How Agentforce Vibes Is Changing How Salesforce Admins Build Apps

By

Today on the Salesforce Admins Podcast, we talk to Tiaan Kruger, Senior Director of Product Management at Salesforce. Join us as we chat about Agentforce Vibes and what it really means to build with AI on the Salesforce platform.

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

Agentforce Vibes is evolving

If you caught the Admin Keynote at Dreamforce ’25, you might recognize Tiaan from his Setup with Agentforce and Slackbot demos. Right now, he’s looking at new ways to help admins expand their toolkits with Agentforce Vibes.

As Tiaan explains, Agentforce Vibes is developing so rapidly that it often surprises the people who build it. When one of his team members was building a React demo that featured a dashboard, they realized it would look pretty boring without some sample data. They asked the Vibes Agent to spin something up, and it delivered (after a few tries).

“These agents surprise even us,” Tiaan says, “we’re still discovering where its power lies and where its potentials are.” And that’s why it’s a good time to look ahead at how Vibes will change the game for admins.

AI as a tool for reducing technical debt

“When I was a customer developer, we always had a two-year backlog,” Tiaan says. Even with a big team, there’s always too much to do and you’re constantly accumulating technical debt. This kind of grunt work is where he sees Agentforce Vibes making a big difference, which gives you more space to look at the bigger picture.

Tiaan points to Code Analyzer as an example. You can use it to quickly scan your org and identify where there are performance or security issues that you should take a closer look at. An admin might not be able to fix it themselves, but they can hand it off to a developer team and speed up the process.

There are also some exciting things coming with React and Agentforce, which will allow you to create compelling, high-quality UIs for your apps. It’s all about giving you more tools to get the ball rolling and build apps faster than ever before.

How to help Agentforce Vibes help you

In order to get the most out of these new features, you need to make it easy for agents to understand what’s going on in your org.

“If you don’t have good descriptions on everything in your org,” Tiaan says, “please, for the love of donuts, fill those fields in.” Having good metadata will help you get the most out of everything Agentforce has to offer. The future is bright, so Salesforce Admins need to get ready.

Make sure to listen to the full episode for more from Tiaan about how Agentforce Vibes will help admins. And don’t forget to subscribe to the Salesforce Admins Podcast to catch us every Thursday.

Podcast swag

Learn more

Admin Trailblazers Group

Social

Full show transcript

Mike Gerholdt:
Welcome to the Salesforce Admins Podcast. So today, I’m joined by Tiaan Kruger and we’re talking about something that’s showing up literally everywhere right now. And that is Agentforce Vibe coding and what it really means to build with AI on the Salesforce platform. We get into how admins can use agents to move faster without losing best practices, and why oh why understanding your org still matters more than ever. And one of the things we tackle is where this whole thing is headed sooner or later than you think. So if you’ve ever been curious about Agentforce Vibes and not sure where it fits into your day-to-day admin life, I promise you this episode is for you. So with that, let’s get Tiaan on the podcast.

So Tiaan, welcome to the podcast.

Tiaan Kruger:
Thank you, Mike. Thank you for having me.

Mike Gerholdt:
Well, it was not that long ago that we had Cheryl Feldman on. And Cheryl’s circle of influence outside and inside Salesforce is a gravitational pull. And of course, we all know that set up with Agentforce is going to be huge at TDX and it’s huge right now. So we wanted to talk to more people that were making admins lives wonderful. And she suggested you. Plus you were in the admin keynote. And I feel like I have to have admin keynoters on the podcast too. So there’s the really long intro for… But let’s talk about you. How did you get started at Salesforce and what’s the cool thing you’re working on that we’re going to talk about?

Tiaan Kruger:
Yeah. So I’ve actually been at Salesforce, I guess coming up on seven years, which is absolutely mind-boggling. I was a customer for 10 years. Spent a lot of time as an admin and Apex developer. I wrote Visualforce. Some of you may have heard of that before. But I work as part of a large team at an enterprise company trying to make Salesforce make a huge difference for our customers. And in the process, learned an incredible amount of Salesforce and got an opportunity to come over here and try to make products better for our admins and developers out there. And I grabbed the opportunity and I’ve loved every minute of it.

Mike Gerholdt:
Yeah. Well, we’ve talked to you about some cool stuff. Plus you did a really cool demo in the admin keynote at Dreamforce too.

Tiaan Kruger:
Yeah. Yeah. We got to do some really amazing things at the admin keynote, showing some of the things that the Agent for Setup is going to be able to do. We got to show some amazing insights into Slackbot. And we got to foreshadow a little bit the work I’m currently working on, which is focused on how we actually build with AI. So I’m working very deeply with the Vibes team to make sure that the metadata that gets created is better and great quality, not just for vibing individual pieces of metadata, but vibing entire solutions. I’m getting to work with our internal teams to figure out how we bring that same power to our admins.

So yeah, absolutely. So the work that we showed on stage continues at the same pace internally. And we’re just trying to figure out how do we bring AI to everybody, make sure that everybody can feel the lift of this incredible technology.

Mike Gerholdt:
Yeah. So I did a podcast with Daryl Moon, which I’m sure you’ve listened to. Daryl is fantastic. He’s down in Australia. The only reason he listens to the podcast, it’s a half hour between his house and the dock, so he can listen to it to go fishing. But he tried out Agentforce Vibes and was thoroughly impressed. And he literally threw the kitchen sink at it. And I think what he loved about it was it got him so much farther into the application building than what he anticipated.

Tiaan Kruger:
Yeah, absolutely. Vibes is quite incredible, specifically the agent that’s in there. It’s a very wide open agent. It’s very good at problem solving. One of my favorite stories at the booth that we told at Dreamforce is one of our product managers was actually vibing a React app. We’re not bearing the lead. We actually talked about this at Dreamforce. React’s coming to the platform, incredibly exciting, creating pixel perfect UIs, reusing a lot of the things that’s out there. And so he was actually experimenting and vibing this application.

And he realized that a dashboard with no data is really boring. It’s just not very interesting to look at. And so he thought, “Maybe I should ask Vibes if it can produce me some sample data.” Not knowing if it could or not. And so he asked it to do it and it went off and turns out it generated an anonymous Apex file that actually tried it to load all the records with data. And it tried it, failed, found a bug, fixed it, did it again, found a bug, fixed it, and then successfully inserted a bunch of data. And that’s just one of the stories of how these agents surprise even us. We have this technology, we’ve put it in these tools and we see the incredible things it can do. But even with all that, we’re still discovering where its power lays and where its potentials are.

And so that’s really where I’m focused right now is like, how do I make sure that, more often than not, you actually get a good result out of it, right? Make it so it has to retry less or debug less. And so that’s a big focus for us right now.

Mike Gerholdt:
So let me ask you a question. With Vibes and vibe coding… By the way, I’m still getting my head wrapped around you talk about, “He was kind of vibing.” Like I’m just old enough to remember what the 70s were. I feel like you need to vibe code in bell bottoms and stuff. How long until anybody just tells an agent to build something and it won’t matter if you have knowledge of development and coding or knowledge of best practices for enterprise architecture?

Tiaan Kruger:
If you asked me this question six months ago or nine months ago, if we’ve given you a different answer than I would give you today. Based on what we’re seeing, the rate of innovation and change, I don’t think it’s going to be very long at all. We’re going to have to adjust how we build and how we deploy a little bit because agents are already getting very, very good at producing targeted applications. And what I mean by that is, there’s still limits when you are trying to vibe inside of a massive enterprise org with thousands of objects and a lot of complexity that somebody that’s got a ton of experience would need to come into for months to understand. There’s still limits there, right? You still have only so much you can actually provide the agent at a time.

But if you are focused, you are looking at problem solving around a particular application or a particular business case, it is actually getting extraordinarily good at producing not just a good backend and a good data model, but also producing incredibly compelling UI, which is actually why we are bringing React to the platform. If you are an admin and you’re like, “What’s React? I haven’t heard about React. What is this thing?” It’s just a way to build WebUI, but it turns out, because there’s so much examples of it, so many examples of them on the web, the agents are incredibly good at producing incredibly high quality UI with that.

So to answer your question more succinctly, I don’t think that future’s all that far away because it’s already starting to show up. And there’s consumer applications and mobile applications that’s out there where, for the end user, they can start vibing the mobile app that they want just with AI, just having conversations. There’s incredible tools out there that’s starting to emerge. And that exact same wave needs to come to the enterprise. We need to unlock that incredible innovation of the fact that you don’t have necessarily software created for you that kind of meets you where you’re at, but really, truly have agents create software for you that is absolutely what you need locally to solve your problems.

So we’re getting closer to it every single day. And I think the closer we get to it, the faster it’s accelerating.

Mike Gerholdt:
So one of the questions that came up, this feels like a hundred years ago when we launched Process Builder, was it was so easy to create processes that people without technical knowledge could just overprocess and build way too many processes in an org when… The example would be of best practices, just write a really good trigger or deploy some Apex code and take care of it that way. With AI doing so much building, how do we know it’s following and architecting best practices or building good code coverage?

Tiaan Kruger:
That’s a great question, and actually one of the biggest challenges that we are spending a lot of energy on. We know that when we put a tool in the hands of our admins, they have very high expectations. They expect it to create best practices style orgs and configuration and really follow the practices that we want our admins to follow. And so there’s a couple things we’re doing to help with that. I can give you some insights into some of them. We are internally working on ways to… And this is, again, we’re working in Vibes at the moment, so it’s a little bit more pro code focused. But just listen to this and imagine how this is going to come forward to admins, right?

All the work I’m doing right now is more for pro code, but we are going to start exposing it later this year for admins. As the system is producing configuration software, we are taking it through a specification phase, a plan phase. And in that plan phase, we’re feeding it all kinds of rules and guidelines in terms of how we wanted to build an architect for you when it’s constructing the experience, right? Not just the UI, but also the backend. How you want fields named, et cetera. So there’s ways that you can feed rules into the system. So that’s how we’re starting to help it architect from the get go. That’s one way.

Second way is, as part of those experts or rules that we’re providing, we can actually also guide its hand when it’s creating solutions. So if you’re creating a lightning app or you are creating a flow, we can actually guide the agent’s hand a little bit and say like, “This is the right way to build it or the best way to build it.” But even with all that, if you generate an application or you generate a capability, you still need on the flip side to actually test it. And so that’s the other areas where we are looking at, starting to develop more tools and actually add more tools to the pipeline, which is what we call validation.

Once you’ve actually produced… Let’s use the example of a web app, right? I produced a web app that created custom UI for me. How do I know that that is accessible and it is secure and it is performant? These are all things that you can measure and test, right? And so that is absolutely parts of the tools that we are looking at building and we’re developing a number of those that’s going to help people not just build random things, but actually guide the agent’s hand to build good quality, best practices style material.

Mike Gerholdt:
Yeah. I mean, one of the analogies that always comes up, and I was just doing a podcast with Josh, is, we were talking about GPS and maps. Because I feel like there is a portion probably of my listeners that are old enough, myself included, to remember stopping at gas stations when you got into a new state and buying a map for that state because it was always more detailed, and being able to read the map. And now, virtually every car you get into has some sort of GPS either built into it or it’s on your phone. The funny thing is, the other day… Well, this is a few months ago. I was driving back from a bigger city in Iowa, one that’s just big enough that I don’t know it, even though I’ve lived here forever.

And the GPS actually took me down a road. It was a highway and I was like, “Okay, cool. I know this highway connects to the next highway that gets me back home.” But in two miles, it had me get off on an exit and get back on the interstate.

Tiaan Kruger:
Interesting.

Mike Gerholdt:
And I had presence of mind to be like, “Okay, well, I know this road continues. There’s no roadblock or anything.” So I just went straight and let the GPS figure out. But I bring that back to my knowledge and ability to still understand and read maps is what helped me not blindly navigate that problem. And I feel like still having that understanding of, well, reporting structures are going to need this for object setup or relationships and making sure to almost double check your math is what I was getting at for how an AI should work. So long explanation longer. If you were heading into an organization as an admin, developer, architect, what would you focus a lot of your learning on?

Tiaan Kruger:
Wow. By the way, I’m a big fan of maps. I’m old enough as well to have used maps to navigate.

Mike Gerholdt:
You probably remember printing off MapQuest directions.

Tiaan Kruger:
Oh, now that I’d rather forget. I’m a bigger fan of following a paper map. Famously, we were traveling in Boston on the train and I forgot the maps on the seat. And when we got back, the maps were gone. You can imagine that feeling, right? So navigation became a little bit more of a challenge. It was like 11 o’clock at night.

Mike Gerholdt:
Oh.

Tiaan Kruger:
Yeah. Yeah. So yes, maps are very valuable. So I actually love that analogy, right? So agents are a lot like… I think they’re in that MapQuest stage. They’re not quite… They’re getting closer to the GPSs we have today. The odds of the GPS turning you into a lake these days, I feel like it’s a lot lower than it used to be.

Mike Gerholdt:
Right. Right.

Tiaan Kruger:
There’s still the occasional person driving into a lake, but it’s a lot lower risk than it used to be. Agents are still very much at that state of the human still ultimately needs to guide it, check it, see what it’s doing, right? It’s an accelerator, not a replacement. And if I were going into a company now in my old role, my role would be changing. My role is no longer just about cranking out that trigger to work perfectly. It’s about understanding the bigger picture. It’s about understanding what my customer is actually trying to accomplish. And hopefully being able to get more of it done because I become more… I’ll actually give you a different analogy, I think, as you brought up driving.

I’m a big fan of full self-driving capabilities being added to vehicles. And I’ve had some different features on the cars that I’ve owned, and some of them more beta than others. And you have to do the supervising thing, right? The car will do the right thing nine out of 10 times, but that 10th time is really scary. You better be ready to grab your steering wheel. It’s that kind of behavior. You’re getting somewhere a lot more rested because you’re not doing as much of the grunt work, but it frees up some of your mind to focus on some of the bigger picture. And so that’s really what I’m looking and hoping that we’re going to see.

I’m seeing it happen already to developers here internally. I’m hoping that’s going to happen for our customers as well. So as we put more of these tools in their hands, that is going to do some of that grunt work, like building a trigger that is very well set up to do batching, for instance. And I’m sorry, it’s a developer analogy. I get it. But there’s a particular way you build triggers. And if you don’t do it right, it’s going to cause you performance issues or it limits. That’s just going to start being natural. You don’t have to think about that. So now that frees up some of your focus to say, “Okay, how can I build this better for my customer? How can I focus on actually producing more value for them?”

When I was a customer developer, we always had a two-year backlog. And business, we turned business requests away all the time because we could not get to everything. And we had giant teams doing this stuff. So the benefits of learning the platform, understanding the details, understanding if the agent is trying to turn you onto a lake, that doesn’t go away. That continues to be there. Understanding the deep, complex context of what your business needs. The agent’s not there yet. So that transformation, that translation layer is more important than ever to make sure that, again, the grunt work can be done by the agent, but that understanding, that comprehension, the translation, I think, is more and more important than ever for our admins and for our developers out there in the community.

Mike Gerholdt:
Yeah. And you don’t have to apologize. There’ll be vibe coding sessions in the admin track for TDX. I know because I’ve seen you and your team submit some stuff and customers have done it as well. It’s going to be a very useful product. I can see the future for it. Do you anticipate… So a lot of what you’ve talked about for vibe coding in Agentforce Vibes is like the creation. And I think we’re always talking about the creation. But one thing you brought up in your answer was tech backlog and technical debt. That’s always a big thing that comes up. As we’re struggling to create more and more and more and more, there’s also more and more and more technical debt.

Would you envision a time when maybe Agentforce Vibes or another product could say, “Whoa, stop. We need to reevaluate this first and fix the foundation before we go building seven more stories on top of this house?”

Tiaan Kruger:
Yeah. So there’s a couple of different pieces to that puzzle. There are already tools coming out today, and that’s available today in the spirit of Code Analyzer, for instance. It’s a tool you can run in your org that’ll tell you where you have performance issues or security issues you should take a look at. We are expanding that concept to look for things like accessibility issues, et cetera. And AI, the agents we have is already able to go and potentially help you resolve some of those. You can go in Vibes today and say, “Hey, can you help me convert this Aura component to an LWC component? Or can you help me put better accessibility on this component or these pages?”

And it’s actually doing a fair job with that already. Make this React app work for mobile is something else you can do. So there’s already some tech debt you can actually tackle with agents very successfully. They tend to be a little bit more focused right now in the sense that it’s a single page or a single component or a single code file or a couple of code files. I don’t even have to say maybe. The answer is, absolutely that capability is going to continue to go further and further. The more the agent is able to understand what’s going on in your org, the more the agent is able to grasp and understand what your org does.

And this is a call-out for you. If you don’t have good descriptions on everything in your org, please, please, for the love of donuts, fill those fields in. And I know we all love to put description and description. But if you actually go put in good quality descriptions in these things, you can actually get so much better help from the agent actually understanding what your org means. When you just have four fields named contract number, it’s really difficult for the agent to know which one to use. But if you label them more carefully, you put it in the help and the metadata, you’re going to help agents help you.

So tech debt is going to get easier to resolve. No question. It’s already happening. It’s only going to get better. It’s the worst it’s ever going to be right now. I could truly say this because I’m seeing it happen. And I think we’re eventually going to get to a place… Again, I’ll give you an internal example. We have a lot of Aura components. I know I probably shouldn’t say that out loud, but we do. And we are working at fever pace to actually convert them to Lightning Web Components to LWCs. Again, developer specific terms, but just the different component frameworks that we’re moving to more of an open standard.

And we used to have to do it by hand. This is huge tech debt for us. And AI is literally having one team do hundreds of components in a month where they used to be able to do a handful. This is real tech debt help, right? So that is absolutely going to come and it’s going to show up for admins as Cheryl. You talked about Cheryl in the beginning, right? She’s a superhero of mine. I absolutely love working with her. I love watching how she works with the community. I’m always learning something from her. She recently actually helped me with the use cases. If any of the people listening got to talk to her, thank you for giving her some information. She’s just fantastic.

And so as the setup agent becomes more and more capable, it is absolutely going to help you identify tech debt, first of all, but second of all, actually help you resolve that tech debt in faster and easier ways that you’ve ever been able to do.

Mike Gerholdt:
Well, that’s very promising, very exciting. Tiaan, I should have scheduled five more podcasts with you because I feel like we just scratched the surface on everything. But when you’re not vibing at Salesforce or making Vibes, can you vibe Vibes? Oh, that’s like the buffalo buffalo buffalo sentence, isn’t it?

Tiaan Kruger:
Start off all the way down.

Mike Gerholdt:
What is Tiaan like to do outside of Salesforce?

Tiaan Kruger:
So my side hobbies, I drive my wife nuts because I try to learn a new thing every couple of years. And I do anything from fabricating and welding and soldering and doing electronics and building things. But I would say probably the most practical application of all of those things that I just mentioned that I do as a hobby that’s consistent is I work with a FIRST Robotics team. And my son got me into it. He was the robotics captain at the team for two years. They managed to take him to worlds both years. And I absolutely love that. I love the fact that I get to be in there with the kids, help them every year try to solve whatever the year’s challenges and problems are. And they hand build a robot that’s 150 pounds that can drive way faster than it should.

And it’s fascinating to work with these kids and see them learn. And now AI and agents are coming in for these kids, right? And they’re trying to figure out, “Where do I use it? When is it cheating? When is it actually allowed in helping me?” So that’s been a fascinating twist to that. But you know what? There’s still something really satisfying about building things with your hands. We have some people that speed in my neighborhood. Sorry to call them out, but they do.

Mike Gerholdt:
I’m sure they listen to the podcast.

Tiaan Kruger:
I’m sure they do. I’m sure. They feel worse now.

Mike Gerholdt:
Well, I’ll send in stickers.

Tiaan Kruger:
I wanted to build a little sign. When we were in Germany this year, I saw those brilliant little speed limit signs that when you go over the speed limit, they turn orange and then frown at you. So they literally just scowl at you when you’re speeding. And it makes you feel bad.

Mike Gerholdt:
Yeah.

Tiaan Kruger:
And I thought, “You know what? This would be really amazing to build.” And so I actually took some of my robotics experience that I use for the kids. I do a lot of things with LEDs and Arduinos. And I actually bought a little radar board. I built a sign that could do this. And I promise it’s related. I got to use AI to code this thing. And I’ve done the same kind of coding for years before. Mike, I coded this thing in about 15, 20 minutes to get this working for the first time.

Mike Gerholdt:
Is it solar-powered?

Tiaan Kruger:
It can be. It’s USB. It’s very light on power.

Mike Gerholdt:
Okay.

Tiaan Kruger:
But it’s amazing.

Mike Gerholdt:
Wow.

Tiaan Kruger:
I was just able to tell the agent, “Hey, here’s the board that I’m using. Here’s the LED panel I’m using. Here’s the radar.” By the way, it’s a Chinese made radar board, so the documentation is hard to understand and limited.

Mike Gerholdt:
Yeah.

Tiaan Kruger:
And of course, AI had no problem with it. It coded it up like it was nothing. And so it’s transformative. Even in your hobbies, the stuff is transforming how we’re doing it. Right?

Mike Gerholdt:
Wow.

Tiaan Kruger:
So yeah. So that’s what I do. I build stuff because I build software for a living. I like to build stuff with my hands when I’m not at the computer.

Mike Gerholdt:
It doesn’t surprise me. Everybody I ask that question to that’s in software development always has something very tactical that they like to do as a hobby. Because a lot of what we do, you can’t touch and feel.

Tiaan Kruger:
Exactly.

Mike Gerholdt:
Very cool. Tiaan, thank you for coming on the podcast. And I look forward to seeing some of the stuff you’re going to present at TDX this year.

Tiaan Kruger:
Awesome, Mike. I will see you there. And keep watching. This is a field that’s going to continue to blossom.

Mike Gerholdt:
So thanks, Tiaan, for that conversation. I feel like I literally could do four more episodes and still not run out of things to talk about. He is one inspirational person and doing a lot more with AI than sometimes I think. So it’s good to know that someone still needs to know where the road actually goes when the GPS is about to drive you into the lake.
If this episode sparked a few ideas or questions for you, do me a favor, share it with your fellow Salesforce admin. And until next time, we’ll see you in the cloud.

Love our podcasts?

Subscribe today on iTunes, Google Play, Sound Cloud and Spotify!

The Importance of Human in the Loop for Agentforce

Today on the Salesforce Admins Podcast, we talk to Joshua Birk, Senior Director of Admin Evangelism at Salesforce. Join us as we chat about how the human in the loop is key to building reliable, predictable AI. You should subscribe for the full episode, but here are a few takeaways from our conversation with Joshua […]

READ MORE