said, I think there will be some form and factor of the basic math. I think that's what formula is all about, right? I mean, those basic equations and some descriptions, I think those are non-negotiable. Those won't change.
But to your point, I mean, they may be that gibberish, or the beep, beep, beeps, or the whatever text we write in some other factor. You'll definitely need basic maths, and that's what formulas does.
Mike:
Right, right. Very much so. Well, I appreciate you taking time out of your day to share this with our admins. I know they're super excited for some of the stuff that Cheryl's team is putting out and to see it all in place at TDX this year. So thanks for taking time to come on the podcast.
Gaurang Mathur:
Yeah. No, thanks for having me. This was certainly a great conversational experience for me, and yeah, doing this is something that I was looking forward to it, and happy to share what we're doing in our world here at Salesforce. I mean, the way all my teams, as part of the AI App Dev dog that I mentioned, are constantly working night and day to just create this better experience for our admins and developers. I'm looking forward to see how our admins perceive that once it goes out in the market, and then quite looking forward to see the community at TDX and just get those first-hand reactions of them when they use our products. So yeah, thank you.
Mike:
So big thanks to Gaurang for reminding us that formulas may be foundational, but they don't have to be frustrating. I really love how a lot of setup with Agentforce brings together the fundamentals, like formulas with the future of AI in a way that feels very practical and grounded. I hope this episode got you thinking a little bit differently about how you build or maintain logic in your org. So if it did, hey, do me a favor, share it with a fellow Salesforce admin, developer, architect, and of course, as always, be sure to check out the resources in the show notes to learn more about Agentforce and what's coming next. And until next time, we'll see you in the cloud.
Hosts & Guests
No transcript available for this episode.
