Name: HockeyStack Agents: In Action
Uploaded: 2026-02-10T19:01:09.232Z
Duration: 46 min 52 s
Description: HockeyStack Agents: In Action

Transcript for "HockeyStack Agents: In Action": Hello? We're gonna be a couple minutes while everybody's trickling in. By the way, this is officially the most well produced webinar I've ever been a part of. There was a time about a year and a half ago when I had this small room that I was doing webinars in. I didn't have a microphone or anything. It was just a terrible MacBook microphone and camera. And now, Jake, can you zoom out? Yep. Look at this. Incredible. I think I know half the people here, but my name is Bura. I'm the cofounder and CEO of HockeyStack. I'm Hoya. I'm the product marketing lead here at HockeyStack. And I'm Camille, senior field marketing manager at HockeyStack. Yeah. They're the brains behind the event, and we prepped a lot. So very exciting agenda today. While we're waiting for people, if anyone has questions that are top of mind that they wanted to ask that they didn't get a chance to send through Slack, feel free to put it in the q and a right now, and then we'll get bored and answer all of them later. All right. Yep. Looks good. Let me share my slides. I think we've been talking a lot on, like, Slack channels, social media, etcetera, about Akistack agents. To give you the full background about this and, like, give you the full entire product demo, we want to organize this event. Essentially, today, we're proposing a new model for how the enterprise should work and how the enterprise revenue team should actually run. This new model that we will be proposing did not exist in concept at all before 2025. And then in 2025, we started experimenting with AI reasoning, and that really allowed us to develop this new model that we will talk about today. And what I'm really talking about is a model where humans and machines truly collaborate. Humans and machines truly collaborate with them applying their own strengths on the jobs to be done in a business. Today, when you look at how an enterprise runs, you're seeing a lot of chaos in tribal knowledge embedded in people's heads and that's being transferred over to people through coaching and team members not being ramped up fast enough, team members not hitting quota, etcetera. And all of this is because we're still using humans to do things that our software tooling can do better than us. What can humans do about well, basically, they're good at judgment, decision making, creativity, relationships. These are specifically human strengths. But a lot of enterprise work today requires data processing, like looking at lots of data and figuring out patterns and processes that work best out of that data. And a lot of enterprise work also requires deep consistency. You don't want to run a sales process across 2,000 prospects differently from team member to team member, but this is essentially what's happening today. So if we allow humans to do what they do best every single hour of the day and machines to do what they do best every single hour of the day, we think that we will reach a level of efficiency in the enterprise that we've never reached before. So, again, in the old world, specifically in the sales organization, there's lots of coaching to drill tribal knowledge into people's heads. We're trying to implement process by making sales reps fill out fields inside Salesforce and gating them from moving deals between stages to stages, and that is a lossy way to implement process. And then there are some pieces of knowledge that are only stored in people's heads. When you look at a sales organization and you look at the top 10% of the best sales reps, they have special knowledge in their heads about how to navigate their prospects that exist nowhere else. And we are not able to enforce process that that top 10% is doing across the entirety of the sales organization. We're doing our modeling for headcount on 70% attainment. We're doing our modeling on six month ramp times. This variance can be completely eliminated through AI agents, specifically implementing process for the organization autonomously. So in the new world, what will happen is AI will identify the optimal process for your organization without you having to know what the optimal process is. AI will be in context in the sales rep's day, and it will give you give the sales rep recommendations on what to do next. You have an 08:30 call and it's 08:25. You will get a recommendation on what to talk about on that call and what to say, what not to say, who's attending. This is something that reps are having to do manually. A lot of manual research, a lot of enablement required, a lot of support personnel required in the sales organization that we can just automate through AI. And we have this phenomenon because humans don't have this large scale data processing capability that they go into forecast meetings and they look at a deal, and they say, oh, this is gonna close next month probably. And you have your reasons for why you're saying that. And you're saying, hey. This is a best case forecast for me. And you have your reasons for saying that, but you do not actually know that it's gonna close next month. But because AI has access to large scale data processing capability, it can more accurately know when that deal is going to close, and you will have an organization where your sales reps cannot deceive you or cannot tell you incorrect things about deals or incomplete things about deals. Managers have full visibility into their deals. I'm still talking conceptually, but I'll bring it all together in a bit. There is two things we need to solve to go from the old world to the new world. The number one thing is the data gap, and this data gap is not just for customer data. Basically, this data gap exists across entire companies today, and we're seeing lots of other companies executing on fixing the data gap for, for example, people data. Rippling does that. And the same thing exists for customer data, but it hasn't been solved by any company at large scale yet. We claim that we have solved the data gap because, basically, we have that has been the only thing we worked on for the past five years where we wanted to build the most complete and accurate picture of a buyer possible, specifically in the B2B context. And we use that to power our marketing analytics products, which almost all the people, I think, here are on. And then we use that to power a prospecting SDR product called Account Intelligence, which uses the same data to give you, basically, intent scoring and recommended outreaches to accounts and does some workflow automation. But there's a second gap that we haven't solved or we hadn't solved until today, which is the process gap. The process gap is how your sales rep goes into that call at 08:30, and they know exactly what to say. They know exactly how to do discovery. They know exactly how to drive the next step. They know exactly what the next step is. How does that happen? A lot of these things that a rep does is not written down anywhere. And I talked to tons of sales leaders. I asked them, like, tell me lay down exactly how your best deals work. They cannot tell you in detail how their best deals are won. This is not something that can be held in the human mind. Because if you really lay down how a deal is won, it is a gigantic decision tree of things. There's a lot of factors to take into account when you're going into that meeting at 08:30 and saying, here's exactly what I should do in that meeting. And this knowledge, again, occurs in people's heads through constant coaching, constant iteration, learning that we can do at large scale today using AI. Again, it was impossible before 2025 to do this because the process gap can all only be solved with actual thinking and reasoning. When reasoning AI came along and it got really good around mid twenty twenty five, we started experimenting this with this, and we realized we can actually figure out a company's sales process better than their sales leader or even one of their AEs can tell us what it is or better than their internal VP can tell you what it is. And then I tested this actually, I personally tested this with my VP of sales. I wrote this program that figures out the sales process on in a very crude way in the most earliest fashion possible. And then it spit out a decision tree of things to do in any given deal to win and what not to do to win. And then I went to my VP of sales and I asked, hey. Give me all the documentation you have and all the knowledge you have. Like, still sit down and lay it all out for me. And I compared the two. This is not an attack on the VP of sales. This is, like, just the truth about how the human brain works. There's lots of nuance lost in what our VP of sales wrote down versus what AI was able to figure out. That is the reality. AI can look at your entire data. AI can look at every single VIN and figure out what the pattern is. And then it can also execute that pattern at large scale because you don't need a manager to go into a deal review, look at every single deal, watch every call, look at every single email sent. You don't have to do that. You can have AI look at every bit of information about a deal, look at the ideal process, compare, and give you the best next step at any given point in time. That is truly what we're trying to build and introduce here. So what HockeyStack Agents is trying to do is to optimize the sales process by a way of discovering the patterns that create the optimal sales process and to consistently execute it across your sales organization where your sales team members get the best next step on all of their deals and prospects at any given point in time. Essentially, that also allows your managers to be confident that your forecast is correct, that whenever they ask a question, they will get the exact right complete answer, and that you will hit your targets because now your medium level performers are up to your best performance levels. And, basically, you're executing on the ideal process on all fronts. That also has another side benefit, which is when you're determining the best next step for a deal. There's actually a lot of research document generation, etcetera, all these things, busy work, as we wanna call it, that a sales rep has to do, that can be automated today. Like, all your work is done on a laptop, on browser tabs, and in different applications. Today, if you tell AI to do research that a rep has to do before a meeting or for a business case or for a mutual action plan, it will do it, and it will do it better than the rep does on average on average. So we get the benefits of we can automate a lot of manual researching and document generation in the sales process. We can find the ideal sales process. We can execute the ideal sales process by telling reps what to do, and we can give managers visibility into essentially how each deal is progressing and how the book of business is progressing. The way to do this is, number one, you already have this, so connecting all your data sources together. Number two, discovering patterns about the sales process and encoding it inside of what we call AI agents. And number three, introducing your sales reps to Ocusac such that they get their next steps and they run their next steps. And I'll make this more concrete by going into the platform. So we'll first talk about what the rep needs to see, and then we'll talk about what is behind the scenes here. And what the rep sees is actually not that far off from what you would expect from any sales tool that you're implementing. So you have three things a rep needs to care about. Yeah. You actually have four things a rep needs to care about. One is you have what you need to do your tasks. Two is you have your meetings. Three is you have your deals. Four, we haven't built yet. We will build it. It is email activity or, like, general messaging activity. We will have that later. But we have those three components. So you have your tasks. At any given point in time, you have your prioritized task list. Okay. Acme has a deal going on. No. That deal was actually closed off, but they're reengaging. So and they have a new VP of marketing, so we should reach out to them. And we should apparently reach out to them with a call, And here are the talking points we would have on a call. And then you would have another activity here, call Snowflake deals stuck in legal. And it would tell you, basically, if deals are stuck in legal for x much time, then there's definitely a closed loss pattern. Therefore, you need to do something about it. Right? You need to eliminate this list risk. So when your AE has 50 deals on their book of business, they will do a prioritization of which deals they will focus the most on, and then they will execute the sales process perfectly for those deals. But let's say you have a deal like Snowflake, and the AE feels like, oh, the Snowflake deals I I don't know. It might not happen. Right? They will give less attention to that deal, and they will miss signals like this one where the deal is stuck in legal or, hey. How should I push this forward? Should I get a stakeholder demo booked, etcetera? Should I multithread? And a lot of these are basic things that the rep is missing, but also a lot of it is the nuance in which how you execute it. When I'm multithreading, at each deal stage, we discovered in our own data that there's different stakeholders that you should be engaging. And if you engage a stakeholder earlier, then you actually increase your chances of losing the deal. For example, we have a stage one and two. If you engage a CMO between stage one and two, which is very early on, you have a higher chance of the CMO shutting down the deal rather than building a champion and introducing a stage three to four. So these nuances are learned over time by running a lot of deals that AI can learn immediately. And instead of trying to teach it to your reps through traditional coaching methods, it just tells your reps, Hey, you should multithread. This deal is in stage three. Now let's multithread with the IT person or the CMO, etcetera. And here is the exact email you should use. And here's the exact person you should reach out to. Question in the chat. Would this integrate with Gong as well for call recordings and notes and things like that? Yes. Very good question. I will get into the integrations parts in a bit. But, yeah, we integrate with Gong. We integrate with your email. We're gonna integrate with Calendar directly as well. I'll talk about it in a second when I get into the work inner workings of this. Keep the questions coming. We love the questions. Also, by the way, quick break from the content. We have the podcast studio built out, and it's waiting for participants to come in and talk to us. If you wanna have a conversation like this, and we can go live we can go live. We can do a recorded podcast, etcetera, feel free to hit me up, or you just wanna grab coffee and chat. I love talking to customers, and I think I know half of this room, like I said. So feel free to hit me up, and we'll we'll talk. Anyways, back to the content. Meetings. So we have the fortunate situation of RAEs being constantly in meetings all day. And this is resulting in some unexpected droppings of the ball, which is when a meeting happens, I want the follow-up to go out immediately. And before a meeting happens, I want the rep to know exactly what the meeting should be about. Unfortunately, if you have AEs running lots of deals or trying to do both outbound and run deals, it's a suspect they have a lot of things to do, and you have a complex product like ours, you have to sell this in a way that you built the mutual action plan, you built the business case, You do a lot of work to get the deal through. You're just gonna drop the ball on doing some of this research and doing some of the follow-up. We identified if the follow-up is not done in the same day, basically, the close loss probability is I don't remember the exact number. I don't want to pull a number right now, but it shoots up drastically. And it's not a single digit percentage. It's a double digit percentage. And this is insane. Like, it is the most simplest thing to do to send a follow-up after a meeting, but they just don't have the bandwidth. So if you go to your meetings, you want your agent who knows everything about your deal to, when you get out of your meeting, to write your follow-up. And this is your follow-up. The follow-up email is identified through all the previous follow-up emails you've ever written. So it's in your company's tone, in your company's language, and it knows when you're going for from stage two to three. When you have the stakeholder meeting and you have rev ops in the room, this is the type of follow-up you send. She's asking, can brand and marketing I care about this question a lot. Can brand and marketing also get input into those emails to make sure that the messaging's on point? It's what we wanna say just because sometimes AEs or AI says things and, like, recommended emails that are kinda wonky, not super aligned with what happened. Yes. So everything about the agent contract, which is how you configure the agent, is customizable. And then you also have business context in Atlas page that you can upload into. I will go into detail on this in a bit. We care about this a lot too because this is my biggest pet peeve, actually. Like, if NAE sends an email out, I think this is even more crazier than, Zoe, what what you you are doing. But when there's, like, not correct spacing between paragraphs or, like, correct bullet point usage or the email is gigantic after a follow-up, like, I I just go nuts. I don't want this to happen, so I care about and then, yeah, for for meetings coming up, you want to prep for the meetings. This is a slightly more complex thing because it doesn't just rely on internal data. It's also third party data, but agents have access to researching on the web and researching other databases as well, other third party databases, such that it finds people, company context, and other things that you wanted to research. So for this one, a, we wanna here's the suggested agenda based on the deal stage, based on the type of people coming in. Here are the three questions to ask. And, actually, when I talk to sales leaders and I ask them, what is your biggest point of coaching in your team? The biggest number one answer by far is I need them to do better discovery and better qualification. So the three questions to ask seem simple, but it is actually solving a major pain point there because reps can just know how to do correct discovery through these questions. You can also customize this to be, like, branching questions. For example, what's the decision timeline? If they don't know, ask this, etcetera. You can configure this. Question. So Jason's asking, if we're a quickly growing company, our data is quickly changing, and we have a desire to move upstream, can we give the agent that context so that the agent knows what our goal is as opposed to what's just today? We should evaluate this a a little deeper, the question, because it depends on like, do we want to completely abandon the patterns that we have, or do we want to use some of the patterns but overwrite some of them? If you want to completely abandon the patterns, you would write a super customized agent contract. If you want to reuse some of the patterns, you would write basically overrides into the agent contract to enable this. I think with that said, I will get into the deal Copilot later. Since there's a lot of questions about, like, how is this configured, let me get into that one. So there's the data platform, which we now call outlet we we call Atlas. Basically, we ingest data from various third party, first party sources. This includes Gong transcripts. We ingest the full transcript and not only, like, the summaries, which means we can catch nuances in, like, what people say or don't say. And then you have all the usual suspects to CRM. You have sales engagement platforms. You have your custom data. It can be imported through Snowflake S3 or other data warehouses. Some other third party sources that you can add if you have industry providers, etcetera. We have, like, stuff like Outreach. And then we're adding, as we speak, a direct connection to inbox and a direct connection to calendar, which will add much, much more color. Right now, we get email data, email bodies through Salesforce and through sales engagement platforms, but we will add a direct inbox connection if you don't have those. And then you layer on business context. So there's one type of business context, which is just like, what does this business do? Which you should add, like, what are our key products? What are our target customers? The other one is, like, brand guidelines, like Zoya asked. It's essentially examples of examples of emails that you would add such that you give direction on tone. There will be more customization that you can do at the agent specific level. Go ahead. We do pain points, value props, persona info. Yes. You can dump all of that into the business context. Okay. And then the second thing we do, which if you're building in house agents I mean, a lot of companies are trying this. This is very hard, close to impossible to get, right, is the identity resolution and cleaning. Essentially, when Gong talks about a person or a meeting and Salesforce talks about a person in a meeting, they use different objects and different language. So we want to unify that language, and we want to talk about the same person, the same meeting, the same company. So that is what this cleaning step does. Because of our unique data model, which is not built on, these relational objects such as an account object or a contact object, because they're built on an event based model where the main primitives are event and the entity that performed the event, we can do a lot more flexible joining of entities and joining of events than a normal relational database would do. And this is the technical detail, but this is why a lot of times when you're trying to build in house agents, you run into issues of, oh, I have duplicate meetings. Like, I received two notifications for the same meeting, and they are different. Or, like, I it the agent missed some things that happened on the deal time line. These things will definitely happen if you don't have a proper cleaning step. And that cleaning step, I would argue, is only possible to get right if you have the event based data model that I described, which was fairly hard to build. It took us five years. And then you have the categorize step, which is essentially how you layer on more structured business context on top of your data. Sometimes you have things in your Salesforce that like, two things that mean the same thing, but one is outdated, and you didn't delete the field, etcetera. So we overcome that by adding some structure on top of the data, funnel stages, breakdowns, touch points, and then personas. Someone's asking, would the agents be able to look at our sales data and then give us the recommended deal stages and steps for each stage? Very good question. We do this internally a lot. I'll show you this exact agent in a bit. Yep. And then we categorize, so that is the data platform. Now we know what data goes in. The second step is the pattern matching step. Unfortunately, we don't have the most amazing demo for this. This is, like, more marketing oriented, but you can understand from the context of what this is. Basically, we look at the data. We have a conversion outcome. We have a set of behaviors and fit properties, And then we use those behaviors and fit properties to find out which ones of those and in what context do they most improve that conversion goal. Example, In a sales cycle, your conversion goal is close one. And then you also have other micro conversions, such as going from stage to stage. And then an example behavior is, hey, they replied to our email. What type of email? And you have email types. These are like other properties that you add. So you're able to convert the entire sales cycle into a set of behaviors and properties that you feed into an AI model such that it finds which ones of those properties improve conversion outcomes and basically sorts them by their impact. This is a representation of the same thing in a marketing journey, Like, how do we create pipeline? You can imagine the same thing in the sales journey. And then you can imagine the same thing in the broader sales journey and also, like, smaller portions of the sales journey. For example, you have an email and you're optimizing for email reply rates, a very, very micro conversion. But if you have an AI model that optimizes for email reply rates, you will get the best emails possible. Okay. So we have data platform. We have blueprints, which is the intelligence and this pattern finding layer, and then we have the agents, finally, the manner to show. So let's say we want to build a next best step generator for all of our open deals. How do you configure this is you have a trigger. It could be API or scheduled. You have inputs to the agent. For this one, my input is every single deal ever, so input type is none. I will ask the agent to find the deals. You have a set of actions that the agent can take to interface with the outside world. You have to define the actions here so that you have guardrails around what the agent does and is not allowed to do. And you can even configure this further. For example, you can give it Slack access I can send to everybody, but we don't want that. So you can configure which recipient, and you configure, like, what types of messages should it send. And then you have the main meat of the agent, which is the agent contract. Now the agent contract is essentially how the agent should decision and what output should it create. For OpenDeal Best Next Step, I wanna give it a general sense of what to do. Like, I want to analyze the current state. I would like to see the engagement signals. I would like to compare that to the winning patterns that are found through BLUEPRINT and then determine the next best action and then generate specific guidance, I give it a list of things to generate for the specific guidance. And then you write out this contract in natural language. You click validate. It gives you exactly what the agent will do. So the agent has access to data querying tools from our own database, and our own database is an aggregation of all of your data, right, plus all of the intelligence. And then it has access to the outside world through web research and other database searches, and it has access to the actions that you gave it. And based off of your natural language instructions, it will decide on how to complete your task. We want to decide on how to complete the task in a very structured way such that every time this agent runs, it does the same exact thing. Because the drawback with AI agents is that if you give very vague description and if you don't have this kind of workflow generation, you will run into inconsistencies, and that is the exact thing we're trying to prevent, like, in in the sales organization. We want more consistency. So we built this workflow. It knows exactly what tools to call, what to fetch, etcetera. Then the outputs let's run the test. It will run through that workflow on a scheduled basis and get health signals, etcetera. Let's wait for this to run. While it's running, I wanna talk a bit about debugging. This won't matter to your sales leader, your SAs, SDRs. This will matter to your GTM engineer or your DevOps people. You want to be able to dry run your agent, observe what it's doing, and you want to observe what data is accessing and how it's transforming that data, how it's making decisions, its exact thought process, and what actions is it taking. Right? So you can see all of that in here in this activity timeline. This is the testing UI, so your end users won't have access to this. And then it creates this output of this visual output of Iran. I created all my tasks for the reps. The tasks go into the task screen that I showed earlier. And then here's the output that I want to show you, the tester, the GTM engineer, to say, here's essentially what I did for you. Question on this on the activity. Is there a good way to audit that on an ongoing basis so that we can adjust the agent on positive ways? Yes. So two things. One is you can see the entire run history here. Every single run will appear here, and you can audit them. The other thing is we need to build like, that's what we're doing right now, like eval evals, automated evals. So that is our next step for automated audits of agent behavior. Yeah. And then when we're talking about, like, customizations to the sales process, there's obviously this knowledge of winning patterns in the agent set. But the agent contract is what determines what the agents does. So the like, you can add overrides or you can just say, hey. Ignore the winning patterns. Here's exactly what I want you to do. Here's also valid. So then we had a question about deal stages. So opportunity stage validator. This is a big pain point for us because RAEs, they have, like, lots of deals, and then they say, hey. This deal is in stage three, and its forecast is, like, most likely. But they've already gone into procurements. They're just, like, sandbagging because, like and then on the last day of the month, the deal closes, and it's it's moving from, like, stage three to close one. So this has been a very big pain point for us. That's why this is on the live demo. So Similarly, can it predict all the stages pre op and for customers? Say, for example, onboarding or ready for expansion, evangelist, stuff like that. Very good question. Let me get to that in a bit. We also have an example for that. So option stage validator, it has access to data sources, but you need a process for, like, what does a stage mean? So that's what you write into the contract. So, like, if it's on stage zero, these are the things that would be done. Again, this is all natural language writing. And then it would determine based off of this what should the stage be. Similarly, it would determine the forecast. The forecast is slightly more complex because it needs depend on the winning patterns. So it looks at the winning patterns and then compares, finds gaps, and then says, hey. Here are the forecast changes that I need to do. And then this is more for managers. It gives you, like, challenge questions. So it's like, hey. Ask the AE these things because we don't want coaching to stop. Like, we don't want AEs to stop improving. We want the agent to derisk deals, but at the same time, we want the manager to be able to see what the AE would have done wrong and challenge the AE. Right? So just giving you those things. And then in terms of the post sales pipeline, because, like, I'm I have an engineering background, I see everything as just data points, and I do not see a difference between the presales and post sales pipeline. So it's all the same data for us. It's all the same primitives. So then you can build the same thing for expansion opportunities, customer success playbooks. We haven't implemented this one internally yet, so right now, we're implementing it. So this one is like, hey. For for customers, what does at risk mean? And then for expansion opportunities expansion opportunities, an interesting thing is because we have custom data connections, you can plug into product data. We have because we're a web app, we installed the HockeySec script into our own products. So we already have access to product data through the script. But if you guys don't have the script installed on your product and you don't want to install it, you can just plug a custom data source like Snowflake, and then you can pull this from there. And then it looks at product data to find expansion opportunities. Example, hitting API limits. Right? And then with two and a half minutes left, we'll do the question that's racked up a couple of votes as well. Oh. Are the agents only planned on the sales or AE side of the house, or are agents also coming to the marketing analytics side? Also, can we pull sales info into marketing analytics and actions? So while I'm answering, just please tell me more about the last part of the question, like, sales info into the marketing side, and then I'll answer your question while you're typing that one. So we already have a marketing agent. It's called Odin. Odin is a fairly generalized marketing agent, and, basically, it had been before just like prompting q and a. But because we added these scheduled prompts, you can now run these things at a schedule. And, for example, I have an email coming to me every 9AM, Monday, about if Amulya messed anything up in the paid advertising. And, like, sometimes we flag risks because of that may sometimes, like, our spend is not going into the right direction, and I need to know that. So I I get these scheduled prompts in my inbox. So this is exactly the same thing as an agent. We do have to do some work of unifying Golden with agents. Right now, how it works right now, how it works is basically you have an ODIN tool for agents, which is not the correct infrastructure, by the way. Like, we need to change this and unify them, but you you can have an agent that asks Autin a question, and it gets a response back, and it runs like that. So we have this, but we should unify the two products soon. Question just under a minute. Oh, somebody actually gave us more context as well. She's saying pulling sales data into the marketing side, like, hey. We're seeing a new pattern and winning deals, and we need more content like this, which actually ties in with the other question as well, which is, can it help us identify gaps in our content if we have consistent questions that the rep says we don't have good content for instead? Yeah. So this is quite similar to the sales enablement agent example here. So it's like, look at all the deals where we won against Gong and find out what are the objections that come up in those deals, how did we answer them correctly, and then summarize it all into a document. And this is exactly what your reps are asking of you as a product marketer, like some part of your job is automated. By the way, for marketing, for example, for field events, I was traveling to New York City, and I wanted to look at my entire customer base and find VPs of marketing and sales that I needed to meet with. And you can do the same thing with other events that you're organizing. Like, you can just draft up an agent to do that. Literally, you tell the agent, look at all my customers, find people matching this pattern, give me a list back, and it gives you the list back, and then you message them and ask them for coffee. At least that's what I did. I think we're a little bit over, but thank you guys so much for hopping up. We have one more question. If people are willing to hop on, there's been a couple of votes for this question as well, which is, Can you walk through costs and how the credit consumption works? Can we limit credit usage by access rules? Credit consumption is pretty aligned to the actions that the agent is taking. So if you have an agent like this one, don't get scared by these big numbers. Like, we will the the credit numbers look big, but the price per credit is pretty low. So you have how much analysis the agent is doing, how much research is it doing. And there are things like research that are most costly than more more costly than internal analysis, and then creating follow-up task is even less costly. Like, it's basically adjusted to the action itself and how costly it is, and then it would be per, like, data point analyzed. You can estimate costs for any agent you're running by going into the costs tab. And in terms of limiting costs, we don't have that capability yet, but that's a fairly good feature request. So we'll we'll keep that in mind. Awesome. Well, there's questions coming in still. As a follow-up, I'm in all of your Slack channels. So, like, feel free to keep the questions coming as they pop up, and I'll make sure Bora answers them on video for you. Or we can do a follow-up webinar to one of these. Yeah. I'm happy to talk to anyone. I think I have, like, an hour and a half in me of talk. I'm happy to talk to anyone. Feel free to message me on LinkedIn. Message me on Slack. Email, bugra@AUCUSAC.com, and we'll talk. Thank you guys so much for hopping on. Again, if you have questions, feel free to reach out to Bora or any one of us. We'll make sure they get answered, and hope you have a good rest of your afternoon.