SaaStr 840: From 1 Agent to 20+: The Reality of Managing Multiple AI Agents Across Your GTM with SaaStr's CEO and CAIO

0:00Welcome to the official Sastra podcast where you can hear some of the best Sastra speakers. This is where the cloud meets. Up today on the Sastra podcast. If you don't know your FTE, if you don't know the answers to these questions, and you're spending any material amount of money on an agent, you're wasting your energy. And so it will be interesting to see where it goes this year. A lot of the agents we use are pushing down market to be more self-service. So far, that doesn't work. So far, I will say for the most part, agents that require deep training cannot be self-trained. It will come. Agents are getting so much better.

0:37That's frontier one. But wait and see. Be skeptical if you buy a cheap tool that says it's self-trained. Make sure it works. And you know the time. If you buy a more complicated tool like we're talking about, just talk with someone senior enough on deployment, not again, of someone trying to sell you something that doesn't know, and be honest about what it's going to take. Otherwise, it's like going to the doctor and getting a prescription for medicine and never taking it. It's not going to work. It's literally like that for an agent.

1:07Hey, Sasser, imagine having agents for every support tech. One that triages tickets, another that catches duplicates, one that spots churn risk. That'd be pretty amazing, right? HappyFox just made it real with Autopilot. These pre-built AI agents deploy in about 60 seconds and run for as low as two cents per successful action. All of it sits inside the HappyFox Omnichannel AI First support stack. Chatbot, Copilot, and Autopilot working as one. Check them out at happyfox.com slash Saster.

1:39Hey, everybody. Saster Annual will be back. May, 2026. The world's largest SaaS and AI gathering for executives. Just as last May, we hosted 10 ,000 attendees with 68 VP level and above attendees, 36 % CEOs and founders, and 25 % were AI first professionals. It's the very best of S-tier attendees and decision makers that come to Saster Annual and AI Summit each and every year. But here's the reality, folks. The longer you wait, the higher ticket prices get. They're cheap now. They're cheap, so just get them. Early lock in your spot today. Use my code Jason100 for exclusive savings. Get your tickets at podcast.sasterannual.com or just use code Jason100 when you check out.

2:20See you there. Saster Annual and AI Summit 2026. It will rock.

2:27All right, cool. So, yeah, to kick things off for today, we wanted to talk a little bit about where we're at today with all of our agents. And maybe more importantly, now that some of you have deployed at least one agent and are looking at doing multiple agents kind of in the path that we've been. How, what does that really take in reality, in fruition to do this multi-agent management? What does that all mean? so just for quick context i think a lot of you have seen this now but just as a refresher we have about you know 20 plus agents now we've i've got about 12 apps they've been used you know almost a million times which is kind of crazy i think we'll probably cross a million mark before the next ai day um and so that's a lot of usage right and so a lot of the things you'll see us talking about today we also have on sasser.ai slash agents so you can always go there at any time and see like you know some of either the third party tools or the in-house ones that we've built and kind of play around with them there okay um so what have we learned now that we're about you know almost depending on how you count either eight months or a year into this journey i mean i was looking up like when we when you joined delphi jason i think it was like December of 24, but I didn't get on it until like February, March timeframe.

3:49So depending on where you can't like fully deployed though, I say eight months was multiple agents in tow. So we started with Delphi. If you guys haven't read it, you can go to saps.com. It's the Jason bubble there. You can talk to his clone. So we added that and then we quickly learned, you know, it was doing kind of like advice and support and we spent a lot of time training it. We've talked about it on other pieces of content here. But that was our first foray into an agent. And I think support and sale I see now is like the most two common use cases of where people deploy their first agent.

4:21So super, you know, super common thing to do there. And then, you know, then we added an outbound AISDR if we added multiple of those. Now that's not necessarily, I think, a path that everyone needs to take. You could probably deploy on outbound AISDR agents and do it well. And then Secondly, I think, too, we also looked across our sales funnel of where do we want to be in terms of what else we want to deploy and go to market. And so we quickly added multiple outbound AISDRs, an inbound AISDR agent, and then multiple agents across go to market, and then a few custom by-putted apps, some of which we're using now internally, but mostly our external facing, like the pitch check valuation and things like that.

5:08it's a good summary a lot of folks have followed the journey but um we did we did push the limit here as as some folks know after saster ai annual last year in may basically anyone that left our tiny team we replaced them with an agent so we've been on this journey we've deployed you can see our whole uh list we've deployed a bunch of startups are also one of the leaders on agent force and Amelia will touch on that. We'll keep chatting about that because I think it's just helpful for you guys to see these apps in production. And then we'll get into it. And then we've also, we started off vibe coding for fun, but as Amelia will go, when we found, we really recommend you buy something.

5:51Don't build the 90-10 rule Amelia will have here. But the latest thing we've added, which we've talked here is we built our own VP of marketing. And we'll talk about why in a minute. Yeah. So that's, yeah, that's kind of been our journey again it's not necessarily that everybody needs to be at you know be a number of agents we are for a lot of reasons we try different agents right like part of the saster community and ecosystem is we are trying different agents we have different partners and so we have also a like underlying need to just try a bunch of different things i also am just like wanting to try different agents and see what works and what doesn't and so there's like that inherent but that it's not necessarily what you need to do for your business it may not make sense to be at the same level and number of agents we are.

6:36But okay, so eight months later, you know, just kind of like highlight result is, you know, does all this work? I think I've seen kind of now that we're, you know, maybe a year slash eight months into this journey, I'm starting to see a little bit more like skepticism, honestly, weirdly on LinkedIn. I don't know what you're seeing, Jason, on X. But I see some people becoming a little bit disenchanted with AI agents. And I see a little skepticism. So does this work? I think for us, you know, we have now eight months in 4.8 million and counting an additional in additional pipeline source via agents and i'll talk about why it's additional and then about half of that so 2.4 million is closed one revenue that was you know first touch source from an agent across go to market so a lot of a lot of good things there as like a proof point additionally it's yeah we've seen some underlying things also improve via the agents so our deal volume has more than doubled right i count a lot of that towards and credit a lot of that towards the agents working you know 24 7 365 they can always answer a question they can always book a meeting they can always reach back out to you now sometimes you know the humans here like me david jason have to take the meeting so sometimes they're limited by our human capacities but i i credit that to the agents working you know you're you're around all the clock now our win rate has nearly doubled i think that's just in the nature of the agents you know especially when it comes to folks that are inbounding and also how it's doing outbound just having a lot more context right it's a lot better context it's a lot better outreach in some cases but then essentially for us like across me and david specifically itself like it helps us with our conversations because we already know what this person has said to the agent we can see the exact conversation they're having we can see for on the website we can literally see what else our company has been doing with our agents and we use that in the mean right so it saves a lot of time it's also a lot more qualified when we get on that call and so i think a lot of that has to do with some of the nurturing there and i think more importantly too in those stats like it did not cannibalize our other inbound revenue sources these have just been augmented by our agent and then another thing i like to remind folks too is like and it's not that we dropped a bunch of stuff either when we deployed these agents right i think a lot of folks now will be like well okay if i if i just get an outbound ai sdr then maybe i don't need any human sdr and maybe that's true right we we between data and i we do some we do some outbound ourselves so we don't really have a human sdr per se but i think there's a lot of pieces in which you would you might still need both why because we we do personally respond as humans to each message that our ai agents produce and so there's a lot of you know time needed there that i don't necessarily like the ai autopilot responses i still think it's better to come from a person who actually knows the business and i think too in that too i think um you know a lot of this has been augmented by our agents so you know in helping us book again more meanings helping us understand the leads a little bit better again we did not drop the other things we're doing right we still send marketing emails we still do outbound we still send gifts to people like we still invite people to come to sass like all the things we used to do before with the age of like now we can do them a little better but we're still doing them and so i think that might be surprising to some folks but just know i don't i don't think it will cannibalize anything if you do it right but i also think it can definitely augment versus you know you don't need to go after the service we have to necessarily replace we've done that and it's worked but you may see that doing a mix of both kind of work okay so cool but you know here's the um here's maybe the honest truth that that you may not see on linkedin or x and that is that we maintain these apps every day like literally even this morning before getting on ai day and stuff you know i'm like checking our agents and i think the important thing here is like the agents and the humans have to rapidly evolve and change constantly like it's such a mind share killer for myself for jason like we're in these agents you know i think 15 to 20 hours a week each that's each not like between two of us, that's each of us, constantly iterating with our agents, constantly saying, what are they outputting, checking the responses, making sure it doesn't hallucinate, making sure it's talking to folks the way we want it to talk to people, making sure it's adding value, making sure it's not degrading.

11:23Sometimes you see these agents degrade over time. And so I think the important thing here is it is kind of a real killer. It does take a lot of time. I don't think you can replace the time of managing you know we've just seen the time shift like the time we used to spend managing slightly more people on our team we now spend that same amount of time if not more managing the agents but it's just a lot different right like there's no there's no people drama really but the end the agents just work at a much higher capacity and higher scale than a human being that it's hard to eventually keep up with them and i put this here in bold i you know i've been trying for a long time for the last few months to keep up with my agents and then I realized it was futile because I couldn't fordo it but I try and keep up as best as I can and in that what I truly mean is you know anytime we get a response we have a system that it'll slack us from like any of our agents so whenever there's an interaction or agents having a conversation with somebody and we want to reply and you know again we like to reply ourselves it says this is like we do try and respond to those people literally instantaneously, if not in real time.

12:30Sometimes we are asleep. And so we respond to them first thing in the morning. But I've realized I can keep up with my agents. They're smarter than me. I'll tell you just one nuanced learning actually from this week. So I was meeting last night with the CEO of a next, we're already in the next generation of AI go-to-market agents that's already got millions of revenue and is publicly launching in a few weeks, but they already have millions of revenue. And I asked, I've known them for a long time, but I asked what the secret sauce was. And the secret sauce was they do everything. They do the onboarding, they do the tagging, they get the first campaigns running, they do everything.

13:14And they do it almost to such a fault that some of the customers think it's too easy and don't even realize the energy that's going into it if they haven't deployed an agent yet. And the learning from that is just, if you haven't deployed many agents or any for real, you got to have an honest conversation, not with someone in sales that doesn't know how the product works or use it themselves, with a forward deploy engineer, with a leader, and find out what is it going to take to be successful. Up front, the first 14 and 30 days and every day thereafter, and then you got to do it or it will fail.

13:49Yep. And meet with the best of them. Um, and you know, if you don't know your FTE, if you don't know the answers to these questions and you're spending any material amount of money on an agent, you're wasting your energy. And so it will be interesting to see where it goes this year. A lot of the agents we use, um, are pushing down market to be more self-service so far that doesn't work so far. I will say for the most part, agents that require deep training cannot be self-trained. It will come. Agents are getting so much better. That's frontier one. but wait and see, be skeptical. If you buy a cheap tool that says it's self-trained, make sure it works.

14:27And you know, the time, if you buy a more complicated tool, like we're talking about, just talk with someone senior enough on deployment, not again, of someone trying to sell you something that doesn't know and be honest about what it's going to take. Um, otherwise it's like going to the doctor and getting a prescription for medicine and never taking it. It's not going to work. It's literally like that for an agent. Right. But this was the first one I saw that could do the type of stuff Amelia and I are talking about without you doing any work. But it's because they have a huge human team. They call them forward deployed AEs in the beginning, and then other folks take it off.

15:00But that's an extreme case. I haven't seen any other app like this that can just do it with not consistent training and work every single day. Every single day. Yeah, we do spend a lot of time per week actively managing all of our agents. Just, I think, to Jason's point, be prepared, right? It is, again, I think I see too many folks who they see the really good stats and they get, you know, even our stats are, you know, fairly good. And they get a little like mesmerized by them thinking, you know, OK, it's AI. I can just prompt it and it'll it'll do it fairly quickly. I'll say to you, like most of these agents have some sort of prompting in them, but they're not necessarily all built via a prompt.

15:41Right. Like don't think about it the same way you think about putting a prompt into open AI or Claude. it's not going to function the same way. There's a lot of like, there is some prompting in each of these third-party tools, but at the end of the day, you're going to have to like figure out what works, put that into, you know, their prompt builder in whatever format they have and refine it from there. And then there's usually additional steps other than a prompt. So I think some of the tools we've seen are kind of getting to that point. Maybe they'll get there like by the time we're at SAS Reannual or maybe right after in like the second half of the year.

16:13But most of them are not just prompt builders. So I think that's another thing to just bear in mind. Like it's not, if you're used to kind of like this easy path that Claude and Chat and OpenAI have, you know, kind of proprietary put out there, like they're not all like that. They're not all that easy. Now I will say you can use things like Chat and Claude to make your prompts better. I do that all the time. Like I'll put whatever context I'm putting into our agents. I'll run it through Claude. I'll run it through ChatTBT and see what I suggest to make it better. Sometimes I'll take or leave the suggestions.

16:44Sometimes it's not, you know, the AI doesn't know your business the same way you do sometimes. And so I think that's just another thing to be prepared for. So how did we get to this? I think this will address some of the questions as well. How did we get to these results of, you know, 4.8 million and counting an additional revenue, 2.4 close to one, so about half that. And then also, you know, we've now crossed the 60K mark in how many emails and interactions our AI emails have had just in the sales funnel, right? Like that's not even counting the almost close to a million we've had in our proprietary interactions with our VibeCoded apps.

17:19But that's a lot right there. You're going to explain how this works. And given how tiny we are, it's pretty impressive numbers. But if I had to summarize all of this and then challenge me if I'm wrong, because you're doing the real work, right? I think some of the key here is not that all these emails were the best emails that have ever been sent in the history of mankind. I know you think they're great. I actually just think they're okay. I don't think they're bad. I think they're better than most of the outbound emails I'm going to get during this day, during AI day. The ones we send are better, but they're not the best.

17:49They're not something you can spend an hour crafting. I think the number one key, and that's why the 60 ,000 key is cool. We are touching folks, lapsed folks. We forgot to talk to folks. We don't talk to enough more often. We are connecting with more people more often, not spam, right? But we couldn't do 60 ,000 high quality emails manually or even with old school outreach tools. We just couldn't do it. Right. So I think the key to this, tell me if you're wrong and then we'll be quiet. I think it's just more high quality, pretty good interactions. Yep. That's the thing. We're getting scale. And that's why what I think, when I think about everything we've learned and you've done most of the work, Amelia, if I had to advise people that are earlier on their journey, find something in your go-to-market motion that just isn't getting done or is getting done very at a very mediocre level yep then put an agent don't try to replace what's working well do that as your 10th agent or your 20th like literally we're such a tiny team we just weren't reaching out to enough people in our base in our activated base and so that's the low-hanging fruit for us we just could not be sure any you could put something in a dated outreach sales loft cadence but that don't work, right?

19:08But we never would have done this otherwise. So find that low-hanging fruit, the stuff in your jodiment that you're just not getting to, the customers that are too small, the customers that take too long to respond and your team doesn't want to do, the customers that have lower scores, right? But they still have intent, but no one wants to call them back. Do those ones for whatever your low-hanging fruit is because then even if you get some yield, is it's it's magical yep i agree so i think again a little bit of a misconception here related to some of what the chatter is the formula for us is to copy your best human like as you're deploying maybe you've already deployed one agent maybe you're deploying your next agents to jason's point right do something that you could either get a lot of scale out of by adding an agent and do pretty good.

20:02There's just some things like the agents can't slash shouldn't do. Like, obviously we love agents. Like, I love our agents. I use a lot of them. I add, you know, I just added the AIBP marketing. We'll show you guys. But there are some things I'm like, the agent would just suck at that. So I'm just not going to do it. It's just like, there are some things I still need humans to do. Like a lot of the production stuff we're doing for SAS Reannual. I'm like, I still need a human to do that, dude. The agent is not there yet. But what we also mean by copy you're human is if you're going to add AI agents at scale to help you scale, right?

20:35Just on scaling more emails, more meetings, more clicks, more volume, figure out what works first. I see too many people who, you know, okay, they want to automatically give an AI SDR to their SDRs. And I'm like, okay, well, are those SDRs new? Did they just join? One, I don't think it's a good idea to give it to every single SDR. I think that's, you're going to, there's a lot of reasons that would get you into trouble fairly quickly in terms of workflows. But I also think if you don't know what works first, I don't get this mindset of like, oh, it didn't work, but I'm just going to add AI and it'll magically work now.

21:13Like, no, if it didn't work or wasn't working before AI, it's not going to magically work now. And somebody asked me this question yesterday when I was doing a Salesforce webinar, like, okay, what if I'm a super early stage startup and I don't know what works? And I was like, well, do you have any customers? They're like, yeah, we have like, you know, 10 paying customers. I'm like, well, go ask your customers why they bought you. Like everybody has at least a few customers, or maybe if you're super early stage and you've got a few folks on a trial, just go ask them, like figure out what works, figure out what got them in to your product and is getting them hands on product.

21:45figure out what works first right we had so much data that we ran through before we put it into any of our agents on what was working right the best the best a like the best email copy for doing outbound the best responses of how we should follow up with inbounds you know the best um contacts and verbiage about saster you know about saster's events about sponsoring saster like we went through all this data we went through all this context we flagged everything that was the best of everything before we put it into any agents, any AI, etc. So, you know, train the agent on what works best. I think I see too many people now falling into the strap of they want to add AI into something new.

22:28And sometimes you can and it will work to some degree. But I think if you do it and you train it on the best of everything, it will work that much better, right? It will get you to pretty good to Jason's point. Like it'll get you to pretty good emails. They still may not be the best on planet earth but it'll still you it'll i think it'll put you over that bar of pretty good versus crappy ai emails that we've all seen or even crappy human emails that we've all seen so yeah that's that's why i can't get you on it but yeah i think you know you have to train it on the best of everything and if you don't know what that is yet i would take that time take a week figure that out before you know you deploy your first or next agent so that and then see where that gets you i feel like you'll have a better output because we are constantly iterating our agents now to make sure they're they have the best of everything and that they know everything that we know like as we know it right so like as we get you know speakers for saster we have new things that we're doing or now we've got like lounges and stuff or like new things in our sales process like we're i'm constantly making sure the agent knows all this so that i can talk to that.

23:36Okay. So I want to address some of the questions in the chat of like, you know, some of folks are asking about evaluation tools, like what's our processes. And then this is the 90-10 rule that Jason came up with, but I really do agree with, and I think it's a good one, which is, you know, buy 90 % of your AI stack. And I'll talk about the evaluation process we've done in a second and only build the 10 % where there's, where I, where I think there's, you know, there's no vendor that can do this well. And it's either a P1 priority, priority, or as you'll see in like our AI BPM, you know, I built that agent because it was a commodity.

24:15Like it was something where even with all of our agents now, I was like, I just still have so much data from Sastra internally that I want to act on. And I want to deploy this agent in a way that maybe I don't need it to run everything automatically so it's a very specific use case but that was where i kind of built that and that's where that kind of fostered in from right it's like okay i had all this data i wanted to do something that was more internal facing not necessarily external like a lot of our go-to-market agents are and so in that case it made sense to build i'd say for a lot of things it doesn't make sense to build right like if you guys listen to the podcast kyle and jason did i think we put it up like last week or something kyle who's the cro of owner talks about how you know he's also kind kind of roughly followed the 90-10 rule of, you know, he's bought a lot of third-party agents, he's made them work.

25:04And then he hired somebody who was like a, I think it was like a former founder or something, right, Jason, to like build a proprietary in-house tool. And that's like one extreme, right? But like even that 10 % that he's building in-house, like he hired somebody who like was a CEO, was an engineer, like knew how to code. Like I think he was like a CEO of like an LLM company or somebody like knew all this crazy stuff and like could build a proprietary like internal agent but again for a lot of things it probably won't make sense to do

25:34um i think too just to address some of the questions on the chat of like what what's been our criteria um and we've talked about it a little bit before but when you're evaluating these tools for the 90 you want to buy um i think the important thing is to one you know again i don't know why i think in the age of ai people sometimes will will throw things away because they're like oh there's this shiny new object i literally asked all of our all of these ai tools that we now use and deploy for help i was like i need one like one i'm gonna need help like i'm gonna need an fd and two let me talk to people who have used this like i think i see too often folks are like okay it's an ai tool and so i'm not going to ask for a customer reference like ask for a customer reference i do these all the time now like i try to make them as short as possible now because um you know we do these webinars and stuff too but i do this all the time now like marshall from manuel mint kyle from owner like we do this all the time like philippe from persona we do this all the time now people ask us constantly for like you know a customer reference like it's called like ask them for a customer reference and if you can ask them for one like in your vertical see what they say to you right like if they push back maybe don't use that vendor like they you know most of these folks have at least one customer that's slightly like you were, if they don't have one in your vertical, maybe you can give them a pass on that.

26:52But like at least talk to a customer and then see how much they will help you, right? I think a lot of these tools, to their credit for the third-party tools we do use now, have been helping us along the way, right? Like there's some of this, like we've learned from just now deploying so many agents, but some of it was because they put an FDE on our success team, right? Like Salesforce put an FDE on our success team. Orison is unique in that, you know, anytime I have an issue or I have an idea, I just see the CEO or the head of product, you know, qualified. There's an FDE on our success team. Oh, you know, Replit, we have an FDE on our success team.

27:32There's just so many cases here where if you ask them for that, they should give you some level of that service, right? Like to make it work because they should want your business and they should want you to be successful. Now, it doesn't mean that you need to have an FDE like every week. Like now I meet with them a lot less often than getting started. Right. But you should ask them at the very start, at the very least to have some FTE time at the start.

27:57Say one thing on the tools. I know this is versions of things we've said since the beginning. When you're talking to a vendor, if it doesn't feel right, don't buy it. Yeah, it should feel right. It should feel a lot of folks flame me a little bit when I say a lot of agents should almost get you going for free, right? And a lot of agents can't do that. There's economic reasons, there's headcount limits. People can't really train you and deploy you for free. But if you look at like the 20 VC that I did with Harry and Rory when Mark Benioff came on, it was interesting when he said he wished he could.

28:31He said he can't at Salesforce, but he wished he had enough FDEs that everyone could be in production on Agent Force before they had to pay. it's not practical but the best ones take you as far down that journey in the age of ai as they can they're proud of their products they'll show it to you if something doesn't smell right if it doesn't feel right if you don't think it's going to work it won't work buy another one even if the brand's less good even if it's scrappier even if whatever if it smell if it doesn't if your spidey sense says this agent isn't going to work don't buy it yep i agree i think too like um yeah that's that's a the the point you made on the free trial they like a lot of agents cannot set you up for free that's a really good point in the evaluation so yeah we you know we threw down for these agents it makes it hard it makes it hard it makes it harder risk it is interesting i want to say it is interesting that when you look at the prosumer ai tools that we highlight all all the reeves and the gammas they're lucky because you can get so much value for free even forget about 29 bucks a month or 99 bucks you actually the free products are great like try those tools the problem with AIG Tim tools is like even if they want to do it they can't do it right so you've got to take some risk but um maybe not later in the year but you know don't do it if it doesn't feel right all right that's our kind of like build versus buy rule and then once you get to this point in the process like um something I wanted to address which is also the title of this hard talk today was what does that look like in reality once you get into multiple agents right and i'm going to say something today that it's not so simple don't let that scare you don't let that like frighten you off of doing more than one agent and maybe you stick to one and it works really well and that's fine like you do not everybody needs to be i think in a multi-agent management journey but just know that if you are you if you're in that journey today for ourselves and what i've heard from some others is it's kind of all band-aided together.

30:35There's kind of a big reason folks like Salesforce are having a big renaissance because a lot of these third-party tools we use, for instance, and for example, push back to Salesforce or we push all the data back to Salesforce with a Zap or whatever, or some of them have a native that they can push records and update records back to Salesforce. And so a lot of the time right now, it looks like all of our third-party tools, whether we're API-ing into them or not, or using things like a Zapier. Then we have all of our internal data and our VibeCoder navs, right? We're pushing all that back into things like, you know, Cod, Zapier, back into things like Salesforce as our like system of record, just to keep all the records up to date somewhere central.

31:18But, you know, that's not native now, right? For now, that's not native at this moment. And so it takes a lot of webhooks. If you haven't heard this word, you'll probably learn it fast. We have so many webhooks in our Zapier account. I can't even count them, right? We have so many webhooks just firing all the time to push things back. But I'm pushing them again as into one kind of thing. And for now, that's like Salesforce because it can ingest all this data and take all the context for our agents. But, and not to say like, you could say, okay, I don't maybe need that data everywhere all at once.

31:56But I like to have it. I like to build the context of the agents from one agent to another. And so to let it build on itself, we use a lot of webhooks. We use Appier. I know N8N is having a renaissance now because it's kind of the same thing, but just built in the age of AI. But whatever one you use, you're going to see quickly. And I've got a screenshot of it. You end up with a lot of different hooks and kind of hodgepodge things together. but I think it's just for now right I don't think that's a problem for always I think it's just a problem for now you know in the first half of 2026 to to you know have it kind of web hooked into things that you need to make sure you can control the flow of what your agents are are doing and where that data is ultimately pushing back to and pulling from I do think you should pick one source of truth right at the end of the day to store some of this and then build further context for your agents you know we put sales first you could pick up spot or something else i think also to get used to your agents talking to each other on their own you know it happens our agents talk to one another it's fine get used to also as a human like talking to your agents um it is kind of a weird thing to at first get used to and then you'll get used to it and then also get used to you know for now copy pasting context like we do a lot of context sharing between our agents like yeah some of this pushes to salesforce but sometimes i'm like you know what i don't want to push through that flow i'm just going to copy paste something in this context from one agent and then put in the other agent like the way that it it understands context and so again that's not necessarily the simplest or the cleanest path of multi-agent management and so i I just wanted to be for real about that, that in today's world, that's what our reality looks like.

33:44But that's also because, you know, we use a lot of specialized tools. Like there are obviously, I know there's like all-in-one agent builders out there. Some of them are coming to Sastra this May. But for us, like, you know, I like to use the specialized tools. I just still find that the output is a little bit better. I like to use the best of everything in each agent versus an all-in-one tool that can build multiple agents. For us, it works better. For you, you might see success in using an all-in-one tool that could build different agents across the board. But for us, since we use very specialized third-party agents, this is the reality we live in.