We're back. In our first episode of the Chief Customer Officer Podcast, Jay and Jeff break down how to actually build AI agents for your renewal process—not theory, but the specific agents you'd deploy, what platforms to use, and where most teams get stuck. Jeff shares the four agents he's building at his new company, and Jay maps out the landscape of agentic tools from off-the-shelf to custom-built.

00:00 – Welcome to the Chief Customer Officer Podcast
00:56 – Jeff's new role at Junction: building CS from scratch
02:31 – Why this goes beyond CSMs (PS, AM, Support)
04:24 – The idea: AI agents for the renewal process
05:58 – Agent 1: Contract terms (auto-renew, terms for convenience)
07:00 – Agent 2: Value metrics (are they actually getting value?)
07:53 – Agent 3: Right stakeholder (who signed, are they still there?)
09:05 – Agent 4: Intent to renew (sentiment from calls, emails, Slack)
10:38 – External agents vs. internal/operational agents
12:24 – Why you have to map the process before you automate
15:14 – Jay's 2013 flashback: manually reading every contract
17:07 – Build from the bottom up: granular agents first
19:45 – Guardrails and ripcords: knowing when agents break
21:57 – The Salesforce AgentForce cautionary tale
25:45 – RAG explained: real-time data + private data
28:00 – What tools should you actually use?
31:17 – Four categories of agentic platforms
35:21 – IC tools vs. department vs. enterprise decisions
38:21 – What's next for the podcast

Jay Nathan – CEO of Balboa Solutions and co-founder of ChiefCustomerOfficer.io

Jeff Breunsbach – Head of Customer Success at Junction and co-founder of ChiefCustomerOfficer.io

[First episode—more coming soon!]

The Chief Customer Officer is all about keeping the customer at the center—strategies, tactics, and real ideas in the era of AI that leaders can execute. From practitioners actually doing it.

Jay Nathan (00:01.772)
All right. Well, here we are, Jeff. We've, we've, we've done it. We've gotten the band back together. Welcome everybody. This is, this is not the land and expand podcast anymore. This is now the chief customer officer podcast. So if you were on this channel before, you're still in the right place, but, my old partner, Jeff Brunsbach and I have gotten back together and we decided that there's so much to be talking about in the world of customers, customer success.

SaaS, AI that we needed to, we just couldn't stand it anymore to not be in the fray. And, and so we're here to, start a new series, a new podcast. We're really focusing on how AI is really changing the world of customer experience, customer success, and the role of chief customer officer and all the roles that fall under it. So what's up, Jeff? How's it going?

Jeff Breunsbach (00:56.053)
Not much, you know, I just missed hearing my own voice. So, you know, I felt like I needed to come back and, you know, record a podcast,

Jay Nathan (00:58.83)
Yeah.

Jeff Breunsbach (01:03.401)
once a week and make sure I could, you know, listen to my own voice. But, but no, seriously, it's, it's, I think, you know, you and I have continued to write for a long time in terms of our newsletter and some stuff on LinkedIn. And, you know, I think there's been kind of a void of, of content like this, from like a podcast and video perspective and, excited to jump back in and do it. I think it's a fun, fun avenue to try and explore. I think the other thing that to me stands out is, I just joined a company called Junction.

Jay Nathan (01:06.4)
I'm

Jeff Breunsbach (01:33.457)
to lead customer success and so we're building a lot of stuff from the ground up and I think it's a pretty exciting time to build customer success from the ground up because you know we have new technologies we've got AI that can be embedded in workflows and so I think there's just a lot of interesting decisions I'm trying to make right now about like you know how do I

make sure we can build a great customer experience that leads to renewals and retention and great growth and outcomes for our customers. But at the same time, can I actually do that even more efficiently than we've ever done before? Because there's technology and AI that can be put in place and we can really try and maximize like CSM's time with customers in a strategic way. So I think that's just going be an exciting part of this podcast is trying to bring some of these kind of real situations I might be going through into the light. And honestly, like you and I kind of

we've done this forever of like, you know, kind of talking about this stuff behind the scenes and sharing advice and whatnot. And it's kind of like, Hey, let's just do it in public.

Jay Nathan (02:31.948)
Yeah. Yeah, absolutely. And I'm just to be clear too. It goes well beyond CSMs, right? It's professional services, it's account management, even support teams, which is where you're seeing a lot of success right now, I would say in in agentic workflows that that actually touch the customer, which we'll get into. But yeah, and so just to go back like we've been sending this newsletter out for.

Jeff Breunsbach (02:39.551)
Yeah.

Jay Nathan (02:58.894)
two or three years now, chiefcustomerofficer.io. If you're not already signed up, check it out. We'll be consolidating all this content. But about once every two weeks, we get an email back from somebody saying, hey, like, are you guys gonna do the podcast again? We had a podcast called Gang Bro Retain back in the day. Turned into a community that we really loved and enjoyed interacting with. we're not, that is no longer a thing, but.

You know, we are still in the world and the moment has not passed. In fact, I think AI has given us, but to your point, Jeff, like a whole new world of possibility to explore here in terms of how we do this. It's really given in my mind, it's given customer experience, customer success, a whole new lease on life, right? Because I think some of where it was headed had sort of, it felt like it sort of got stagnant a little bit in terms of.

Jeff Breunsbach (03:27.361)
raised.

Jay Nathan (03:55.65)
You know, what's the role of the CSM? What's the role of professional services? What's all the, what are all these post sales roles? It's like, we were just repeating ourselves over and over. So I'm super excited. I'm super excited about what we're trying to do in my company. What we see happening in other SaaS companies and non-SaaS companies too. So my company works with a lot of non-tech firms, which is pretty cool to see how quickly they're adopting AI and these platforms. And so maybe we touch on some of that as we go along as well.

Jeff Breunsbach (04:24.053)
Yeah, yeah, it'd be awesome. All right, well, I wanted to bring up one thing kind of right off the bat with you, which is, you know, I'm...

joining Junction and we're going through it, just trying to start looking at building a kind of sound renewal function and process. And, you know, that starts even at like, how are we tracking them today? Do we have deal records in the right systems? Are we even, you do we have staging? there, we're kind of, you know, trying to hit some of the basics there. And, and while we're doing that, you know, I think some of the things that I'm starting to think about is where's like AI's place in this aspect and how you think about the motions of renewals and the

maybe the critical pieces of information that we believe lead to a Yes on the renewal and then it just got me thinking like can I go build agents that help identify those key pieces of information so you tell me and Let me I'm this is the first time I'm saying this out loud anybody so like you tell me if you think this is the right way to think about it or if you think there's different ways so I think Right now what I've kind of come back with at least in the again kind of where where we are I think we're trying to just nail some of the basics then I think you can

probably, you know, even go beyond this. But I think there's four pieces of information that I want to make sure we know about on a consistent enough basis for all of our renewals coming up. And so here's the four. One is a couple pieces of information from the actual contract that we have. So do we have an auto renew clause with them and or do they have terms for convenience?

Jay Nathan (05:44.398)
Thank

Jeff Breunsbach (05:58.232)
Those are just two things that I think are critical from the contracting perspective. Like the auto renew piece, you know, might change how we think about the renewal motion and what we're going to do with that customer coming up. And then the terms for convenience, you know, I think that's just something that is kind of a risk, right? It's something that we want to know about whether they're coming up with a renewal or not. So that's kind of the first piece is like those two pieces of information. Could I essentially go build an agent that is crawling our contracts on a regular basis for upcoming renewals and essentially like

auto-populating this into certain fields so that we can see that information inside of our our systems and tools. So that's kind of first piece. Second piece is really about I'm saying are they getting value and I know that's gonna be a hot debate because I'm then gonna say like I'm looking at data for us to understand if they have value. I think some people will be you know I think some people out there might say well you need to ask the customer if they're getting value but I think like

Clearly there's some metrics that we have internally that we should be able to understand if they've gotten value from our system or not. And that could go back into kind of order volumes that could go back to adherence percentages. And so there's some critical value metrics that I just think that we could again, can an agent go look at the data behind the scenes, go look at the data in real time and start saying, okay, know, send volumes are either going up or going down adherence volume.

adherence percentages are going up or going down. like, you know, can we basically have trending and can we have an agent essentially looking at that on a more consistent basis? That's number two. Number three is, we engaged with the right buyer? So, you know, do we have the right decision maker? Is this person the one who's really signed the contract? And so could I have an agent that essentially looks at who signed the contract before?

Jay Nathan (07:37.038)
Okay.

Jeff Breunsbach (07:53.932)
Let's go crawl LinkedIn. Is that person still there and in that role? and then let's go kind of cycle and see other people that we might want to introduce. like, you know, I think.

On one hand, it's amazing what these tools like a zoom info and others have done because it's just giving you access to like everyone's information. But like at the same time, if you go to any Salesforce or HubSpot record in any company, there's just like 75 contacts and you're like, okay, I don't actually know any of these. so could you actually train the agent? Yeah. Like, could you actually train the agent that basically says like, here's who my main persona is. Do we have a connection with that person? Here's two to three other personas that we might want to be in contact with. You know, basically can you triangulate that for us and kind of get it down to

Jay Nathan (08:19.054)
Who actually matters?

Jeff Breunsbach (08:33.465)
to maybe more of a buying committee of four to five people that are the right stakeholders. That's kind of agent number three that we would build. And then agent number four is have they confirmed their willingness to renew? And so clearly that's just a question we could ask on a call that's recorded that.

You know, we could then, you know, can we get the transcript and can we get their answer and can we, can we auto populate that? but even if we haven't asked that question, I wonder if we could train the agent to basically go look at call transcripts, Slack threads, emails, and basically like almost have a deterministic probability. Like is this looking like this person would renew based on these, these types of interactions that they've had with us? You can probably put support tickets in there and whatnot. so anyways, that's a long diatribe, but those are four, I guess agents that I'm thinking about.

kind of attached to our renewal motion that could be working behind the scenes to do all that information gathering for the CSM in order for us to then really surface up like what's the right playbook for us to run on this renewal. I'm curious, you know, what's your first take? there, you know, am I thinking about agents in the right way in terms of like trying to do some of this manual maybe behind the scenes work that CSMs have traditionally had to do?

Jay Nathan (09:44.492)
Yeah. So I mean, I love the four categories. You always do such a good job on like framing things like frameworks, but, but yeah, I really liked that. So there's a contractual piece. There's the value piece. There's the stakeholder piece. Are we communicating with the right people? And then there's basically sort of like a sentiment kind of do it right. intention. Yeah. Intent. so

Jeff Breunsbach (10:04.043)
Yeah, almost like intent. Yeah.

Jay Nathan (10:09.676)
Yeah, this is really interesting. And I think this is what we need to start unpacking when we're talking about AI and this big, you know, this is a, there's a lot of stuff here. A lot of, a lot of pieces of that process. And there's a lot of sub tasks, right? So, you know, I've been doing a lot of thinking and experimenting inside our company with what agents actually are like, what, what do you build? So, and then there's there's, so there's two main flavors of them. Number one is.

Jeff Breunsbach (10:19.893)
Yeah.

Jay Nathan (10:38.838)
A customer experience, an external facing agent, which is something that goes and interacts with the customer. Right. And then number two is something like an operational agent, an internal agent that we build to help the person who's going to go have this renewal conversation to do that more successfully, to spend less time preparing so on and so forth. It's productivity and quality tool. So that's the first two categories I would think about agents. then like, I want to talk about how you actually.

Jeff Breunsbach (11:00.597)
Yeah.

Jay Nathan (11:08.332)
select an agentic platform and tools and all that kind of stuff too. That's this other topic that you and I had on the list. Like that's sort of where my head was on it. But I think first things first, no matter what type of agents you're talking about, what type of process it is, whether it's renewals or whether it's something completely different, right? Maybe it's end of year close for finance. You really have to understand what the current process is before you go innovate it with

tooling, right? That has not changed since the beginning of time. You can't go automate and fix something if you don't know what it is. And so, you know, we we've emphasized this with our clients and even internally with our processes, like we're trying to dissect our sales process right now of how we go get new clients at Balboa. And part of what we're doing is mapping out like, OK, like we've done this for a year. What are the exact steps that we typically go through when when it works well? What are we doing?

every step of the way, because if you don't have that mapped out, trying to throw technology at it is going to be weird. And it's not going to give you the outcome that you would expect. So my question for you is, do you have the renewal process all mapped out yet?

Jeff Breunsbach (12:24.363)
yes. And so yes. And that's where like this list of four things came from is that we documented the, the process of how we essentially get a renewal to close. And then what I go, what I went to go do on that kind of process flow and the process map was essentially look at like.

to your point earlier, where can we go build operational agents? Like where can I go essentially on the back end? Look, look at, I guess the way I asked the question to myself was like,

At what moments is the CSM having to go do something that is not in front of the customer? And so like, if you kind of go back through my list, right? Like the contracting piece, that means they're reading a contract behind the scenes. Like they're looking for certain information. Have they gotten value? It means they're looking in a database or they're looking on a dashboard, right? So I tried to find these moments where they were, were essentially they're not in front of the customer. And I just thought, okay, like here are four critical moments that they're not in front of the customer that they're spending time.

that like it seems like an agent could go in like you like you were alluding to. Right. It seems like those are those are almost like stringent enough tasks that you could essentially like build the right workflow with checks and balances around each of them that you could say, OK, like now I've removed the CSM from having to do that manual activity. And that that means, you know, that means I'm trying to, guess, like minimize their operational overhead and kind of maximize their are they in front of the customer?

Jay Nathan (13:40.995)
Right.

Jay Nathan (13:50.882)
Yeah. And that's the key point, right? Which is why are we doing all this? To get that person in front of the right customer at the right time more than they have been maybe in the past and not just doing the rote standard things that we say every post sale person should do with a customer. Right. And I'm intentionally staying away from the term CSM just because that's a whole separate podcast we'll do, but like

Jeff Breunsbach (14:00.16)
Yes.

Jay Nathan (14:15.704)
Because it's more than just CSMs, right? It's renewal managers. It's your professional services, engagement managers, consultants, all that kind of stuff. your whole contract point brought up a really funny memory for me. It's not really funny. It's just interesting. when I first started at in my first VP of customer success role at a SaaS company here in Charleston, based in Charleston, South Carolina,

One of the first things I had to do when I got there, this is back in 2013, I literally spent days and days and days sifting through every single contract that we had with customers because, know, in the early days when you're building, like customers are all different. had hundreds of customers already. So I didn't know where we had term for convenience. I didn't know what the agreement was, the renewal date. None of that was in Salesforce yet. So I literally had to go do it myself and I was thinking just as I was

Jeff Breunsbach (15:09.003)
Yeah.

Jay Nathan (15:14.606)
I don't know where I was driving somewhere else thinking, man, boy, would it been great to have AI tools back then to go sift through all that and at least give me a running start on it. But I think, you know, that's probably one of the easy to go back to your four pillars, contract, you know, parameters, getting all that pulled out in a way and validated that sort of puts it in the CRM and you can run sort of more deterministic processes on it.

Jeff Breunsbach (15:23.702)
Yeah.

Jay Nathan (15:42.604)
or more standard repeatable things is really helpful. So getting back to agents with all these things, like to me, the best place to start on these is to build agents that are the most granular in nature. So the word agent is even overloaded in AI. If you go out and read about this stuff, like,

Jeff Breunsbach (15:42.891)
Yeah.

Jay Nathan (16:12.288)
If you go read open AI's version of what an agent is, they're really thinking like an autonomous actor that acts on your behalf to go make decisions about all the different subtasks that need to be built. I think that's cool. And we're going to get there right at some point. There would be a renewal agent, for example, that orchestrates everything that you just said, right? It would call the right tools at the right time. Go get the right information.

Jeff Breunsbach (16:30.284)
Yeah.

Jeff Breunsbach (16:35.98)
Yeah.

Jay Nathan (16:40.382)
from all the different places where that information exists, handle the sentiment, all that kind of stuff. But I think what we have to do now is actually build those from the bottom up. Right. So instead of a renewal agent, we're building a contract terms review agent, right. And a value metrics analysis agent and, you know, stakeholder analysis agent. And by the way, that might not be the

Jeff Breunsbach (16:52.684)
Yeah.

Jeff Breunsbach (16:58.22)
Yeah.

Jay Nathan (17:07.084)
bottom level, there might be sub agents underneath those that need to be built for very specific tasks. That makes sense.

Jeff Breunsbach (17:13.783)
Yeah. And that's, yeah, it does. Um, because like the thing, you know, I think like the, thing that's just always in the back of my mind, right. Is how do you create, um, how do you call, how do you create rip courts might be the right way I'm thinking about this, but like, right. You, I think like in my mind, like I want these agents to go, like, like you said, do these very specific tasks to bring back information, but I also want to be able to have, um,

Jay Nathan (17:28.844)
Yeah, I knew what you- I knew what going.

Jeff Breunsbach (17:40.896)
a ripcord so to speak where let's just say it's going to review the contract and maybe it's not so sure or maybe it doesn't find the contract right like we need to be able to essentially understand where those breakdowns are and go figure out okay are those the exceptions right or did we not build kind of the rules in the right way or is it not acting in the way that we needed to but I think like we need to understand I guess where there might be ambiguity and or where we might need to like pull the ripcord so like I guess having some of these breakpoints where like

your point, right? I kind of want these four agents to go do things in their own lanes before I bring it all back to something that I guess then is determining it based off of all four of these. Like right now, I think what I want to just try and do is let's just, Hey, step number one is almost like, let's use the agents to make our CSMs more efficient and effective. And then almost step number two would be, okay, now how can you go, um, like orchestrate this from almost a main agent, I guess is the way I'm thinking about it. Um,

Jay Nathan (18:37.898)
In the yeah.

Jeff Breunsbach (18:39.711)
Yeah, that would then kind of be almost always on versus us maybe kind of running these things at very specific moments in time.

Jay Nathan (18:49.698)
Yeah, in some of them, for example, like you talked about, confirm willingness to re to renew, like typically what we would do there is we'd have a survey that goes out six months ahead of time. It could be part of the process, but that's like, that's a customer. That's an external agent, right? Something that actually interacts with the customer. And the cool thing is what, what we'd love to be able to do there. What have we typically done, Jeff? We've typically sent a survey via email.

Right? Where it gets ignored, you get 5 % response rate if you're lucky. But what if the, if you train the agent to, well, not even train the agent, but what if you enable the agent to communicate with the key stakeholders that you identified on whatever channel that they're in, like the executive sponsor, probably an email because where else are you going to get their attention? Right? The administrator or the person who's like the

Jeff Breunsbach (19:18.145)
Yeah.

Jeff Breunsbach (19:42.037)
Yeah.

Jay Nathan (19:44.896)
internal champion user for the platform, maybe in the app, you go get that information, right? But it's the same, it's the same data being captured in funnel to the same place, right? It doesn't matter what channel it's on necessarily. So you talked about rip cord. That's, that's what I think people like about us, by the way, is we use like really plain language here, but open AI calls them guardrails. And this is where training comes in. So

Jeff Breunsbach (19:58.497)
Yeah.

Jay Nathan (20:13.578)
I think, you know, I don't know that this is a mistake that people are making, but I think it's something that you really have to be careful of as you go design and deploy these agents. This is a whole idea behind for deployed engineering in AI firms and agents is that you have to go deploy an agent to handle the tasks that you want to handle, but then you need to spend time tuning it and training it further to make sure it's doing the thing that you want.

The whole idea behind agents and AI in general is this ideal is it's probabilistic, right? Which means it's going to do what it thinks the best is to do with the information it has versus a workflow, which is deterministic, meaning we do the same thing at this point in the process every single time you prop interesting tidbit here. Have you seen the headlines about Salesforce and agent force and their whole

cutting 4,000 jobs and support. Well, what we decided to do, and this is again, like I'm not inside Salesforce. don't, you know, maybe I don't have all the detail here, but from the headlines, ostensibly what happened is they went down the full agentic path saying, we're going to let an agent, a support agent take care of all these support requests, decide how to handle them, decide when to go capture feedback and see set from the customer. They recently, and then they laid off 4,000 people.

That wasn't the reason, but we'll talk about that some other time too. So then they've backtracked recently and said, well, now we have to have this blend of, you know, durable standard process and agentic flows within it, right? Is maybe the easiest way to describe that, which that makes total sense, right? Like there's just some things you want to happen every single time a customer contacts you for help, no matter what.

Jeff Breunsbach (21:57.856)
Yeah.

Jeff Breunsbach (22:08.363)
Yeah.

Jay Nathan (22:09.326)
And you can't leave that up to chance, whether the language model or the model behind that agent decides that it's a good idea or not. So anyway, it's a combination of workflow. And I want to get into tools here in a second too. I want to understand what you're looking at. But workflow and agents together, some things that happen predictably every time and some things that happen probabilistically.

Jeff Breunsbach (22:26.838)
Yeah.

Jeff Breunsbach (22:37.195)
Yeah. And I think that makes, I think like that's the way, I think that's the way I'm thinking about it right now, right? It's like we have our, we have a renewal process that we want to happen every time. And then let's go deploy these agents at very specific moments right now, or very, very specific tasks. so that we can run our process more efficiently and effectively. And then I think like, to your point, like maybe the future does look like.

the agent is almost like coming up with what our playbook should be. Right? Like right now it's our playbooks almost are standardized, right? Based on the responses that we get from those four agents. But maybe what happens in the future is that the playbook is actually built by the AI after it goes and determines what's, Hey, I captured all this information. Here's the playbook that I think is best for us to go, you know, deploy in order for us to win the renewal. And I think that's what like the next, I guess to your point, that's what the future looks like to me is that like,

Jay Nathan (23:15.949)
Yeah.

Jeff Breunsbach (23:29.599)
once it's captured the information, let it determine what it thinks the best playbook is and put that in front of the CSM. And therefore it's not a, you know, it's not a renewal process that is like the same every single time for every single customer. It's, Hey, no, we've, we've kind of built this playbook for this specific renewal at this specific time with the information that we've gathered. And like that could still be changing, right? Like you could still have calls and it might change the playbook and things might happen. like,

Jay Nathan (23:49.454)
Yeah.

What about what about having an identify when an early renewal is appropriate? Right? You'd love to be able to do that or renew somebody early. Get that off your plate. Move on to the next one. Right. So yeah, it actually so you're you're talking about something that I just heard somebody talk about the other day. I mentioned the external agents, internal agents, and then almost like the hybrid of like operational improvement agent. Right. Even

Jeff Breunsbach (23:56.246)
Yeah.

Jeff Breunsbach (24:01.717)
Yes.

Jeff Breunsbach (24:18.741)
Yeah.

Jay Nathan (24:19.796)
even before you get there, because I think that's further down the line. Maybe I'm being too pessimistic on that, but think about this. The way you talk to a customer about their renewal can be 100 % personalized by customer now too. I mean, that's a good starting point even today.

Jeff Breunsbach (24:33.408)
Yeah.

Yes. And like, and the other thing I haven't added it here, but you just maybe think of it right. As like, part of a renewal that we should probably be putting in there is like, how's their business doing? How's their industry doing? Right? Like that's just, again, like another agent that's pretty easy to stand up that like,

go look up this company, go look up the market, go look up the trends and like pull that back and basically tell us, you know, are they doing well or they're not? Cause that's also like another point that I think is pretty critical. And again, to me is seemingly, this is another moment where you'd have a CSM go, let's go Google the company or let's set up a Google alert, right? Like that's the old way. And I think like this way, I actually wrote a prompt that we used quite often at my last job where we,

built a prompt in chat GBT that would go do market research on the customer. and it would look across like six or seven factors. So, you know, what's the, what's the financial performance look like? What's the leadership, other key leadership turnover, like all this kind of stuff that would pull it back. in a summarize, like one pager with like six or seven bullet points. And I think like this to me seems like another area to do that too.

Jay Nathan (25:45.422)
Here's something you got to be careful with there. If you go ask Jim and I, even chat GPT, Claude, like tell me when you were last updated with data, it will tell you it was January of 2025. We're in January of 2026 now. So you've got to be careful. So this gets into sort of some of the technical goo of like what gets passed in to these large language models that are behind the agents.

Jeff Breunsbach (25:48.267)
Yep.

Jeff Breunsbach (25:56.459)
Yeah.

Jay Nathan (26:11.426)
Like the company and industry information. I think this is why Google has a really important advantage over a lot of these other language model companies because they've got the real time search results. Now, like if you go update your earnings on your website today, Google is going to know about it in five minutes. It's going to serve those results back to people who are searching for your results. So that this is called rag, right? The big concept is called rag retrieval.

Jeff Breunsbach (26:30.869)
Yeah.

Jay Nathan (26:41.346)
Retrieval augmented generation. And that is the LLM knows a bunch of stuff, right? It knows about the whole internet, right? Cause that's where it's Ubered up data from, but I might have to get more real-time data and I might have to get private data into the context that I'm searching for or trying to make sense of, right? And so that might be a real-time query from Google or in the case of your customer data that

Jeff Breunsbach (27:03.873)
Got it. Yep.

Jay Nathan (27:11.264)
Means you have to be able to take your contracts, which you do not want to be public and will never be public, hopefully. And you pull them into the context as you're trying to make a decision using that large language model behind, behind the agent. So like, this is where like all these tools now start become really, really important workflow tools, agent builders, things like where do you store the data?

Jeff Breunsbach (27:26.251)
Got it. Yep.

Jay Nathan (27:39.394)
that is private to your company that your agents can see, but they shouldn't be, it shouldn't be distributed on the internet, right? So all these pieces, then your mind starts to explode around like, okay, how do we actually go build this? So my question for you is, what tools are you looking at today to go do this?

Jeff Breunsbach (27:49.237)
Yeah.

Jeff Breunsbach (28:00.248)
I haven't actually. That's a good, I mean, I've messed around with tools myself, but I haven't, I think at like this, like a company level, like you said, right? Like thinking about some of the compliance or data privacy things, creating like kind of own environments. So I need to go do some research.

what we can go use or what we have access to. I do know that we've got access to a couple tools today, right? Like we use HubSpot, we use Google and Gemini, we've got ChatGPT, we have Zapier. So like we definitely have tools at our disposal in order for us to go build, I think, some of these things. But I think like you bring up a good question too, which is essentially like,

Jay Nathan (28:24.078)
Yep.

Jeff Breunsbach (28:39.317)
we need to have a technology map of like, okay, how are we going to make sure like the data that goes into any of these tools is secure and safe and right. And how's it going to access this stuff on a regular basis if we needed to, which I think is the part I probably need to go do more research on. But I think, between some of the CRM and like Zapier and I've never known if it's Zapier or Zapier. Yeah. Happy with Zapier and Zap, right? Zap, not you wouldn't. Yeah.

Jay Nathan (29:02.446)
Zapier, happier with Zapier. Zap, yeah.

Jeff Breunsbach (29:09.291)
so yeah, but I think, you know, from what I've seen, I think, Zapier has done a good job of, of, you know, kind of staying in this workflow business. They've been there for a long time and they're starting to overlay AI, but I definitely think I've heard like N8N come on board. Auto is another that I think is, it's kind of right there as well. So I feel like those are at least three workflow tools, with like AI capabilities that I've, I've heard about.

Jay Nathan (29:35.106)
Yeah. Okay. So you just totally validated an article that I'm writing right now about like, what is the spectrum of agentic tools that you even have to choose from? Like if you're a Microsoft shop, adding on Microsoft co-pilot studio where you can build agents is actually a pretty trivial thing, right? And it might already be sanctioned within your organization. It knows how to connect to, you know, your data sets and your raw data and SharePoint, your whatever data sets you pointed at. So

I did some, was doing some research in this last week myself, just to try to, cause my mind is blown. Everybody's mind should be blown right now with the number of tools and options out there. Think about the list that you just made. Jeff, you talked about hyperscaler cloud platforms, Google, right? You talked about, SAS tool vendors like HubSpot. You talked about integration, middleware tools, cloud middleware, like Zapier or Cato like

Those are all different categories of technology. so, and they're all trying to play in this agent game and this agent agentic world. So I came up with four categories of agent platforms and we could see if these are right. I don't even know yet if these are the only four, but I think it's pretty close. So number one is basic off the shelf tools that have like an agent built into them.

So like a simple chat bot, what else here? Like something like, like Finn, right? In intercom, like it's just a, it's a support platform. Now they may come back and say, well, no, we can do so much more than support. I'm sure that's true. Right. But I'm just using an example. Finn is intercom. Then there's configurable agentic products. And so names that come to mind, Sierra that AI, maven.ai, these are two platforms that

We have two customers, two separate customers of ours are using and they're using them for CX processes and workflows. I don't think they're limited to that, right? That's the confusing part is what is a, what is the workflow already know about from a domain knowledge experience versus just being a generic builder for AI. Then there's platform embedded and that would be like,

Jeff Breunsbach (31:42.667)
Yep.

Jay Nathan (31:59.598)
Okay. We're a Salesforce shop through and through. They've got agent force natively connects to Salesforce, which is the corpus of my, you know, a key corpus of data that I have, but can also get data from other sources. And then you have custom magenta platforms, which is like you said, I'm going to start, I'm going to, I'm going to, I'm going to use an eight N for my workflow engine. And I am going to use, uh, maybe I use that for my agent builder as well, because you can build agents and an eight N and eight N is an open source platform.

gaining popularity. think it's a little bit more mid market today. Maybe somebody can correct me on that if I'm wrong. But then you've got the big agentic platforms like you can build agents in open AI. You can build agents in all these platforms in Microsoft Co-Pilot. think Google has an agent builder studio. So that's like if you want to own the infrastructure in custom build your agents, you can do that, right?

Whereas the other ones, the configurable products and the platform embedded elements have like prebuilt agents that you just sort of go implement and then integrate with other pieces. So I'll pause there. I know it's probably not real clean list, but that's.

Jeff Breunsbach (33:10.956)
Got it.

Jeff Breunsbach (33:17.973)
feel like you can come up with better names, know, more sexy names, but no, think categories, the categories make sense to me. And like, like you said, like as you went through them off the shelf configurable platform embedded and then custom agentic, like I could kind of see like tools and technology that I've used or that we have like in-house kind of across those things. Right. And then I think that's where I think that is maybe where it gets complex as you start to map this out is, is

Jay Nathan (33:21.006)
you

Jay Nathan (33:40.323)
Yeah.

Jeff Breunsbach (33:46.604)
You know, do I have to, if I build custom agentic, do all of them, do all my processes then have to be custom agentic or if I build them platform embedded, do they need to be in the platform? You know, it's like, I think that that's where it probably starts to get really complicated. Cause then you start thinking, okay, like if, if each four sets of these products or tools have the capabilities, you can see how all of a this becomes a very big tangled web. like, and I think this is honestly like something I experienced, at a previous company too, is that you had.

Jay Nathan (34:08.428)
Yeah. Yes.

Jeff Breunsbach (34:16.968)
you kind of had the mandate hey let's use ai and then you had individual contributors who were like cool i'm gonna go use ai and they would build some super cool agents and other stuff but then it wouldn't

How you know it's validated? Is it talking to our systems and tools? What data did you put in? And all of a you'd pretty quickly, right? When you, when you have an organization of thousands of people and a bunch of individual contributors that are trying to help use AI, it's helpful in some aspects where like on their day to day, you might think it's helpful. But then all of sudden when you start to think about, okay, how do we roll this out team wide?

Jay Nathan (34:40.152)
Yeah.

Jeff Breunsbach (34:49.495)
Um, you know, how do we roll this out? Like across the entire CS team, you can start to see how that's like a web because people are all at different stages using different, almost like they're each using different parts of the four categories you talked about. And that's where you're like, okay, somebody needs to map this out. I'm like, okay, if we're going to go build agents for these tasks, how's it going to be done? Where's it going to be done? Let's standardize this. Um, and I think that's probably where like a lot of people are stuck today is that they don't.

Jay Nathan (35:07.533)
Yeah.

Jeff Breunsbach (35:17.867)
they don't have maybe like, they don't know where to standardize those.

Jay Nathan (35:21.368)
Yes, 100 % true. Now here's something really important about all of this that I think we have to keep in mind is that the closer you can get a business user to agent development, the better. So Walmart is a great example of this. Walmart, they're head of, I can't remember his title, basically they're head of internal business systems kind of guy. I think it's external as well.

His mantra is basically, want to give agentic tools to every Walmart associate to work with, right? So that they can build, because they know what the problems are that they're trying to solve. And you want them to be really well-versed on what these tools can do. So that's layer number one is individual contributors should have access to some kind of tool where they can build a basic agent, right? And use plain language to define it, which is a hallmark of these things.

So on and so forth. Then I think there are departmental level purchases, which really fall into more of this configurable agentic product category. Those are going to be a little bit more deliberate in terms of selecting that solution. So, you know, our customers selected Sierra for their AI platform for their customer facing CX agent, mainly a support use case today. Another customer chose Maven AGI for mainly their

renewal flow agent today, right? And then I think there's enterprise capability, like, okay, we have a platform where we can build enterprise grade agents to go do the most complex, most important things within the whole company. And if you're going to go build something, you're going to use that. So I think those are the three stratifications, individual contributor tools. Everybody should have access to those in 2026 so that

Jeff Breunsbach (37:10.092)
Yeah.

Jay Nathan (37:18.668)
the organization can become well-versed. It's beyond just having access to chat GPT or Claude now, right? You have to be able to, like that's table stakes. You have to actually be able to build something that you start to manage your little team of agents, right? And everybody needs that platform. Then there's department and enterprise level. So maybe that's the way to think about the stratification of it, just to simplify it a little bit.

Jeff Breunsbach (37:25.429)
Yeah.

Jeff Breunsbach (37:33.633)
Yeah.

Jeff Breunsbach (37:43.798)
Yeah. Well, I know we're coming up here kind of on our, our window of how we want to keep these, but,

I think it's a good, maybe good kind of launching point for, what we're trying to do here. So I think like early in this conversation, we talked about, maybe a specific use case around renewals, kind of four to five agents that you could go build that would essentially bring back information for the CSM to go have a better playbook and a better conversation with the customer about the renewal. And then I think like that launched us into the topic of conversation, about the types of platforms that are out there. And so you kind of categorize them as off the shelf configurable platform embedded customer agent tick, but.

There's definitely starting.

Jay Nathan (38:21.826)
Configurable product, configurable product agents. So that's very specific because it's like a pre-built predefined product. Sorry, but yeah, you go and you're on the right track.

Jeff Breunsbach (38:25.92)
Yes.

Jeff Breunsbach (38:29.429)
Yep. And just the.

understanding those platform or understanding those dynamics, right? And I think then trying to overlay on top of that, like what types of platforms are you going to give access to your ICs? How are going to think about making decisions on your department level? And then, you know, if you're an executive, how are you going to start to have the conversation with the other executives about like, you know, what are we going to start to push kind of company wide? So I think this is cool. This is I think like the thing that just came to my mind about this podcast is I think it's going to be fun. think I think you what people maybe

I don't know about you is that in a former life, you were a software engineer, you were deep into the weeds and you love to nerd out over this stuff. so I actually think this is almost going to be like a, I'm going to be the young grasshopper and you're going to be bringing me along and understanding what some of these things are, especially because you've gone to do the research. So think it's going to be fun to tease these conversations out and bring out like.

Jay Nathan (39:23.885)
All three.

Jeff Breunsbach (39:31.317)
I think real world applications of how these things can be put into your business. because again, I think that based on some of the responses that we've gotten from our newsletter and, just from LinkedIn messages and things that I've gotten, like, think there's a lot of people who are at that precipice of like, okay, I'm being told I should use AI. I know I should use it. I know my teams should use it, but like, how do we actually go do it?

Jay Nathan (39:54.062)
100%. I had a call with a friend of mine who is also a customer over the weekend and I, I called him up cause I was sort of looking at their architecture for what we're trying to accomplish with them. was like, you know, I think we're missing a piece here. And he was like, you know what? I realized the same thing this week independently. Yes, we're missing that piece. Let's go figure it out together. And so the cool thing is that everybody is working to figure this stuff out in real time. Like I can't wait to see what you come up with.

careers, right? And I can't wait to share what our customers are doing, both enterprise and mid market and some even SMB. It's a little different, but it's at the end of the day, everybody is trying to figure out how it applies to them. like we always do, Jeff, we'll just keep learning and sharing and hopefully it'll help somebody along the way.

Jeff Breunsbach (40:42.839)
Cool, well, appreciate the conversation. We'll see you all next time, next week.

Jay Nathan (40:45.206)
All right.