In this FNTV interview, Salesforce’s David Fan offers a grounded, practical look at how agentic AI is redefining enterprise workflows. Speaking with Steve Saunders, Fan unveils Agentforce, Salesforce’s digital labor platform that blends AI agents with CRM and data layers to automate customer service, accelerate sales, and reduce operational costs—especially in telecom.
Fan explains how Salesforce differentiates with industry-specific, trusted AI agents that are easy to deploy and safe by design. These agents not only streamline quoting and order handling but also reduce AI hallucinations through data grounding and human-in-the-loop safeguards.
With telecom leaders urgently looking to drive margin and revenue, Fan makes the case for Agentforce as the pragmatic, vertical-specific solution that aligns with today's commercial realities—empowering sellers while preserving trust, speed, and control.
Steve Saunders:
Hey, David. There is a lot of hype around AI, obviously, but also Agentic AI at the moment. I wondered if you could share how Salesforce is approaching this technology right now.
David Fan:
Last year, we introduced Agentforce, our digital labor platform at Salesforce. This is our Agentic AI layer on top of our data and CRM applications. What it does is it leverages conversational AI to really transform how business operates. If you can think about it, essentially lets our organization build intelligent, trusted, customizable AI agents. These agents work with humans together and combine, they drive more efficiency and enhance the customer interactions.
Steve Saunders:
What sets you apart from everybody else who's doing Agentic AI in business processes at the moment?
David Fan:
Yeah, I think what sets us apart is these agents, the way that we design them, they're easy to use. They're not just technology. They're fully embedded into something that is manageable and deployable. We ship pre-built agents with skills and actions, so that you can easily stand them up to solve customer questions, like look up a order information, verify customer information, summarize a case. Those are very standard use cases that we see, but the way that we approach that, it's very differently. Agentic AI, specifically, the way that we build our Agentforce is built on the foundation of trust. I think for enterprises, they want to make sure that when the human agents are using AI, information is protected. So, when you send information to the LLM, the information in a way that it is not retained and for future training purposes. At the same time, the conversation coming back and forth, you need to make sure that there's no hallucination, there's no toxicity.
Steve Saunders:
How do you deal with hallucination? Because I think that there's a misconception that hallucination goes away if you do enough training. It doesn't. It's a factor of AI structure really in the way AIs work. Is it that you just make sure that you still have a human being involved in the process?
David Fan:
Yeah, I think the accuracy comes in multiple parts. I do think that when you ground the agent, you'll get more accuracy. I think depending on which LLM you use, there's always some variance coming from it, and you want to make sure that those variants are filtered. So, in a sense that if there are responses that are not coming back the way that you should be, you need to have a way to be able to filter them, and this is where the first layer does.
Steve Saunders:
In the telecom market, is that a big opportunity here?
David Fan:
I think that's a huge market. I think by the nature of the communications industry, it is a hyper-competitive market. It's very fierce. I think everybody is trying to improve the EBITDA to drive margin. Now, I think where AI can help is really come in to help the service provider. One is to grow revenue, second is really to reduce the operation costs. In terms of grow revenue, AI agent can also come in to help the sellers, especially in the B2B SMB space, where the seller needs to be able to respond to customary requests, to be able to generate quotes very quickly, to be able to deliver the service very quickly, so that they can receive the revenue very quickly. That is a big use case for that, as well.
Steve Saunders:
And that is absolutely the key right now. That's one of the biggest trends, not a technology trend, but an attitudinal trend in the carrier market right now is, hey, we're not going to save our way to the end of the 21st Century. We've got to work out how we can actually efficiently make more money and sell more products.
David Fan:
Yeah, for sure, for sure. I still remember, one of the customer tells us, "What is the best experience for sellers when it comes to creating a quote?" It's not about going through a bunch of flow screens to capture what the customer need, configure what the customer need. The best experience is for the seller to be able to show that a machine and say, "This is what I want. You do it for me." I think we are getting into this by leveraging Agentforce agentic experience to be able to take this conversation, take this voice to be able to convert into something that is meaningful for the seller, as well as for their customers.
Steve Saunders:
It's refreshing for me to hear you talk about this in such a pragmatic way because the thing which is different here is that there's still a human involved in that sales process.
David Fan:
From our product planning perspective, we will be launching a number of pre-built agents with skills and actions that our customer can use right out the box. So, these actions will be industry-specific. So, think about these digital workers. They know the industry really well, so when you tell them that when they looked up information, they see the data with a bunch of numbers, the agent will know that this is actually a phone number or this is actually a SIM card number, so that metadata is really important.
Steve Saunders:
And you do that on a vertical basis. You do that industry by industry. That's incredibly important, isn't it? Because actually, telecom, which is what I do, is the layer underneath the vertical industries, but you have to have that bespoke knowledge.
David Fan:
This is all done in a very easy administration UI, as well. Any Salesforce user will be familiar with the experience. They can go in and create tests, deploy them very quickly.
Steve Saunders:
David, it's a pleasure to talk to you today. It's very exciting. Obviously, this is a huge opportunity for Salesforce. You're sitting on top of this huge customer base, but also gigantic data set, and it's almost like AI is giving you this shim layer to pull the value out of that for your customers. Fantastic. Wonderfully explained, as well. Thank you so much.
David Fan:
Yeah, thanks for your time.