Name: Analytics where you work: Hex Agent in Slack
Uploaded: 2025-11-05T20:16:59.797Z
Duration: 58 min 20 s
Description: Analytics where you work: Hex Agent in Slack

Transcript for "Analytics where you work: Hex Agent in Slack": Hi, everyone. Welcome. Welcome. Sorry about the one minute delay there. Appreciate your patience. We will get started in about a minute or two and just give folks a chance to trickle in. For those of you who've joined our webinars before, I just wanna address the elephant in the room, which is that we are using a new platform for hosting these live virtual events. So we're no longer on Zoom, and we hope that this will be a better experience for all of you. But we will, as usual, send a post webinar survey. So we'd really love to hear your thoughts after about what you think about the the new setup and all. And so while we are waiting, would love to see where folks are tuning in from. So please feel free to share that in the chat. I'm based in New York City. We have some folks traveling into our New York City office this week. I know Sarah is actually taking the call from our office, Hex office in in the city. And so, yeah, we'd love to hear where other people are joining from. I'm seeing oh my gosh. Someone's joining from Egypt. That's quite the time zone difference. Thanks for thanks for tuning in. And then also, let's also launch a quick poll as well just for folks have, interactive question to answer. So there's a poll tab that should have just popped up. You should be able to see. And we just have a one question poll, and would be really curious to hear, where folks in your company are going to to to ask data questions when when they need help from the data team. So Okay. Everyone the most it out is, is not triggering for anyone. Yeah. I mean, I guess, Katie, could you speak to where you feel like the questions, are that you get at Hex? Where are people asking you all the questions? I mean, that that's gonna be a little bit of a spoiler, but we we see a mix, although there's definitely skews in certain directions that I'll cover later. Amazing. Okay. Cool. Let's just take a look at, hopefully, if people had a chance to to submit their answer. So let's see what the results are. Wonder where it's popping up. Pulls tab. Well, I am not seeing the results show up here in this tab. You know, technical difficulties perhaps with a new platform. So we are just going to pretend that, like, lots of people chose Slack, and that you maybe have a data channel in Slack, question mark, where people are submitting a lot of their questions and they run into roadblocks. They need help from the data team. I imagine probably you get questions too sometimes over DM when someone has maybe, like, a dumb question they're embarrassed about and don't want to ask, publicly. I sometimes am guilty of that myself. I know I should put it in a public channel, but sometimes DM, someone instead. So that is, the topic we're covering today is we will be talking about the, Hex agent in Slack. So hopefully that wasn't too much of a spoiler. Next slide, please, Sarah. So to quickly touch on intros, so I am Nicole. I'm your host for today's live event. We have two really wonderful speakers today. Sarah is a PM here at Hex. She focuses a lot on our, like, self-service workflows, including the Hex agent in ways that it can help you, answer questions and self serve. And then Katie is our head of data. And, of course, our wonderful data team is always dogfooding, our latest and greatest text features. So very excited for her to join and really share some of her tips and tricks. Next slide, please. So I'll keep this brief, but just wanna share how we're gonna spend the next forty minutes together. So, Sarah will start by giving us an overview of Hex, and really our unique approach to AI. And then we'll jump straight into a live product demo of the Hex agent. And Sarah will not only cover how it works in Slack, what that looks like, but she'll also show you how you can actually dig into the results to really audit and validate the agent's work. And then after the demo, Katie will share how our data team at Hex actually approached rolling this out, as well as best practices that she would recommend for data teams looking to use this integration to really scale self-service. And then we'll plan to end with about ten to fifteen minutes of live q and a. So please put your questions in the q and a box. That's, like, a little tab to the through the right of the chat box, and we will try to address, some of those questions as we go along and then get to the rest at the end. And then the last note that I'll share is that the session is being recorded, and we'll make sure to share the recording with all of you over email afterwards. So with that, I will pass it over to Sarah to take it away. Cool. Thank you, Nicole. Hi, everyone. Super excited to, talk today about, some features that we've launched recently. Like Nicole mentioned, I'm, on our product team here at HEX, and focusing on, workflows for less technical stakeholders and making sure that HEX is a great place for them to ask and answer their own questions, and, of course, keeping the data team central, for, helping them answer those questions as well. So, we have a lot of new folks tuning in, so I'm gonna start by giving a brief overview of HEX. For those of you more familiar with what we do, HEX is a connected platform for using AI to work with data. It's a place where the data team can do deep dive analysis, build data apps, and interactive dashboards, and it's also a place where, we allow you to curate context to unlock self serve for, business stakeholders. All of this is done with AI agents that are deeply integrated into your workflows, so you can ask questions in natural language and get those insights faster. And we do that differently, depending on where you're at in the platform and what your workflows are. So I will dig into that a little bit deeper here. But zooming back out a little bit, what makes Hex really powerful is that our platform, is built to follow and, like, supercharge the data analytics process. So we have this notion here of this, like, cycle that is kind of creating compounding value for you as a data team and, value and compounding value for your stakeholders and the questions that they're asking and the answers that they're receiving. So first off, in Hex, the data team can use our notebook environment to explore, like, new deep, like, gnarly questions, in a very technical platform where we can combine, SQL, Python, and AI, to, allow the data teams to, like, focus on strategic data problems that they can use to drive decisions. Second, they can actually, like, publish those those results and, like, canonize the output of the hard work that they're doing, as data apps and, semantic models. So that is, like, helping, the data can create this context that is, everlasting in the workspace and compounding over time. Adding all that context in the workspace is then what unlocks self-service for the rest of the organization and enables everyone to be able to use AI, and UI, to answer their own questions and self serve. But, of course, we know that there's this is not like a perfect world where we're always gonna be able to feel, like, curating the perfect amount of context to answer every single question. So once the, stakeholders, hit a wall with their self-service, we have a really seamless way for them to, like, volley that question back to the data team, where they can use, a familiar, like, Hex environment to continue working and continue, answering, the difficult problems. So we refer to this internally as, like, the, virtuous cycle. And when all these pieces work together, answers get better over time because every asset the data team is creating just becomes more and more context that makes, both humans and agents more accurate. And, yeah, Pex is really good at, like, making this, compounding value possible in ways that, is really difficult with more fragmented workflows, or, like, standalone systems. Cool. So how do we actually, like, enable this virtuous virtuous cycle in reality, and, like, make it actually happen in Hex? So we offer several, different capabilities, like, along this process, that bring, everything together into one platform. So for deep analysis, like I mentioned, we have our powerful notebook environment and our notebook agent. So this enables individual analysis and analysts and, data teams, to, have an environment where they can, again, like, get really deep, use an environment that is going to be flexible enough to allow them to answer deep and powerful questions. And, yeah, users use the notebook for this. And many of you probably know, Hex for our notebook product. And while that is, like, the grounding piece that ultimately powers this entire virtuous cycle, We do a lot more than that, so I'm happy to talk, and excited to talk more about that later on. But, yeah, as the data team, produces these insights, they will publish them as interactive apps, that run on the powerful analysis that the data team performed, with the notebook. And, you know, these are, like, reusable, assets that that, people can revisit over time, and just become part of that, like, trusted context in the workspace. We also, for those who have, like, watched previous webinars, you you probably know this, but we, launched earlier this year the ability to define semantic models within hex as well. So that's another way to add context to your workspace. And we actually have a modeling agent there as well that will help you, create semantic models based on existing, content in your workspace. And then for self-service, we have a visual point and click, drag and drop, UI for no code data exploration. It's really great for, like, very data savvy, business stakeholders who really wanna get their hands dirty with, their own data for people that are, like, really familiar with, the specific data that they need to answer their questions. And this is, like, traditionally how self-service has worked, in the industry in general. But really what we see as, like, the biggest unlock for self-service is, threads, which we launched a little bit over a month ago. And threads is basically our, natural language interface to allow stakeholders to chat with their data and get trusted answers back, in a conversational interface. And I will dig a little bit further into that soon. So, again, the focus of the webinar today, is that we're actually, like, expanding this out even further, and we're making self serve even more accessible by bringing threads to places where your stakeholders are already working, which is Slack. So just continuing, the theme of allowing you to use Hex agents wherever you work. We also launched, our MCP server, last week, which allows you to hook into Hex from Cloud, Cursor, and really any tool that supports MCP integration. But today, our focus is going to be on the Hex agent in Slack and the workflows that that unlocks. And I'm really excited to get into that. So with that, let's get right into a demo. I am going to swap over to my full screen. So just one moment. I'm also already starting to see some really great questions in the q and a that I'm very excited for us to to get to as well. So I'll try to weave some of those into, the demos there as it makes sense. I think I might be having any issue with sharing my screen. Just a moment. I'm gonna try refreshing my page. No worries. Again, trying out a new platform. This is our first webinar on, yeah, a different tool. So there'll always be some hiccups here and there. But I'm really actually loving this new interface for for the q and a, and it's nice you guys can actually upvote, each other's, questions so we can get a sense for what everyone else is also thinking about. Let's see. Is there a pop back out, back in? Hopefully, we can get Slack up and running soon. Also, actually looking at the chat, it's, like, really fun to see where everyone is tuning in from. I noticed there are some other folks in New York City, which is fun. Feel like you should just drop by the Hex office sometime. We love when our customers drop by. We always have, like, swag and merch that we love giving out as well. So, yeah, if you're in New York City or SF, please do, like, reach out and up. Same same is true for San Francisco. We've got a pretty good contingent there, and there's a whole closet full of swag that we can potentially distribute to the masses. For all of you who, are are in Europe and beyond, I'm very impressed by your stamina for crossing time zones. We're very happy to have you here. Thank you for coming. Yeah. We so appreciate that. And then, hopefully, for the folks who weren't able to make it, we typically try to schedule the webinar at this time because it, like, works relatively well for both East Coast, West Coast folks. But, yeah, not always great for folks in Europe, but we will send out a recording. That way you can also, like, play it back later if you can't stay for the whole thing. Hi, Sarah. We see you again. Any any luck with the screen sharing? How's this looking? Yes. We can see your browser now. Okay. Amazing. Sorry for that delay, guys. Oh, good. Okay. Might need you to zoom in just a little bit if possible. And then I think folks can also zoom a little extra on this platform as well if you're still having a hard time seeing. But, yeah, that is better. That's definitely better. Okay. Amazing. Cool. So, thanks for dealing with that delay, everyone. Cool. So, I know I mentioned where your the focus of today is going to be, our Slack integration, which I'm super excited to get to. But first, just for folks who might not be aware of what our, threads interface and our threads features look like, I thought I'd do a quick recap because I think we'll ground the rest of our conversation. So I mentioned, a little bit over a month ago, we launched our, like, chat interface, for stakeholders, and this is what it looks like. So here we have, a prompt bar where users can, enter questions. So in this example, I asked a question, do we have any deals that have landed at over $1,000,000? By the way, this is dummy data. But, basically, we have, an agent network here that is going to, crawl over all the data that I have access to. It's going to prioritize certain, data sources. So it's always going to search for, model data first. That will be the top priority. If there's any semantic models that are hooked into my workspace, those will always be treated as source of truth. And I will also, the or sorry. Not me. The agent will also, prioritize endorsed data as well. So we have a bunch of different, curation methods, in Hex where you can, promote and kind of encourage both humans and agents to, access certain, warehouses, schemas, data tables, and things like that. But, yeah, I can I can kind of scroll through, all the tool calls that the agent is doing and really understand, how it is going about solving my question, how it's going about thinking about, how to answer the question, and, just all the the logic really that it's that it's using to, assess the question that I asked? After it is done thinking, it will give me a little summary of its findings. So here, it's just giving me a a straight answer that we have not closed any deals, over $1,000,000, and it will summarize, the findings that it that it did find when looking at this specific semantic model. So it's giving me a couple of related metrics. It's giving me a visualization that I can interact with and understand, and it's also giving me a table here where I can see, the 10 largest, deals that were won. So something specific about threads. You know, I asked a pretty pointed question, but the agent, really gave me a lot of context to answer the question. It's showing me the data sources that it's using, and it's giving me, a lot of context, that is useful for me when I'm, receiving an answer, and I can kind of, view the the response that it gave in, yeah, in context with, with this other information. Sarah, can you jump in real quick? If you scroll back up to the top, I noticed, like, the way you asked this question and and framed it is, like, very much sort of how you'd probably even ask, like, a member of our data team if you had this question. I'm noticing someone in the chat, had a question about, like, you know, what is the ideal way to sort of prompt the Hex agent? Like, is there a particular format that you feel like optimizes responses? And, you know, just curious if you have any, like, best practices there. Yeah. Great question. Honestly, we our goal is that, we shouldn't have to tell stakeholders how to prompt and answer their question because, that's a high high standard and, we want we want this to be as if you're interfacing with a human or someone, that was working on the data team. So there are certain things that you can do to optimize, the results. So, for example, like, if I had a specific data source in mind that I wanted to tag in, I can search for it by using, like, the at symbol, finding, like, a specific data source. This will just, like, narrow the context for the question that I'm asking and, like, ensure that the agent is gonna use a specific data source. But, you know, in reality, your stakeholders, just making an assumption, probably don't know where all the data lives. They might not know the names of all of the warehouse tables that they have, available to them or, semantic models and things like that. So we don't make that a requirement, but it is something that users can do to, further specify specify their prompt. I'll jump in briefly here to also add that it's quite similar to prompting a coding agent in that if you give it more specific instructions, it's more likely to do the specific thing that you asked for. So if you're looking for particular segments to be included in an analysis or you want a certain time frame, definitely specify. But one of the things that's really nice about the agent is that this is an interactive workflow. So you can always tell it, like, oh, you did that wrong or, like, I want something slightly different. Can you redo it? And it will do it quite readily. Yes. Great point. Yeah. So I can always, of course, add, follow-up questions and continue prompting to get closer to, what I'm looking for if the first prompt was not sufficient. Cool. So yeah. We launched this again, like I mentioned, a little bit over a month ago, But, we really wanted to make it even more accessible to for stakeholders. You know, sometimes even, like, opening Hex and navigating your way to this UI is, quite a lot to ask. So we really wanted to bring this super powerful workflow exactly, where folks are already asking data questions and interfacing with the data team. And that is Slack. So let's hop over to Slack here. I'll do a quick demo of what this looks like in practice. So I am just going to type in hex, the name of our Slack bot. I will tag it, and then I'm just gonna ask a prompt or ask a question the exact same way I would enter a prompt, in, the threads UI. So I might ask, what were the five most recent deals we've closed and their land size. Cool. So I will kick that off. Right away, I'm going to get a little indicator, in form of an emoji that the agent has received the question and is going to start, answering it. And, basically, I'm gonna get a notification when the agent is done thinking and has a response ready for me. So we will revisit this question, later on in the demo. But, for the sake of showing you something right now, I'm just gonna open up this, thread that I kicked off, earlier today. So here I asked a question, what lead sources are working well for new customer acquisition? You can see, we got this same, like, starter message here. So just a confirmation that the question was received. And then a couple minutes later, it looks like two minutes later, I got this response from, the agent, and the response is just in the form of a message that's appended, in the thread. So we obfuscate a bit of the thread's UI, so we don't have the thinking text. We're really just giving you the response that the agent came up with. And I can scroll through here. We've been, like, attached, the visualizations that we've, created as part of the prompt as well as, the summary. We have some recommendations in here. And then, actually, something I forgot to mention in our, in that, thread that I showed off earlier is that, in threads, we're always also going to deflect to existing projects in the workspace if the agent assesses that the, projects it finds answer the question or maybe they could get you close to answering that type of question. So we really wanna avoid, like, reinventing the wheel if we don't have to. We'd rather point you to trusted, sources, and assets that that the data team has already created, if they exist. So here, it looks like, the agent found two, projects and hacks or published apps that, it thinks might be relevant and linked them here. Since this is a thread, like, under the hood, I can again reprompt for follow-up questions. So here, I just prompt it again, and I said, for deals with a trial lead source, which ones were our CEO involved in? And, again, question received, and I get a response, about a minute later. Here, the response was that, the agent really didn't have sufficient data to answer the question. So it's explaining to me, the data that it does have available to it and where the gaps were, with, you know, what the information it did not have available to it that it would need to answer the question. So, yeah, the real value of this, honestly, is that, first, the conversation is taking place in Slack where users are used to to working. So like I mentioned, the barrier to entry is much lower. And secondly, now that we're in Slack, I can just, like, have conversations with other members of my team. Maybe I have, the data team coming in and, assessing some of the responses that I'm getting. In this case, we can see Nicole jumped in. She is giving me a suggestion. Maybe we should ask rev ops, about tracking CEO and product team involvement in Salesforce. Great suggestion, Nicole. So, really just kind of, like, bringing the threads agent along with workflows that people are familiar with and making this, like, a super collaborative experience. Let's say, Katie or someone on our data team was taking a look at this thread and, maybe I asked for verification or something or, there was something that needed to be, like, audited a little bit further with this thread. So Katie or anyone really can, click this view in Hex button, and that will just open up, like, the thread representation of this in the Hex UI. So since, again, I mentioned this is the thread under the hood, we can just see my question, the full response, and then my follow-up question that I asked, will be further along in the thread here as well. So here if I will jump in with a a couple of quick kind of Slack related questions. I think that you'd already sort of touched on this, but just to reiterate, you are able to interact with and kind of progress the chat in the Slack thread itself. Right? So, like, folks don't have to open it up and necessarily continue it in Hex. Like, they could have that back and forth directly in the Yeah. That's that's correct. You can have the entire experience in in Slack if you want it. So if I, the the requirement is just that we have you tag the Hex again. That way, we're not, like, adding a bunch of, the agent isn't just, like, accepting any normal conversation that you might be having in the thread. That's, like, commentary on the results. So as long as you tag the, the Slack in your follow-up question, you can you can continue the conversation purely in Slack. But it will be represented as, messages in the thread UI. That makes sense. So that way, you could still have the back and forth with, like, a teammate and be addressing some of maybe their questions within the thread. But if you specifically want the Hex agent to kind of follow-up and, kind of answer a follow a following question, you just need to, like, tag it in the thread. And then if you wanna see the thinking process and everything, then that's when you would go and open it in Hex where and this is the UI right now that you're in where you could then, like, dig into the thinking process. Yeah. Yeah. We wanna include as much detail as possible in in Slack without it being overwhelming. So there are certain things that we're not including like, like you mentioned, the tool calls. So to get that information, we're gonna have you go to Hex, which is like a much richer UI where you can really interrogate, how the agent arrived at the question, and, yeah, what data sources it's using, etcetera. So once I'm here, I can I can see all of the tool calls that we used, the data, the, existing projects that it pulled from? Awesome. And then, could you remind, the audience, like, what model is powering, these, like, AI agents under the hood? Yes. We are using SONNET, to power the, the agent. Okay. And then, on is there, like, anything on the road map for folks to be able to kinda select or try out different models down the road, or is it just like SONNET 4.5, for the foreseeable future? Yeah. We're considering that. We'd love to hear people's opinions or or thoughts on how they would want to control, the model that we're using. For now, for simplicity's sake, we, we choose the model, and we we choose the the the ways that we're going to get, like, the most accurate responses and based on our own internal testing and evaluation. But, very open to feedback on that and how we evolve our our AI products in general. Great. Amazing. Yeah. Sorry to derail. I'll let you get back to the demo. Yeah. No problem. Cool. So let's actually head back to that original question I kicked off. Cool. So I asked a pretty straightforward question straightforward question. What were the five most recent deals we've closed? And here, I've gotten a a pretty concise response, actually. So I really just have this table with the five account names. The table is embedded here, and I'm getting a quick summary of, the land sizes, which I ask for as well. Here, it's telling me the data source that it used to to arrive at the question or sorry, arrive at the response. And, again, if I want to audit this in Hex, I can just click view thread. And one other thing I'll touch on is that, if for for some reason the, or for whatever reason the data team wants to, really, like, closely observe exactly how the agent, arrived at this response, a thread is just a Hex notebook under the hood. So if I click this save as project button, I will be able to see, firstly, a summary of the thread, the thread's, prompts and responses. But secondly, I'll be able to see exactly which cells the agent produced, to arrive at a response. So in this case, I'm actually not using a semantic model. The agent just, wrote a super simple query against this endorsed, table that it has access to. This is a very simple example. You can imagine that, closely auditing something, more complex would be a more compelling reason to open this in a project. But this is really just meant to demonstrate that, kind of like closing that loop where I talked about earlier. You know, there are going to be certain points at which, stakeholders kind of reach the limits of what they can self serve and need to, like, volley, a question back to the data team. This is, like, a really seamless way to do this. The data team can, like, use this as a starting point. They can change the way that the, SQL query is written if there's something that needs to be adjusted. You could even, like, take this as a starting point, publish a project, and then you've, like, canonized this result, and it will just be used as context in the workspace for future questions that users answer or users ask. And it can it can be used as context for responses in the future. So, yeah, that concludes, our quick demo of, threads, Slackbot, and, honestly, a bunch of other recent features that we've launched. But, yeah, I would love to kick back over oops. Sorry. Sarah, real quick. I know you're about to hand it off to to Katie, and we definitely, are excited to hear from Katie about sort of how we rolled this out at Hex. I might have just another quick maybe two follow-up questions I do think are particularly relevant that we can maybe just answer now before we throw things over to Katie. So one was about, you mentioned, of course, threats being able to actually reference and use semantic models. And I think someone had pointed out that, they're aware that, you know, they can adjust that in the settings. But is there a way for them to have the Slack integration, for example, use semantic models only, but then for threads and Hex to be able to actually use all types of, like, data sources? Like, is there a way to kind of, like, I guess, further curate that a little bit more? Yeah. That's a great question. I would actually love to dig into that part with you, but maybe we can take that offline. But, yeah, there is not a way to do that right now. We'll just respect the existing thread settings that you have configured and just, carry those over. So you can think of this as, like, the exact same workflow just from kicked off from a different, surface area. Amazing. And then another question that that I definitely wanted to address sooner rather than later is I think there are some folks who are very excited to get their hands on this, and I'm curious, what plan types it is available on. So maybe if you could address that for both, like, Slack and also MCP. Yeah. All these features are available on our team's plan and higher. Yeah. Yeah. And then as long as you have an editor or explorer seat, then you can ask the questions. Right? And then with the viewer, then you just have to view only access? Yes. Correct. You need an explorer seat or higher to kick off threads and, basically do most, self-service in HECS. Awesome. Wonderful. Okay. Well, I'll hand it over to Katie then. Thanks, Sarah. Cool. Cool. Thank you. I will try to move through my slides relatively quickly here without shortchanging them so we can do a little more q and a at the end. But hello. I'm the head of data here, and I am here to talk about what it's actually like being on a data team that has rolled out threads. And I'll start by contextualizing and just talking about what the data team at HEX is like. This is the entire thing. There's only five of us counting myself. Two analytics engineers and two analysts, which is not a crazy ratio, given our company size. But what is unusual about HEX is there are many very data savvy and data hungry people, and everyone is an editor at HEX. So there's a lot happening all the time for us to keep track of. And we're not that different from other data teams in that we don't wanna spend all of our time keeping track and tabs on what everyone else is doing. We really do want to spend more time on high ROI work, such as the things pictured on this slide. If my head of product asks me what an OKR target should be, I wanna talk to him about how we hit it, not just pull a number for him. We have lots of infrastructure that powers tons of our business processes that we wanna improve and scale, and the data models as well that are used in Hex. We wanna experiment with new workflows and and learn new tools and become better at our craft and adapt to the times. And, of course, we wanna dogfood all of Hex new product features, because we love HEX here at HEX. Before, we had our Slack bot. We were in a pretty similar situation to a lot of different data teams. If you look at these questions on this slide, they're pretty standard. I've gotten questions like this at nearly every company that I've worked for. And as you can see, some of them are, a little bit of a a can of worms, where there's a lot of sub questions and, you know, there's gonna end up being a lot of follow-up questions. This is very hard for data teams to keep up with, and our data channel, where folks reach out to us for support was really inundated with a lot of incremental asks like this before we rolled out our Slack bot. But, the Slack bot has really changed thing things for us, in ways that are kind of hard for me to believe, having spent as much time as I have working in the data world. We basically decided we were going to roll this out, in a somewhat controlled fashion, but not too controlled. At Hex, we have two values that I think are important to emphasize, which are the every second counts. And for those of you who like Dune, the other one is fear is the mind killer. So we decided we were gonna really get this in front of people internally and see how it worked to learn about it rather than try to come up with every contingency plan in advance. And I I do generally recommend being aggressive as you roll it out, at a company, but there are, like, some small things that you can do to make it go a little bit smoother, and this is similar to what we did at Hex. The first step is just testing it yourself. As the data team, you're kind of responsible for the quality of things that come through Hex. So you really want to think about whether you trust what it's giving you, and you also want to think to yourself about how conservative you wanna be as you roll out something like this. Some companies, there's greater consequences to someone getting wrong answers or there are certain teams that you may be we wanna make sure they always get consistent answers, like finance teams, for example. And you really wanna think as you're testing this, how often is it wrong? Am I comfortable with it being wrong? Although you will probably find in most cases that it is pretty accurate, if you've done the relevant work, to prepare the agent, which I'll talk about in a little bit. Once you believe it yourself, start rolling it out to some trusted business partners. These might be your teams that kinda can self-service on their own. They're they're more savvy. They spend a lot of time looking at data and know when to be skeptical of answers being, maybe a little fishier, maybe a little bit off so they can help you find edge cases in areas where your agent needs a little more context to perform well. And there are also people who can then go and be your advocates, when you roll this out more broadly. And then finally, you do roll it out more broadly. You give it to as many people as makes sense to have have it, at your company. And I as I said before, I do recommend doing this slightly before you feel comfortable doing it. You will learn so much by seeing how people actually use your Slack bot, and the best way to learn. I mean, like, we're we're all data people here. We love data. So the best way to understand behavior is to actually look at what you see and observe, when you give a feature to someone. In terms of best practices for actually making your Slack agent helpful, and effective and accurate, the most important one is giving the agent context. And context is sort of a a different way of saying data, although it's also more more than just that. But there are a lot of features in Hex you can use, and I've got them listed on this slide kind of in order of how easy they are to roll out. Endorsing tables is just looking at tables in your data warehouse and telling the Slack bot to either not use them at all or to prefer them when it's generating queries or or answering questions for people. There was a question earlier in the chat about, endorsing projects. You can give them statuses, not properly endorse them, but statuses will allow them to be surfaced and referenced more by the agent. Next up, once you've got endorsements in, you wanna update the rules file that's available in your Hex workspace. And for those of you who are unfamiliar, they're just a markdown file where you can give your Slack bot, or the Hex agent generally, just business context where it might be, we have an offset fiscal quarter. So please, when I say q three, actually reference these months instead of what people normally think, for example. And once you've gotten good mileage out of those, start dabbling with your semantic models. Or if you already have semantic models in something like the d t metrics flow, you can load those into Hex. All of these also really benefit from having well documented tables in your data warehouse, and metadata on your columns because the agent likes to read metadata, and it performs much better when you have clean metadata. And in general, if a table is really easy for someone who is a person to use, it's gonna be easy for your AI to use. Another best practice I recommend as a data practitioner is to lead by example. So when you are posting or when you're seeing a lot of posts in your data bot channel, you should ask questions in there too. It will really normalize and help people feel comfortable with the fact that they are asking questions in a public setting if they see people on the data team doing it too, and it helps teach them how to prompt better and how to use it better. And also lots of people benefit from seeing what other people are asking. Another thing you really should do is set aside time to monitor what your, agent is getting asked. And I do recommend carving out more time than maybe you expect for this, because reviewing threads, like, some of them are really long. It can be hard to to really understand everything that happened in it without carefully reading through. I mean, you saw some of the the thinking traces that Sarah shared earlier, and it's really worth reading through those and seeing what people are doing with the agent because they're gonna do things that you don't expect that are pretty cool. And then finally, seek feedback from the people that you've rolled it out to and iterate. People may not proactively give you information, about what they're using the agent for, or or whether they like it or whether they think it's maybe a little too eager with it. You're absolutely right. So it's worth just sending out a survey or directly asking people that you work with in one on ones, and getting a sense of how they are feeling about your agent as they're rolling it out. I can just in real quick. I think that yeah. That was really great. You pointed out this process that's worked well for you for getting, like, feedback internally about, like, the responses that folks are getting from from the Hex agent. And I think someone had a quick follow-up question about the rules file. And could you share a little bit more about just how you think about the rules file and also for some folks who are, let's say, like, also using DBT, does a markdown file in a DBT repo do the trick? Or, like, how how do you kinda think about, I guess, like, navigating the rules file and where this information should live? Yeah. Great question. I'll start by saying that context engineering is pretty new. So no one truly knows the best practices here. I will say a markdown file in your dbt repo will not be read by HEX, but you can version control it there if you want. There is a native editor within HEX, and you can upload files from elsewhere in there. Having a version controlled rules file within HEX is something that is forthcoming. In terms of what goes in the rules file, we have found it as useful for things like I I specified where it's like there are certain dates that are important. For example, we have a list of product launch dates in our roles file at HEX so that when someone asks about a particular product feature, the agent can know when to constrain dates to, which is really helpful. We've used it to specify, like, if someone is asking something that's financialized, define a metric this way versus everyone else. It's a good way to clarify metrics. And another really interesting use case for it that we are intending to experiment with more is it's a good first pass place to define a really new concept in your ecosystem, before you have time to go and properly model it in, say, dbt. So, for example, for threads itself, before we have had a chance to properly put that in a data model or into our semantic models, we set in our rules file, like, these are the events that you should use for this. These are the tables, and the agent was able to do pretty effective analysis. One thing I will say about rules files is they can overload the context of your agent pretty fast. There's definitely people posting in in other context, not just about data, about how overloading a rules file or overloading the context of an agent can reduce performance because, like people, if they hear a lot of information, they can kinda lose sight of what's important in a in a huge tirade. So you do wanna make a point of moving things when they are complicated enough or formalized enough into a data model or semantic model, rather than keeping them in your rules file forever. Cool. And I'll just quickly close out with this mirror image of this is what our data support channel is like now, after we rolled out the Hex Slackbot. And it's basically just people asking all kinds of questions, and getting answers right away. And the things that I think are really cool about this selection of questions in particular, like, one, they all came out relatively close to each other and these people got answers faster. This is a a wide variety of different functional roles asking these questions. We've got people, on our customer team. We've got salespeople. We've got engineers. And a lot of those are roles that data teams are often not staffed to directly support, and they are now big consumers of data, some of the biggest consumers of data. So one thing that has been really exciting for me to see about having the Slackbot and having the agent in general is that we are serving more types of people within Hex, as a data team by curating context and giving it to our agent to scale what we're doing. And we're also answering more questions and answering questions more deeply for people faster, and carving out a lot of time for ourselves to do other things. So the Databot has changed a lot of stuff for us as a data team, and I, honestly, as a head of data, don't know how I'll survive without it. And just to really close it in, the benefits of having a Slack bot like this is knowledge is shared automatically by default. Everything is in public. You've got broader coverage of the types of people that you work for and the types of questions you're able to answer, and it's a lot easier to experiment with context, and and giving people information that you otherwise had to document, outside of your main data workspace tool. Alright. And with that, Nicole, I don't know if we have more time for q and a or if people wanna stick around, but we are done with the presentation. Amazing. Yeah. Thanks, Katie. So I will say, clearly, we should have maybe scheduled it for a full hour. There are so many amazing questions. So I clearly did not anticipate just how many questions we would get. So I think we can stick around for a bit, because I think Katie and Sarah, hopefully, we can both of us all can stay for, like, a little bit of extra time. And for people who have to jump, again, this is being recorded. But I I think what I'm noticing is there are a couple buckets of questions we can address. So for for folks who can stay, I think we can touch on pricing, observability, a little bit about, like, data access. I think there's some sensitivity around, like, who is seeing some of these answers, in public Slack channels. So I think maybe we can touch on those three topics real quick, do a do a lightning round. And then, again, this will be this is recorded. So, for folks who wanna watch this later, they can. Okay. So let's get into it for those who are able to stick around. Maybe let's start with, pricing. So there's I think people are, you know, curious because, of course, a lot of these AI agents are relatively new to Hex, all things considered. And so we, you know, don't have, like, specific pricing, publicly displayed on our pricing page. And so could you speak a little bit to how how we're kind of thinking about that and approaching it and and for folks who are wondering if this is going to kind of affect, their their cost of using Hex as as they try these agents out? Yeah. Great great questions. So I think the the first thing to keep in mind is just the seat access like like we touched on earlier. So, the the users that will be able to use these features, are, again, Explorers, and higher. So if you have editors on your plan already, they will be able to use all these features right now. If you have Explore receipts on your plan, you can extend access to the explorer seats as well. If you don't have explorer seats and you're interested, you can reach out, to our sales team, and we can get started with a conversation about how to, like, get that seat enabled for your workspace. But, generally, right now, we have, a a credit system associated with each plan as well when it comes to using, agents and, just their their use in the workspace. So users on our team's plan have extended credit, capabilities. And for the foreseeable future, we are not, charging for, extending over those, credit limits. Like a lot of other teams, we are thinking a lot about how we're, pricing this, and, how we want to, continue to to, allow allow users to access these features, with with our pricing model. But, basically, just to kind of maybe assuage people's concerns about pricing, it is like our top priority at Hex to have people using agent workflows, like, as their main method of using Hex. So we are not, we want to make this extremely accessible to people, and this will be, like, built into the overall framing and, like, pricing of, of Hex in general. So, yeah, that's the the short answer. And if you want to talk further about any of these, I would just recommend reaching out, to us. Awesome. Yeah. And just to emphasize again, if you're interested in that explorer role, please do reach out to us. If there's already someone on our sales team that you are already in conversation with, they can help to enable that role in your workspace. Or if you don't already have someone that you feel like you get in contact at Hex, I can drop something in the chat so that you can kind of reach out to us, and we can make sure to follow-up on that. And then, Katie, I think you were maybe going to jump in and say something before I cut you off. Yeah. One thing I was gonna call out, and I think I saw a question about this in either the q and a tab or chat. But the agent does run queries in the background. And if you have more people using it, like, whatever your database or data warehouse vendor is may see some usage, increases due to that. It has not been anything crazy for us in particular. Like, I I don't find running interactive queries is a big driver of costs on technologies like that, but it is something to be mindful of as well. Yeah. Thanks so much for calling that out. Really, really good point. Okay. So the other kind of theme I noticed is that there are just a bunch of questions about observability. So just wondering if you guys could share a little bit more, about, like, the context studio and where is this getting logged, how can people even kinda see the questions that, business users and stakeholders of the company are asking. So either, like, Sarah or Katie, if you wanna kinda jump into how you think about observability and what's on the road map for us. Yeah. I can maybe jump in with just, like, how we're thinking about from the product side, and then I'm I'm sure everyone would love to hear Katie's, take on this as well. But, yeah, Not surprised that we're getting questions on this. It's, of course, makes tons of sense that data teams need to be able to keep tabs on the types of questions that users are asking. And, again, they need to be super in the loop with, checking checking the responses of these questions and making sure that the context that they're curating in the workspace is working the way they need it to. So Katie knows this. Nicole knows this. But this is, like, our top priority, basically, for the company is, like, how are we making these workflows really excellent? How are we making it easy for the data team to understand, like, how context that they're adding is, working in the way that they'd expect? How are we making it easier for the data team to, keep tabs on the types of questions that users are asking and where context is, like, insufficient? So in the very near near term, we are going to be coming out with, what we're calling our observability tooling for admins and workspace managers, where they'll be able to see the questions that users are asking and the threads that the, agent is creating. So that will be just basic monitoring on, the types of questions and, the topics, that the that the questions are centered around. I'll go let Katie speak to this, but I think also the value that we see with the Slack bot is that all it tends to encourage, like, building with the garage door open. Like, everyone's kind of has great observability and the types of questions that people are asking, and it encourages, a lot of collaboration. But, yeah, we'd love to hear Katie's take as well. Yeah. I'll speak to a little bit just the context studio as we have rolled it out right now as well, which I will say upfront. Like, I usually end up looking at, like, two to three times a week and and just kinda scrolling through all of the threads that have happened. It's something that, like, I'm not looking at every single thread, but getting the span of types of things that people are asking for is super useful. So you can look at, all the questions that are being asked of your agent. That's not just your Slack bot. That's when people enter into the agent flow directly in the Hex homepage. You can see what people are asking, in notebooks. You can see what people are asking in MCP, if you have anyone doing that. So you can see a lot of different stuff. You can see who your top users are by volume. You can see what kind of roles and seats they're in. So there's actually a a pretty decent amount of information. You can even see, like, whether people have thumbs up or thumbs down, the threads. But right now right now, it's a little coarse, but the data team at Hex has been spending a lot of time with the product and engineering team, dogfooding and and figuring out what is actually helpful for monitoring these things at scale. And I feel pretty excited about the stuff that we're building. And, like, I love getting to see what people are actually doing with my tools as a head of data. It's it's something that's hard to find in some tools. So having as much as I do already, has been super exciting. It makes it way more possible to be a very empirically and numbers driven data team. Awesome. And then I think this will be the last question that I posed for the group. And then what we'll do is for all the questions we didn't end up addressing, if this if your question did not make it into the last question, we will just be reaching out individually so you can definitely expect an email. So we will make sure that your question does get addressed, async. So for the last question, maybe, Sarah Tayeri, could you speak a little bit to how the kind of permissions and authorization of, like, access to underlying data works? Like, let's say, I have access to certain projects that maybe, like, you don't have access to or vice versa, and I'm asking a question, to the Hex agent in Slack. I guess, how how does that kind of data access work, and how is it kind of, figuring out what kind of, answers it's able to to give each of us? Yeah. It's a great question. So we built, our Slackbot with feeling in mind that if you are the type of data person that is comfortable adding a public Slackbot question response experience in hex, then you're probably, okay with having people view the outputs of those responses. Some would say that's the value of having the feature in the first place. So when you enable the Slackbot, we are always going to be using the default data connection, that you specified for the threads agent, in your settings. So when you hook that up, anyone that is in the Slack channel is going to be able to see the outputs of the responses that the threads agent generates. So that is something to keep in mind as you're as you're rolling out. Typically, like I mentioned, I I I think people that are interested in this feature see that the the value of the feature comes down to sharing their responses and having everything be in the open, but that is something to keep in mind. Also, something I didn't mention on the demo is that, when you kick off a question from a a public channel in Slack, we, share that thread by default with the workspace. So anyone will be able to open that thread in Hex and audit, the response or just view the response in more depth. Again, you can already see the outputs, like, in Slack, so we just didn't wanna create friction with, like, requiring someone to share that thread in order for other people to view it. And I will say one final thing, which is, again, I did not demo this. There's a lot to the feature, but you can also kick off, questions in, like, your your DMs or something like that in Slack, and that can just be, like, another entry point. When you do that, the thread is not shared by default. So we'll only share it, when you're executing that, command from a public Slack channel. Awesome. But in general, so, yeah, it's always just distilling down to, like, your user permissions and what you have access to in the workspace. Yeah. And, like, I've seen people add it to private channels, as well within our Hex Slack. Yeah. And just to reiterate for the private channels, the threads created there are also going to be accessible to the entire workspace in the same way that they would be with public channels. So it's only with DMs that it would be something that is private, to you. And then I saw Asia came in with another question. So just to clarify, yes. You would just be selecting one specific data connection. That's the same for both threads inside of the Hex UI as well as what the Hex agent is able to use from Slack. Same same setting. That's correct. I will also just, clarify that if you kick off the thread from, within Hex, you can change the data connection manually. In Slack, we just don't have that UI ported over, so we're always going to use the default. So that's just one nuance to keep in mind. Amazing. Okay. Great. Again, thanks for all the questions. We love it, and we'll definitely make sure to, get back to you all async with answers. We will also be hosting another webinar on MCP specifically, and I think so many of these questions, right, are they overlap between our Slack integration and our MCP server. So we can also get to a lot of these questions live, in the MCP, virtual event that we'll be running. So keep an eye out for that and hope that you will all join. So, just to lastly quickly wrap this up, I'm gonna throw some resources in the chat right now. So if you toggle over to the chat tab, I'm posting both our, blog post announcement for for these integrations as well as the technical docs for our Slack agent. And then I'm also going to drop our LinkedIns for folks who want to, connect with us. So thanks again so much for joining. Apologies for running a bit overtime, and bearing with us with this new platform. But we so appreciate the engagement and excited to see you guys next time. That's it. Alright.