Zack Terry

But they have completely overhauled the Agent Force Builder. So this is where you, as an admin or developer, are going in and you’re actually creating the agent. You’re writing the instructions, you’re creating the topics and the actions, and you’re telling the agent what to do when questions are asked or how to go and retrieve information. This whole experience has been completely revamped. It is, I think, a lot more intuitive. It looks more like a working document or a canvas, if you’ve ever used something like a Slack canvas or Quip. It’s very similar to a Quip document, where you’re writing things in plain language. And what’s really interesting is you can interact with Agent Force while you’re building an agent in Agent Force, to have Agent Force help you create an agent in Agent Force.

Eric Housh

Hey, welcome back to AI in Action. I’m Eric Housh. As always, joined by Fast Slow Motion’s Director of AI, Zack Terry. Zack, how are you today?

Zack Terry

I’m doing great, Eric. How are you today?

Eric Housh

Doing well, dude. Dreamforce ’25 in the books, just wrapped up. This year’s theme was loud and clear: the agentic enterprise has arrived. From Salesforce’s partnership with OpenAI to Agent Force 360’s new capabilities, this event redefined how AI will live inside the tools we already use.

Zack Terry

Yeah, this Dreamforce wasn’t just about shiny demos. It was also about renaming every single product to use the word “Agent Force.” So I’m being a little bit facetious, but they really did rename almost every product to use the word Agent Force, which I think we may have predicted a few months back. But this particular Dreamforce was all about connection, so it’s really connecting AI agents and data and voice capability, and people, and really putting everything into one system. And so it really goes from last year, where Agent Force was introduced, which is crazy to think that was just one year ago, to where we’re talking about just this whole tool. It’s very new, it’s very novel, we’re all trying to figure out how to use it, to now that same tool being integrated. And I say the same tool, we’re on the fourth version of Agent Force, which we’ll talk about more as we get into it. But in one year we’ve seen four different iterations. But we’re taking that tool and we’re really weaving it throughout the entire platform. Every single formerly-known-as-cloud that Salesforce has is using Agent Force in some way. So we’ve got a lot to talk about, but before we do that, Eric, what do you say we talk about some news items?

Eric Housh

Yeah, as always, never a slow week in the world of AI. I’ve got a lot to talk about this week. So yeah, before we get into Salesforce, let’s hit some of the headlines. Adobe launches the AI Foundry for branded models. What does that mean for brands?

Zack Terry

So this is pretty interesting. Adobe is jumping into enterprise gen AI with a new service that they’re calling the AI Foundry. And the idea behind this is that companies can now train custom models on their own IP. So images, audio, video, graphics, things that belong to their brand voice, things like that, they can now generate on-brand content at scale using generative AI.

Home Depot and Disney are already a part of this, and specifically it’s the Disney Imagineering team. So they’re using that for some pretty interesting things, I imagine. But definitely a clear signal that enterprises want to own their AI and not just rent it. And also, this is pretty interesting: I pulled up an excerpt from TechCrunch, because I was wondering what is so different and unique about this, right? And according to TechCrunch, this is an offering that allows enterprises to work with the company to build custom gen AI models trained on their branding and intellectual property, which is a very different selling point than already-trained AI models that are grounded on your company data. So what it sounds like is Adobe is empowering enterprises to actually train AI models, like little AI models, on their brand and their information. So I thought that was interesting. I’m not exactly sure what that means, or if that is in fact how that’s working, but definitely an interesting update. And I think if you’re on the Adobe suite, it’s going to be an exciting thing to experiment with.

Eric Housh

Yeah, it seems like a big move, especially for these firms looking to protect their IP. Protecting their brand voice will be absolutely something that we keep our eye on.

Zack Terry

Absolutely. So rolling into the next news item here, Andrej Karpathy, if you’ve ever heard of him… yeah, silent J there.

Eric Housh

Well done on the pronunciation there, by the way.

Zack Terry

For sure. But Andrej, if you’re not familiar with who he is, he is a computer scientist. He formerly served as the Director of Artificial Intelligence and Autopilot for Computer Vision at Tesla. He co-founded and formerly worked at OpenAI.

He has a whole list of things he’s working on today, including Eureka Labs. It’s a new AI education company that he’s working with. So, quite a big name in the world of artificial intelligence and large language models. Well, on a recent podcast, he said that true agentic AI is still a decade away. So very interesting to see one of the godfathers of AI essentially cooling that hype a little bit, taking it down a notch.

So he is arguing that we lack multimodality, computer interaction, and continual learning. And so he does not expect fully autonomous agents anytime soon, which is a very interesting reality check for any leader planning next year’s AI budget. And I think this is a very cogent take on where we’re at in the current landscape of generative AI and large language models. I think he’s probably right, because we are going to need more tech. We’re going to need advancements in how generative AI works. And it may be a totally different mechanism than the transformer and what we’re using today that enables us to get there. But I do think a lot of the points that he argues here are really good. And I mean, I don’t know if 10 years, a decade, I don’t know if that’s the right timeline. But I think the points he’s making, around we are gonna need to make more advances in order to get to truly autonomous agents, I think that’s absolutely true.

Eric Housh

Yeah, I’m on the same page as you. I think maybe we do need to tamper down the hype cycle a little bit. I think 10 years is a little hyperbolic. I would be surprised if it were that long before we got truly agentic AI out there, but definitely interesting to hear a voice, especially him, with so much credibility, come out and say, “Jets, cool your jets, we’ve got some work to do.” Let’s talk about the FTC.

Zack Terry

Yeah, we got the Federal Trade Commission in the house today, Eric. That’s right.

Eric Housh

Let’s go.

Zack Terry

So Wired reported that the FTC took down hundreds of AI policy blog posts from Chair Lina Khan’s tenure. So this is pretty interesting. So I think it’s raising some questions around transparency. Is there a regulatory tone shift going on in the United States around AI and its use?

I think for companies evaluating AI or using AI, it means that governance and compliance rules may not be everything that we thought they were. Things may be shifting or changing. I don’t know in which direction things are shifting or changing right now, but the fact that the FTC is scrubbing a lot of that from their site is certainly something to keep an eye on.

Eric Housh

Definitely feels like the rule book is still being written there. Taking down blog posts is an interesting tactic though, because you’d assume, especially with all the models that are being trained, that this content is out there somewhere. It really is. So I know.

Zack Terry

The internet is forever, Eric, so…

Eric Housh

I took down that blog post. Wasn’t me. Anyway.

Zack Terry

It’s on the Wayback Machine somewhere.

Eric Housh

Yes, there it is. The LLMs are training on the Wayback Machine. We know that.

Zack Terry

That’s right.

Eric Housh

Let’s get into Dreamforce, the headline everybody’s talking about. Wow, two Goliaths, OpenAI and Salesforce, hand in hand, partnership. What does this mean?

Zack Terry

This is really a continuation of an already-existing partnership. So it’s taking what they were already doing and it’s just really cranking it up to 11. So we were already working… OpenAI was already working with Salesforce. That was already what was powering the Atlas reasoning engine for Agent Force. That’s how the agent is actually making a decision on: well, what am I being asked, and what do I have the capability to do? And that’s what it was actually using to go and do that. Now that’s opened up a little bit. You can do that with Anthropic now. You can… I think yesterday they announced that you can start running that on Gemini. So some of that is expanding, but specifically there were a bunch of announcements related to OpenAI and Salesforce, including Agent Force 360, which is just now the word for Agent Force. So that’s sort of the whole Agent Force platform now is called Agent Force 360. Just think about it as Agent Force, right? So Agent Force is now going to be able to be inside of ChatGPT. So that will bring Salesforce CRM data and Tableau insights that you could surface directly inside of ChatGPT chats. Now, I imagine that this is coming on the heels of MCP and agent-to-agent, and being able to actually leverage connectability and interoperability into those systems. So I don’t know exactly when this is coming, I don’t think it’s there yet, but you should be able to connect to Salesforce at some point inside of ChatGPT and be able to surface CRM data and ask questions about your CRM data, which is really cool. They are bringing commerce, Agent Force Commerce, which is Commerce Cloud renamed Agent Force Commerce. So they’re bringing instant checkout to Agent Force Commerce, which is the ability for merchants to sell to ChatGPT users while Salesforce handles secure orders and payments. ChatGPT is coming to Slack. So ChatGPT will be able to summarize threads, draft messages, pull insights, without ever having to leave Slack. So really weaving that even further into the systems that Salesforce has to offer, including Slack, which… really, you’ll see this throughout all of the Dreamforce marketing. They are positioning Slack as the agentic operating system for everything in Agent Force. So they really envision Slack as sort of the central location where you’re interacting with Agent Force, which makes a lot of sense, because that’s where you’re chatting with your coworkers. And by extension, if I’m just able to chat with agents to have them do things for me, it makes a lot of sense.

All right, two more items here related to OpenAI and their partnership with Salesforce. Codex is coming to Slack. So Codex is the developer-specific coding agent and coding capabilities that ChatGPT has. And so now developers, when this gets generally available, developers will be able to tag @Codex inside of Slack to then generate or debug code on demand. So bringing some interesting developer tools directly inside of Slack with Codex.

And then finally, OpenAI is also going to be directly inside of Agent Force 360. So ChatGPT-5 is already available, that model has already been brought into Agent Force, you can use that inside of prompts now. And it’s going to become the preferred model for the reasoning engine itself. So not just for prompt templates and running, you’ve always been able to select a bunch of different models there. GPT-5 is available now, I’ve seen that already. But we’ll be able to use ChatGPT-5 as the preferred model for the reasoning engine itself, which should be an upgrade in the reasoning capabilities, in terms of what that model is able to do and how it operates. So I think we’ll see some additional efficiency and hopefully some expanded capabilities by upgrading that reasoning engine to the latest model. And you know, if the rumors are true, maybe that’s gonna be GPT-6 by the end of the year. I don’t know, we’ll see.

Eric Housh

We’re definitely on that path. Feels like… let’s stay on Agent Force 360 for a bit. Dreamforce is always about the platform evolution. What’s going on with the platform? What did we see around Agent Force 360 at this year’s event?

Zack Terry

I’m very excited about this, as someone who works with Agent Force almost every day. The speed at which Salesforce has been pushing out updates to this platform is wild. So we mentioned this at the beginning, this was a year ago that Agent Force was announced, and we’ve gone through four different iterations of this toolset since then. So we’ve got four major releases for Agent Force within a one-year timeframe. And this latest iteration, which again they’re calling Agent Force 360, I kind of think about it as Agent Force 4, because it’s the fourth iteration here. But they have completely overhauled the Agent Force Builder. So this is where you, as an admin or developer, are going in and you’re actually creating the agent, you’re writing the instructions, you’re creating the topics and the actions, and you’re telling the agent what to do when questions are asked or how to go and retrieve information. This whole experience has been completely revamped. It is, I think, a lot more intuitive. It looks more like a working document or a canvas, if you’ve ever used like a Slack canvas or Quip. It’s very similar to a Quip document, where you’re writing things in plain language. And what’s really interesting is you can interact with Agent Force while you’re building an agent in Agent Force, to have Agent Force help you create an agent in Agent Force. Yeah.

Eric Housh

Super meta, super meta.

Zack Terry

So it is, it is literally like meta-prompting. So you’re able to actually have the agent help you create the instructions and the topics that you want to be able to leverage within that agent. And this goes into another announcement: there is a new scripting language, a proprietary scripting language that Salesforce has developed, called Agent Script, which can actually sit underneath what you create in the Agent Builder. So if I type something in natural language, it’s actually generating script behind the scenes to tell that agent what to do. And so what this does is it actually injects a lot more determinism into this process. And that’s historically been something very difficult to achieve with agents and large language models, and with Agent Force specifically. If you wanted it to do something deterministic, you had to do some interesting workarounds, like hiding certain capabilities until some variable gets populated.

Well, now with this scripting language, it allows us to create much more robust if-then logic. So it’s really combining the usefulness and the unpredictability of a large language model with more deterministic logic through this scripting language, which is a big benefit. But because we have this scripting language now as sort of the bedrock of our agent, Agent Force itself, as a tool to help us build the agents, has the ability to generate that script for us. And so I can take an instruction and say, “It’s not quite doing what I want it to do, how do I get it to do this other thing?” And it will just write that script for you and populate it directly inside the builder. So I really think it’s going to speed things up and help make things more reliable over time. Anyway, that’s the whole thing about that, is just go check out some demos and some screenshots of what this builder looks like, because it really is very different than what we’ve been used to.

And then, in addition to that, we have Agent Force Vibes, which is… it’s a vibe-coding app. It’s really just the IDE. They’ve got this ability to essentially leverage VS Code, but from a browser. And now they’re injecting the ability to use natural language to have Agent Force create things for you. And so if you’re a developer, this is kind of like if you were using Cursor. It’s specific to Salesforce, however, and I believe you get 50 premium requests per day per org. And then after those 50 are used, it brings you down to a Salesforce proprietary model. So it’s a little bit more limiting. So that’s exciting, that’s really interesting. I think people will be able to build some things. It’s also maybe a little bit dangerous, because they launched it directly in production. So, you know, if you’re on Salesforce and you don’t want people vibe coding in production, you might want to limit access to that, or at least make people aware of it.

And then finally is observability. So we have some new dashboards and a lot more reporting capability around agents, around topics, around actions, around tags, around utterances, and things that are happening. So a lot of stuff that was hard to report on previously is now, out of the box, going into Data Cloud and getting populated on some pretty robust dashboards. It’s in beta at the moment, I think it’s targeting GA around November. But if you’re using Agent Force, it’s a public beta, so you should be able to enable it and check it out. And I’ve looked at this for a couple of clients, and it is really, really good. It adds just so much more information directly at your fingertips that you were not able to really see before, without a significant amount of development or a bunch of other work that you would have to do. So definitely check it out, I think you’ll find it very useful.

Eric Housh

So Salesforce continuing to think enterprise-grade. Move fast, let’s roll out these enhancements to our clients, let’s make this Agent Force 360 really something that these larger clients can leverage. We talked about the new builder, we talked about Vibes. One thing we didn’t hit on yet is Agent Force Voice, what’s up with that?

Zack Terry

Yes. So the very long-awaited voice capability, I think Salesforce has been talking about this, well, frankly, since last Dreamforce. So there was a whole demo last year where I think it was Saks Fifth Avenue, maybe somebody calling in, and an agent actually handling that whole conversation, right?

Eric Housh

Yeah, I recall that.

Zack Terry

Yeah. So very much a part of the initial launch of Agent Force, but the capability was not there. So now, a year later, we do actually have voice capability with Agent Force.

Eric Housh

Wait a minute, wait a minute, you’re telling me the demo was staged last year?

Zack Terry

I would never make such a claim. Let’s say it was streamlined, Eric. It was very streamlined.

Eric Housh

There we go. You never do a live demo, everybody knows that.

Zack Terry

That’s right, that’s right. Well, now, now it’s real. So this is a real product and a real capability that you can bring to Agent Force agents. And it is voice-native AI assistance. So I can essentially provide voice audio as an input, and then when the AI gives me a response, it is also a voice. So it’s generating audible responses to me. So, of course, the use case there is I’m going to get on the phone and I’m gonna chat with an AI, and the AI is able to understand what I’m saying. And so it’s changing this whole concept from decision trees and IVR systems, where I’ve got to pre-program everything. And it’s like, well, if you wanna talk to customer service, press one, and if you wanna do this, press two. Changes that whole deal to where it’s more like a natural conversation, and it’s able to interpret that. So that’s one piece of voice. The other piece of it is not just with phone conversations, but you’ll be able to use like a microphone on your computer or on your phone and interact directly with Agent Force. In some ways, we’ve always been able to do that, because we can use our phone’s transcription to transcribe our voice and then put that into the chat window, but this is different, I’m not transcribing it, I’m providing that audio file, that I’m getting audio back, and so it’s a new way to interact with it. And a lot of this is still on the roadmap. So I do believe the voice capability for a phone is available, but the capability for that agent audio is coming a little bit later. But it’s something that people have been asking for for a long time. I know I have several clients that have waited to implement Agent Force until voice was a capability. So I think we’ll see a lot more people adopting Agent Force that have really been waiting for this specific function. And now it’s here.

Eric Housh

It’s how their clients want to interact, right? We don’t want to call in and dial one and dial seven, we want to talk to someone that understands us, ultimately. So you gotta get that right, that’ll be interesting as our clients start to roll that out, how we help them build that. Zack, moving on, we heard a lot about Agent Exchange and A2A. What do we need to know there?

Zack Terry

Yes, so the Agent Exchange is essentially like the App Exchange, but for Agent Force. So it’s where developers can create and distribute applications specifically for use with Agent Force. And a lot of what that means is using MCP, model context protocol, to connect to other systems, so that the agent can interact with those other systems.

Well, Salesforce is taking this a step farther, because they’re also bringing A2A, which means agent-to-agent. And that is another protocol, but this particular protocol is the ability for agents to essentially communicate with each other. So if I have an Agent Force agent that needs to reach out to an agent on a third-party platform you’re maybe working with — let’s say PayPal — and I need my Agent Force agent to communicate something to PayPal, and then that PayPal agent needs to go and take the information and actually perform something, and return the output back. It’s essentially a way for these agents to handle those conversations kind of within themselves, right? And so it’s not having to, you know, create a stop-gap in between. Really, Salesforce’s vision here is to be able to connect with anything.

And so they envision a future where Agent Force agents are communicating with other agents, they’re communicating with other servers, they’re getting information into the system. And so by bringing both MCP and A2A to Agent Force as native capabilities, it really is opening up a lot of interoperability between Salesforce, Agent Force, and a number of other platforms that also operate on these protocols. The idea, I think, is really about orchestration, not just within Salesforce, but with any other tool that you have, so that you can collaborate with those tools. Agents can leverage those tools, you can share context from one tool to another, pass tasks back and forth, and really kind of have these agents sort of working as teams.

So, kind of going back to the comment from Andrej Karpathy, where we’re not gonna see truly agentic automation for a decade, well, this is interesting, because these are the pieces of the puzzle we need to create that type of automation. Now, will we see the full vision of this in the near future? Maybe not. It’s gonna take a little bit of time for this to mature and for systems to be able to connect, and not just connect, but do it securely, and in a way that’s trusted and is not risky. But this is how we get there, right? So we’re putting those pieces in place to enable Salesforce to be at the forefront of agents communicating with other systems and with other agents to accomplish tasks outside of just the CRM itself.

Eric Housh

Very cool stuff. Let’s wrap it up, Zack. Let’s talk about one final thing: Setup, powered by Agent Force. What is this?

Zack Terry

So if you have ever been to a True to the Core session at Dreamforce — and if you haven’t, I’ll tell you what it is — True to the Core is where customers and partners can get in a live forum with some of the biggest decision-makers at Salesforce, executives, product managers, and ask questions and surface issues. And it’s a pretty cool forum, because it’s real time, people are not, you know, pre-prepped with what they’re allowed to bring or not. So it ranges from really simple requests to diatribes of complaints. And what’s really interesting is Salesforce often takes that feedback and incorporates it directly into their products. So I think for several years, True to the Core has had a lot of complaints and a lot of issues with the setup experience. So this is for admins: when you go to the back end, and you’re managing users, you’re managing metadata, you’re creating objects, you’re creating fields, you know, the whole process by which you actually manage and administrate Salesforce is happening in Setup. The complaints have been, it’s archaic, it hasn’t seen a lot of updates. We did this whole update to Lightning, but Setup is still in Classic. And so over the past few years, we’ve seen a lot of updates being brought to Setup. So we’ve got Salesforce Go now, with all these streamlined walkthrough versions of how to set up different features and clouds. We’ve got all the user summaries and access summaries, where we’re able to see things at a glance, instead of having to look through a needle in a haystack and just find the one permission that’s missing. So all of that has led to this, which is now an Agent Force experience for Setup, which is an agent that you can chat with that can actually do things for you on the back end.

So it can query information, it can figure out why somebody has access to something but somebody else doesn’t have access to it. So it’s looking at permissions. It can help you fix formulas, it can help you create flows, it can help you manage users, reset passwords, all kinds of stuff like that, right? And the idea is that that capability will continue to grow. So it’s in closed pilot right now, but when this becomes available, it will essentially let admins run a lot of their day-to-day activities through Agent Force, using natural language, as opposed to having to use clicks throughout the system, which I think will be really interesting. So my expectation here is it’s gonna be probably cool, but not incredibly powerful initially. I think over time this will become one of the primary ways that admins interact with Setup, because a lot of the things that take us dozens and dozens of clicks, if I can just type one sentence and get an answer, I think that’s a great experience. So I expect this to be something Salesforce continues to focus on and really builds out over the next several releases.

Eric Housh

Should admins feel threatened by that?

Zack Terry

That is an interesting question, Eric. So…

Eric Housh

You…

Zack Terry

I think no, because the ability to have Agent Force do some of these things for you does not remove the need to understand the data model and make good architectural decisions. We are still gonna need people to be able to drive those decisions and drive the build process. Now, if Agent Force can help me create a custom object with a bunch of custom fields in a matter of seconds to minutes, as opposed to me having to go through that click path and spend an hour or more doing that, that is a huge advantage for me as an admin. But the value is not in my ability to do those clicks, the value is in my ability to understand why I need that object and why I need those fields, and the right data types for that, and how that fits into the rest of the system. And two different admins with completely different skill sets could get completely different results interacting with the same agent, right? So I think ultimately it’s about augmenting our admins and giving them more capabilities to do more with what they have. But I don’t think it is gonna get rid of the need for good admins, because ultimately we’re still making those decisions, and we still need really good architectural understanding and guidance.

Eric Housh

Great point, great point. All right, wrapping us up: four things leaders need to do right now that Dreamforce 2025 is…

Zack Terry

All right, first up, sign up for the Agent Force Builder beta and go test out the new builder. Check it out. Check out Agent Script and just use it on a low-risk workflow. Throw it in a sandbox and just check it out. I think that’s one of the most compelling things to do, is sign up for that beta, check it out. The second piece is watch out for voice pilots and use cases, and people building things with Agent Force Voice. Just check out what’s going on, and if you have a use case, consider getting Agent Force Voice and maybe building out a pilot to see how it works. Third, and this is really important, audit your data structure if you haven’t already. It’s a little far into the conversation to just be thinking about that, but if you haven’t, it’s not too late to start, right? So really make sure you understand your data, make sure you have good data, make sure you have good processes. Data 360, which is the new name for Data Cloud, that is really becoming like the plumbing for all of Agent Force and Salesforce and how everything works together. And so having really good data for Data 360, AKA Data Cloud, to work with is going to be really important. And then the fourth thing is, if you’re using ChatGPT at all in your business, think through how you might want to leverage connecting that with Salesforce, if you want to at all, right? If you get utility from working with ChatGPT and being able to pull in Salesforce data and ask questions, that may be a great thing to look out for. Now again, I don’t think this is available yet, but when that goes live, if you’ve already been thinking through a strategy for how you might wanna work with that, you’ll be well-positioned to really hit the ground running when that’s available.

Eric Housh

I think Dreamforce 2025 drove one point home for me: AI is no longer an add-on, it’s at the core of how we work, it’s at the core of how Salesforce is going to innovate over the next foreseeable future. But there’s a lot to digest. So really want to invite our audience, if you want help mapping any of this out, understanding this, going deeper, seeing how this applies to your business, really want to invite you to reach out to us. Fastslowmotion.com/AI, would love to have that conversation with you.

Zack Terry

Exactly, it really is all about experimenting, measuring, then scaling what works. That’s AI in Action.

Eric Housh

There it is. Zack, great time as always. We’ll see you guys next time.