Who will be the adult in the room on AI?

What does it take for companies to use agentic AI to transform the enterprise, without losing control of the technology?

In this episode of The Deep View Conversations, we sit down with Shibani Ahuja, SVP of data and AI strategy at Salesforce, to discuss how one of the world's leading software companies is applying AI with practical use cases, matching governance to risk, and building toward larger transformations ahead. 

Salesforce has surprisingly embraced a "headless" AI strategy that lets customers use any AI to access their Salesforce data safely and securely. That includes its own Slackbot, which sits inside one of the world's most widely used business messaging systems. In this interview, we learn more about why Salesforce wants to give customers optionality.

Shibani also lays out Salesforce’s four modes of enterprise AI, from everyday assistive tools to agents that can reshape end-to-end operations. We also discuss Koa, Salesforce’s new CRM reasoning model, why the model-plus-harness approach is so critical, and why adaptability may be the defining enterprise skill of the AI era.

Topics covered:

• Why organizations should start with practical, level-one AI use cases
• How Salesforce matches governance and ROI expectations to the risk of an AI deployment
• What Koa, Salesforce's AI model built on NVIDIA Nemotron, changes for enterprise AI
• Why operating models, process expertise, and professional services matter as much as the latest technology
• Shibani's case for AQ: the adaptability quotient for technology stacks and teams

If you’re trying to make AI more efficient, safer, and more ROI-driven, this conversation offers a practical framework for how to build it, how to govern it, and where to start.

Subscribe to Deep View Conversations for interviews with the leaders shaping the future of AI, business, and technology: tdv.transitor.fm 

And don't forget to sign up for The Deep View daily newsletter. We don’t just cover AI, we decode it. In a world flooded with hype, we deliver sharp, no-nonsense insights to keep you ahead of the curve and help you put AI to work every day: subscribe.thedeepview.com

Creators and Guests

Host

Jason Hiner

Editor-in-Chief of The Deep View

What is The Deep View: Conversations?

From frontier labs and enterprise platforms to emerging startups reshaping entire industries, The Deep View: Conversations podcast interviews the brightest minds and the most influential leaders in AI.

Jason Hiner: Welcome to The Deep View Conversations podcast. Shibani, so good to see you again.

Shibani Ahuja: Nice to see you again.

Jason Hiner: For those who don't know, talk about who you are and what your role at Salesforce is.

Shibani Ahuja: Well, my role at Salesforce just changed about a month ago. So I think when we last chatted, I was leading Enterprise IT Strategy and what that meant was I was meeting with C-suite leaders, particularly CIOs around the world advising on AI transformation. Well, turns out I gave some really good advice and the company said, that advice that you've been giving to other folks, can you do it for us within the company? And so now I'm leading data and AI strategy for COFO. So COFO is all of our CFO and COO functions. And so now what I'm doing is all that advice I'm applying and we are driving, we're building the agentic enterprise. I'm, me and my team are building the agentic enterprise for Salesforce.

Jason Hiner: Wow, okay. Yeah, I was going to start this by saying, since we talked in February, a couple of things have happened.

Shibani Ahuja: Couple things.

Jason Hiner: But that wasn't even in my purview.

Shibani Ahuja: Surprise.

Jason Hiner: That's good. I do remember seeing it on LinkedIn. I remember seeing something change on LinkedIn. But that's huge. I mean, you think about, even since February, you all have launched your own AI model.

Shibani Ahuja: Yes.

Jason Hiner: Yeah. You have, yeah, the agents have gone parabolic. The inmates have escaped the prison.

Shibani Ahuja: Come on now.

Jason Hiner: Oh, the AI inmates have escaped the prison. Like a lot has happened, right?

Shibani Ahuja: So we've also built a beautiful harness that ensures safety, security, governance. We've evolved from probabilistic and unpredictable agents to probabilistic plus deterministic. I think, listen, Jason, we've got to paint a balanced picture.

Jason Hiner: Okay. I appreciate it. So a lot of people have told me this week, you know, which is good.

Shibani Ahuja: Yeah.

Jason Hiner: Which is good. Your team is very on message with this. Like, because I've used the inmates, I'm not the person I've used that joke this week. And like, I've gotten the same thing, like we've got it under control. Like that's what we're here for. I will say that my, to be fair, my headline was that Salesforce might be the adult in the room on AI. So.

Shibani Ahuja: Jason, I'm down for that.

Jason Hiner: So it tracks. It tracks. What is it? What does it look like for your role? Because last time we talked, you'd been talking to CIOs, a lot of IT leaders, a lot of different enterprises. They're trying to get their arms around AI and they were, you know, with varied rates of success and confusion and all of that. And, you know, you are dedicated to helping them. I think what you were like, I've talked to, I can't even remember. I was a lot.

Shibani Ahuja: 587. It's funny.

Jason Hiner: I was going to say over 500 and I thought that can't be right. That's just too many.

Shibani Ahuja: It was 587. Like it would be one to one was one thing, but it was like one to many. It would be a lot of conferences. And it was also at third party conferences, not just our own.

Jason Hiner: It's still, that's a lot of folks. That's a lot of folks. Okay. So you talked to over 500, almost 600 CIOs about AI.

Shibani Ahuja: Yeah.

Jason Hiner: What is your, when did your role change and what's it looked like since then?

Shibani Ahuja: So about about a month ago when I shifted into this new role and it was funny. I am now giving myself the, and I'm trying to remind myself of the advice that I've been giving to CIOs along the way. That we're saying exactly what you were saying is that like, this is a little bit overwhelming. I've got everyone that's got an AI solution. I've got tech debt. I've got data that needs to be harmonized. Like I know I got to do something about that. I know I have to figure out agents. Like it was a lot coming at them.

Jason Hiner: Sure.

Shibani Ahuja: And I remember back then what I would say now, and I think now I've tweaked it even further in terms of like as I'm, what I'm learning about what we are doing at Salesforce. Back then I would have said, you know, like take a breath. We, I think we talked about the agentic maturity model.

Jason Hiner: Yes.

Shibani Ahuja: I had said like map, you probably have 100, 170 use cases, ideas of agents that you have thought about. Take that and map it against the agentic maturity model. And the agentic maturity model was a model that we had built or it was a framework that we had built that showed a chat box on one side. Then you got co-pilots or like generative AI and then you got the spectrum of agentic AI. And I would say just map all the use cases you can think of against that maturity model and then start at level one.

Jason Hiner: Right.

Shibani Ahuja: Because I think that that was a bit of the overwhelm is people are like, how do I build this utopian agent that's going to speak to other agents that's going to traverse the virtual halls? And I was like, no, no, no. Start at level one. And so I'm kind of coming back to those basic principles to say, where do we already have level one use cases where we should be understanding what we've done, how we've had those successes. And so that's what I'm coming back to to say, let's look at those level one, level two use cases, and then work backwards to say, okay, what's the minimum critical data elements that we need to get perfect for that agent? What's the tech that we need to connect? And I'm using simple terms for that agent to become active. And in the process, you're harmonizing and cleansing the data elements that are going to drive impact in the process, you're starting to look at technical architecture from the past that you might need to like refresh and revamp. So it's not like a two year project. You're now doing it in pieces and you're starting to see value and benefit at every increment. Right.

Jason Hiner: Two-year projects are not much of a thing anymore.

Shibani Ahuja: No, it's like it's it is it's from going from project like waterfall to agile delivery. Yeah, it's sprints. It's bringing teams together.

Jason Hiner: Yeah.

Shibani Ahuja: Technology, the business users, bringing all these folks together to be able to do this collectively and everyone is learning from each other at the table is cross functional pods.

Jason Hiner: So what is it like being the first and internally who has to go to to go to teams to like help them? Because I I expect some of your job is like helping empower maybe in some places like hastening a little. So what's the list that look like?

Shibani Ahuja: Yeah, like one of the earliest things that I've discovered it because we've got to actually define AI. Everyone's like we on one hand you're hearing everyone say we've got to democratize AI. Every AI for everyone in an organization. And then on the other hand, and this is still conversation I'm having with X remember I've been in my job for a month. I'm having with with customers is but like the cost of AI and token token costs right like so those are two ends of the spectrum. So I'm reflecting on that internally for me to say like within our organization going how do I how do I want to handle this? And I'm pausing to say I think I have an opportunity to define AI. Similar to the maturity model. How do I define AI and the maturity model is really for technologists to understand consumption, tech stacks, data cleansing, all that good stuff. But I'm defining AI in our organization is like level one AI or mode one AI. Okay, is where we've given Gemini we've given Slack and Slackbot. Slackbot to everyone. We have officially democratized AI because everyone has access to it.

Jason Hiner: The whole organization.

Shibani Ahuja: Everyone. Everyone has Slackbot built into Slack as we use Slack. So everyone is able to leverage AI today and it's an individual assistive. It's assistive AI.

Jason Hiner: Okay.

Shibani Ahuja: So in that one I have everyone's got it and it's not like we are expecting ROI in a tangible way.

Jason Hiner: Okay.

Shibani Ahuja: It is assisting and aiding a colleague. Level two is where we're saying, okay, now we've got things like Claude Cowork and Claude Code that are more advanced models. Not everyone in the organization needs that, but I'm going across the corporate functions and saying, who needs this access? Why do they need it? But I'm not limiting it. For that level, all we're doing is an inventory of the skills, agents, and capabilities that are being built, and we're looking at consumption today. Where are we finding amazing power users that we want to showcase? Then you get to mode three of AI, where we're saying, okay, I think that we've got some processes within our organization, within, say, finance at L3, that we could automate and agentify. We bring the process experts from finance to sit with us for a two-week sprint. I will bring experts with Agentforce Operations, formerly known as Regrello. We'll help you create the blueprints, automate, and streamline this, and now we can actually start to quantify value and benefits.

Jason Hiner: Okay.

Shibani Ahuja: Then you got the last mode of AI which is where organizations are some are stepping back and saying what if we could rearchitect our entire company? What if what if like lead to cash the point at which you have a lead or a prospect right down to where you get cash in hand that we could imagine a world where we've got agents that could traverse multiple departments, multiple functions and perform certain actions. Like that is a slower build. There's a different governance. There's a different blast radius should something go bump in the night.

Jason Hiner: Sure.

Shibani Ahuja: So I'm starting to look at this in four different planes and appropriately understanding what is the ROI? What is the appropriate who should be building these agents?

Jason Hiner: Okay.

Shibani Ahuja: The governance of these agents. And so I think that this is now starting to make a bit more sense in are we are we building in the right spaces and are we doing a little bit in all of the categories?

Jason Hiner: So that's a lot of categories.

Shibani Ahuja: It is.

Jason Hiner: That's good. It is. Now it's time for a word from this week's sponsor, Deel. Global hiring and payroll can be a serious headache. We get it. Most teams juggle five, six, sometimes 10 different HR tools just to hire and pay people abroad. Deel replaces those tools with one AI native platform for HR, IT and payroll companies like Shopify, Revolut and ElevenLabs already scale internationally with deal and Airwallex. The leading AI finance platform manages and pays over 1100 employees across 16 countries. To build your global team with deal and start expanding your business today, visit Deel.com slash deep view. That's Deel.com slash deep view. We thank Deel for their support of the deep view. And now back to the show. When we talk last time, you said this thing that stuck with me. You're like AI is the smallest, biggest word.

Shibani Ahuja: How do you remember that?

Jason Hiner: That's amazing. Because you're memorable. But it was such a wise thing to say, right? Because people define it so differently. Because the expectations of it and the fear of it are like these two ends of the spectrum.

Shibani Ahuja: Exactly.

Jason Hiner: That are and then the argument gets defined by like these two very, there's a lot of outcomes between AI is going to kill humanity and AI is going to cure all disease and bring an end to poverty and have unlimited abundance. Like there's a large set of outcomes, potential outcomes in between that.

Shibani Ahuja: Yeah.

Jason Hiner: Right? And so I wouldn't, I've been looking forward to asking you about this, like how Salesforce thinks about this because I think that's part of the challenge right now. It's helping people realize like it's not, and also these two things are not mutually exclusive like the bad outcomes and the good outcomes, right? AI is a tool that people use. Salesforce is a tool company. You know, how do you all, when you're working with your customers, help them have a vision for what AI is? Because I'm sure that's part of it. Like there's all the confusion. There's the fear. You know, I think, and one of the things I got from you when we talked before is like that was what you saw as part of your job was to go and help people, you know, get their arms around this like as a mental model. Like how do I understand this and what it means for us? Yeah.

Shibani Ahuja: Yeah.

Jason Hiner: Or is that what you're doing internally too? What, how's that look?

Shibani Ahuja: So, it's so funny that you say AI is the smallest, biggest word because that's exactly what I'm saying. And even it's like as I talked to you about these modes.

Jason Hiner: Yeah.

Shibani Ahuja: Why am I even thinking in this notion of modes is for exactly what you're describing. Are we expecting that when we give someone Slackbot or Gemini that they are going to, we are going to be able to calculate the ROI and we need to be able to govern and control this and like you're going, no, actually no. At a certain point, at mode one, don't expect that return that is something that you can hold on to. And also don't over govern it because there's a limited amount of things that folks can do with like everyday assistive AI.

Jason Hiner: Yeah.

Shibani Ahuja: And so I think like right sizing expectations like you said, whether it's on ROI or the guardrails or the security, the trust, the governance that you put in place in terms of if I put a mode four agent in the wild, I certainly want to know that there are appropriate guardrails and that there is separation of duties and that if it goes bump in the night, there's clarity of who owns that agent. But that's not how you treat assistive AI that is meant to be for the masses. So that is exactly what we're doing is that's why a mental model, I've created a mental model for ourselves to make sure that we are taking the appropriate steps as we look at the spectrum and the full spectrum of what AI is in our company.

Jason Hiner: Excellent. All right. Y'all launched your own model. Very interesting, very cool. Koa is the model that Salesforce did built on NVIDIA's Nemotron. This is so interesting because it's one of the most interesting conversations I've had over the past couple months with enterprises is enterprises, there's a few of them that have had this light bulb moment where they're like, we need to own our own AI. This is our own intelligence. This is the most valuable, this is our crown jewels. This is the most valuable part of our business or will be and we can't just sort of outsource that to somebody that also our competitors are using and that we don't necessarily know who controls the data and all of that. I've been having this conversation and I'm like, oh yeah, that's really interesting. Thomson Reuters made their own model. CrowdStrike a couple of weeks ago, they announced their own AI model also based on Nemotron, just like Salesforce's. Then of course I get here and one of the first things I hear, we're launching our own AI model. I would love your thoughts on that because it is so timely and also if you would have asked me six months ago, like in our conversation in February, seven months ago, if I would have asked you or you would have asked me, what about us doing our own AI model? It would have never, it would just not have even been in the realm of possibility, but a lot has changed and I think it shows the acceleration of things and Nemotron 3 Super that wasn't even released yet that this model is built on, which shows how banana is like the timeline is right now. Your thoughts on what does that enable for Salesforce, being able to have its own model that is specialized for you and your customers and your business?

Shibani Ahuja: The way I'm looking at it is it's kind of following a pattern of what we're doing with all of our products, like this whole notion of headless. We're creating options, but it's a choose your own adventure. No different than we've said, choose your own lake, choose your own warehouse, choose your own applications, choose your surface. This is just one more that we're saying, choose your own model. We also have, like you also probably saw the spectrum of agents that we've introduced. We've got agents that you can, we've got the platforms where you can build agents as you wish or you can get agents like we've qualified or finned. These are specialized agents. So now we have specialized models that will allow customers to just have optionality. That's really what we're trying to do.

Jason Hiner: Hey, everybody, thanks for listening to this episode. Quick note, and then we'll get you back to the conversation. We love bringing you this content every week. We're always trying to figure out how we can deliver you the most value to help you understand how AI is transforming business and transforming the world. If you're enjoying the show, there's an easy way for you to give a little value back and help others learn about the show as well. If you're on Apple Podcasts or Spotify, drop us a rating and leave us a review. And if you're on YouTube, hit like, subscribe, or leave us a comment. That's it. It only takes a minute, but it's a huge help. So thanks in advance for pitching in. And now, back to the show. One of the interesting things that's come up is that the capabilities of AI have advanced enormously in recent months. You see more people actually using it in business, which is great. I was in a meeting this weekend with a teacher, a head of a nonprofit, a professor, and an artist. I did not even bring up AI. One of them mentioned, "Oh, hey, you do something with AI, right?" Then all of them wanted to tell me all the things they've been doing with AI. It happened with my neighbors too. I'm like, this is crazy. It is mainstream. It is here. We're living this. It's the topic everybody wants to talk about, which is amazing. Is it so helpful?

Shibani Ahuja: Yeah. It's like, I think that there were folks at first. We've gone through this change where at first people were like, I don't know what this is. To then going, I feel like I'm cheating by using it. Now going, I can't, like, this is incredible. I can't live without this. I can't live without Slackbot. Slackbot keeps my head on. Slackbot, where am I going next? Hey, who am I speaking to next? Slackbot is everything.

Jason Hiner: What did Slackbot prep you for for this meeting?

Shibani Ahuja: I didn't need any prep, Jason. I was like, Slackbot, I'm excited about meeting with Jason. That's what I said.

Jason Hiner: You're so good. You're so good.

Shibani Ahuja: I happen to, well, you said that I was memorable. I have to stay memorable.

Jason Hiner: That's right. All right. Very good. Very good. So this, being able to do your own model, maybe, is it showing for like the second time in my life? So the cool thing about having your own model though is like, when you put a lot of the stuff out there, there's like, there also, now we're getting to the point, it's so useful that now there's sort of like big questions. Like, okay, what about our data? Like, what data is it, are they training on our data? Are the, is it going to hallucinate, you know, because based on this, because if it does, like, this is really important business critical stuff, like, I'm going to be in trouble if it's hallucinating. And we've all had the thing like, where you ask it something, and then the next time you ask it, it tells you a different one. You're like, no, I wanted you to do it just like you did the last time. And it's different, right? So by having this like domain specific model, train on 30 years of business processes, what you might call business wisdom, or learnings, or best practices, then it's eliminating a lot of this stuff that might get you in trouble, right? But world knowledge, you don't need that for sort of business solutions. But it's also then injecting a lot of this like expertise and best practices around processes and that that feels like the sort of the game changer. Is that how you think about it? Because you get one better accuracy, it's going to use fewer tokens because it doesn't have to use a big, you know, enterprise frontier model. And then also like it's more, it's more accurate and less likely to hallucinate. Like, is that how you think about it?

Shibani Ahuja: The way I would look at it is, was it about like a year ago, or two years ago, a lot of folks came out with like the model is the thing, the model is the most important thing. And we as us as an organization, what we said was an LLM is not enough, the model alone is not enough. Right? I think that even for us thinking that just us creating our own model is going to be the silver lining and silver bullet, no, it's actually like, it's the harness. I think you might have heard Rohan talk about the harness. Yep. It is the model in conjunction with the harness. So what's the harness to me is like you, what you just talked about was like probabilistic behavior models by nature on their own have a probabilistic nature to them. To bring that determinism is when you get the model plus the workflows plus the applications plus the data layer that has the metadata, the semantics, that's when you start to get something that's contained that's more that can be that can reason and be probabilistic when you need it to be, but deterministic when you have no choice. So I don't know that the model alone, it is the model plus the harness. So now we're just saying like, we're fine tuning and fine tuning and fine tuning to make sure that our customers have confidence in what we are deploying.

Jason Hiner: Very cool. What's next? What do you wish that AI could do for Salesforce and for your customers? Like, what's the thing that you can't wait until AI can do?

Shibani Ahuja: Jason, you're going too far. What's next? What do I want AI to do for me? Plan my next vacation. Okay, that's what I'm going to leave it with.

Jason Hiner: That's what I'm looking for. So smart. Very good. I mean, can Slackbot do that? Actually, it probably does. Yeah, really. Slackbot knows everything about me. Oh, all the context. It knows what we're doing. You're a wild adventurer

Shibani Ahuja: who wants to find something remote in the jungles of nature. You should go do this, and here's the best time to do it because I've accessed your calendar.

Jason Hiner: I'm just going to say, like, that's most important: these weeks, not these weeks, you know? Yeah. Very good. Oh my goodness, I'm going to have to try that with Slackbot. It does know a lot about me. I've used it for a long time. So good. What else should I know, or what do people not know about Salesforce and its journey with AI that may be underestimated, that I should make sure that I understand?

Shibani Ahuja: What I would say is, we've talked a lot about technology. We're a tech firm, right? But I think a space that's worth noting is actually our professional services. Agentic AI is about as old as I've been at the company—22 or 23 months, really. Organizations like Salesforce have been close to the successes and failures of our own launches and deployments with customers. A lot of folks are starting to say: once upon a time, I might have wanted someone independent from Salesforce to come in and do the implementation. But CIOs and C-suite leaders are saying, Salesforce, you are closest to seeing what great looks like. You're closest to understanding where the opportunities are and where people have fallen flat on their faces. The professional-services offerings we have—the forward-deployed engineers—that's what's becoming a game changer. Before we even talk about the technology, we're starting to talk to customers: Do you have the right operating models or structures? Do you have the right mindset? Do you have the right leadership skills? Do you have the right processes? Are you thinking about AI in these modes? The technology is the cherry on top, but we are bringing the patterns of successful customer behavior from the last 22, 23, 24 months to the table more and more.

Jason Hiner: And that helps Salesforce too. Like you get to see all these best practices. So that's very cool. All right, while we meet at Dreamforce 2026 next year, what do you hope we're talking about when it comes to AI next year?

Shibani Ahuja: Oh my goodness. Jason, I think I may have even used this line last time. Like people ask me to predict the future all the time. And I go, I don't know, you're giving me too much credit to be able to predict the future. Here's what I would say. Like, you know, my focus as I look into the future is, you know, there's EQ, there's IQ. Focus on AQ, the adaptability quotient. And I think I may have said this last time too, like the adaptability quotient to me is not just in mindset, but it is absolutely a mindset, but the adaptability quotient of your tech stacks as people are starting to move from build versus buy, size, SaaS versus hyperscaler, now moving into the realm of like composable components separating the UI from the underlying data. The advice that I have for customers is start to think about the adaptability quotient of your tech stacks and how well you've got partners in your ecosystem that are playing well with one another. You could say playing well in the sandbox.

Jason Hiner: Yeah, yeah. AQ. I like it. I like it. It's so smart. And it's so, you know, for every sphere of life, like right now, it's so critical. Very good. Shibani, thank you. So good to have you on the show again.

Shibani Ahuja: A three-peat.

Jason Hiner: I love it. I love it. Thanks so much for your time.

Shibani Ahuja: Thanks, Jason.