Name: Build Hour: Workspace agents in ChatGPT
Uploaded: 2026-04-29T18:32:12.936Z
Duration: 37 min 56 s
Description: Build Hour: Workspace agents in ChatGPT

Transcript for "Build Hour: Workspace agents in ChatGPT": Hello, and welcome back to Build Hours. I'm Victoria from product marketing, and I'm here with Christina and Hojun. Hi. I'm Christina, and I work on the engineering team for Workspace agents. Hi. I'm Hojun. I'm on the solutions engineering team. Awesome. And today, we're gonna show you Workspace agents in ChachaPT. So if this is your first build hour, build hours is all about helping teams get more practical value from OpenAid products with real examples and tips you can use after the session. So today is a special session focused on building Workspace agents in ChatChuPT. And Workspace agents are available in research preview today to ChatChuPT Business, Enterprise EDU, and ChatChuPT for Teachers plans. So here's what we're gonna cover. First, Christina is gonna start off with a quick intro to Workspace agents. Then Hojun is gonna walk through two demos. The first is a meeting prep agent that checks your calendar, does research, and creates a meeting brief. The second is a software review agent that helps handle employee requests in Slack and roots next steps based on your policies. After that, Christina will come back to share a few tips on getting started, how to think about agents and GPTs, and go into some detail in enterprise admin controls. And then we'll wrap up by answering your questions live. So on the right side of the screen, there's a Q and A chat box where you can submit questions at any time during this session. So now I'll hand it over to Christina. Great. So before we get into the demos, I wanted to take a minute to to level set on what Workspace agents actually are. So at a high level, Workspace agents are codex powered agents in ChatChBT. They're built to handle complex, long running work that spans multiple systems, and they have access to files, code, and tools. And so what makes this really exciting is that they're not just helping one person with one prompt, but they can actually gather the right context, keep going in the background, and be shared either in ChatGPT or Slack so a whole team can use them together. And they also have memory, so they can be guided in conversations and actually improve over time. So getting started is quite simple. You click agents in the ChatGPT sidebar, describe a workflow that your team already does, and ChatGPT helps turn that into an agent. So today we're focused on Workspace agents. But since I mentioned codecs, I wanted to also quickly explain how Workspace agents fit in with our other agentic products. The simplest way to think about it is Workspace agents are for teams. They're built for shared work for tasks that run-in the Cloud even when your computer is closed. Codex is great for individuals working with a personal agent to get work done, and the agents SDK is for teams that wanna build custom agents directly into their own products and customer experiences. At OpenAI, many teams are already using Workspace agents across a few different functions. As a couple of examples, our marketing team has built an agent that turns a product brief directly into a website. This pulls requirements from Google Docs and code. Our accounting team built an agent that helps prepare for month end close faster and more consistently. And our finance team has built an agent for vendor risk review, which researches vendors, assesses signals like sanctions exposure, financial health, and reputational risk, and produces a structured report. So today, Hojun is gonna walk through building two of these examples, the meeting prep agent and the software reviewer. So I'll hand it over to him. Yeah. Thanks, Christina. So, we'll be walking through how to build these agents, and we're gonna start off with this meeting prep agent. Now I'm sure folks who are customer facing on this call, you have just a lot of, customer meetings to get to every single day. So this is an agent that I built to help with the manual multi system work of customer meeting prep. So as you can see on the right side here, every single morning, I get a email from my agent, named Otto, which checks my calendar, does research on the customer, whether it's on the web or within my Google Drive, and it'll actually create a meeting brief for each meeting that it will link and provide to me. So, you know, the hours I would spend every evening building out my meeting prep docs, docs that I wanna share with my team members so, we're aligned on the agenda, that's all done for me behind the scenes. And the great part of this is that this agent, can be shared, so that others can customize it for their specific workflow, and get the similar value that I'm seeing. So let's jump in. What we'll do is actually we'll go through this cookbook, which, actually walks you through how to create the agent that I just described. So you can follow along with what I'm doing on screen, or you can also, you know, reference this and know that you can go back to it, later on after this session. What I'll also introduce you to are some of the demo assets that I'll be using here today. Don't worry. These are just mock events, with some fake customers, but I just wanted to show you a real live desk run, towards the end of this demo section. So here, we're going to go through, my calendar, and then we'll also have, some, you know, sample assets, to show you how it all comes together. So some account notes, other company details, and then a sales meeting prep to bring it all together. So let's dive into ChatCBT and kick off this agent build process. Workspace agents, is going to allow you to build agents, just using natural language. So you can start from a blank slate, or you can utilize some of our friendly templates down below, to get started with your agent. But I wanna show everyone how easy it is to build these agents from scratch. So let's get started with a prompt, that just outlines what we discussed, in terms of the use case and will get me started, building and working with ChatGPT. So here, my prompt just calls out that it's gonna help me with sales meeting prep. It calls out the various tools and apps that I might need. In this case, it also says, here, I have a template for you to use. And then last of all, you know, just directing it to send an email with a summary of my upcoming meetings and links to the full briefs. So if that if I enter in that prompt, what you'll see is that ChatGPT is gonna guide you throughout the build process. It's going to create an outline for the agent plan that I can review before it gets started building, but I really can just work with ChatGPT here. Just use natural language, give it feedback, iterate together, and we can build a really powerful complex you know, the agent plan, everything I need, the capabilities look good, so let's get started with the build process. So what you'll notice in a couple of seconds here, is that the agent is going to or sorry. Chat TpT is going to continue to live in the left side pane here, while it works on the right side to, wire up all of the, you know, scheduling, the apps, the tools, and also create a fine tune instruction set. So this is a really great no code way to build agents, that still, as Christina mentioned, work like our codex, agent. So it can run across long running tasks, behind the scenes, access data, and take action across systems. So pretty surely here, we're gonna see, ChatGPT finish up wiring up the agent, and we'll be able to actually do a test run and show you all of this live. Alright. So we'll give it, a little more time here, and I'll start to poke around as it, finishes configuring some of the different applications and tools and steps. Alright. So you can see here, it's finished up the instructions. It's going to, you know, continue to build out the agent. At this point, I can go in here and make modifications directly, or again, I can provide that feedback to chat on the left side, and it'll take care of all of that configuration. Alright. So almost done there. Let me actually wire in some tools here. So for Google Calendar, and this is helpful. I wanted to do this manually to show you the configuration step. So any of the applications that you're utilizing, you have the ability to, you know, lock down what the agent actually has access to. So in this case, we're looking at Google Calendar, and I just need the agent to reference my calendar. I don't need it to take any right actions or anything like that. So I'm actually going to disable, you know, those actions, for the agent to take those actions, And I can know confidently that anytime it's actually seen my calendar, it's not going to, you know, adjust or change any events. It's just going to go about the workflow that I've described. I'm also going to quickly add in Google Drive, as well as Gmail. Actually, it seems like okay. There we go. So I'm going to add in Gmail here. You know, similar configuration. In this case, I actually want it to run an email to me, but maybe, you know, I can disable some of these actions that it actually does not need to do as a part of this workflow. Workspace agents is all about having the control over what these agents are able to do, providing them a playbook, and letting them execute on these tasks on, you know, a schedule or being able to interface with them in tools that your teams use today. Let's do another configuration step here, and I'm actually going to add in a skill. So we have the apps and tools that the agent can use. Now let's actually give it some additional direction on how to best support, my meeting pref creation. So the great thing about skills within ChatGPT and Workspace agents is that, first and foremost, you can bring in skills from other tools and platforms that you use today. So the best practices and processes that you've codified, you can bring them in and have your agent use the same, so that it can scale the expertise, and consistency across your organization. You can also choose popular skills within your organization that already exist, or one of my favorite, kind of tools here is actually being able to create a new skill and have ChatGPT wired it up much like it did for all of the other agent configuration. So let's have it generate a skill here and add it into, this agent's playbook. It'll kinda keep going through this. So we'll give it a few seconds here, but I also have, a kind of refined agent that we can skip ahead to, and be able to jump into a preview run. But I always like to observe what ChatGPT and the agent is doing as a part of the build process. So we'll take a look at what it's doing for a little bit longer here. Alright. So it's finding that template within my skill or sorry. Finding that template within my Google Docs, and adding it in with some formatting instructions so that, you know and I'll show you the final product, coming up here, but it's going to beautifully format these meeting briefs for me with tables, headers, bullet points so that I can get the information I need at a glance, especially if I'm looking at it on my phone on the way in my morning commute. Alright. So this looks, like it's working out well. I'm actually going to skip ahead to an agent version, that's maybe refined a bit more. You know, I've had some time to work with it, and I wanna show you what it could look like to easily run a preview test as well. So here, I'm going to start up a new session. Again, the same agent, maybe with just a little more refinement. I've given, ChatGPT some feedback, and we've done some preview runs. So I feel pretty good about this agent, and I'm at that step where I can schedule and distribute it. But let's let's actually do a quick preview test run. So I'm going to ask Chet to do that. Can you run a preview with my calendar for tomorrow? So much like that experience of, or the no code experience of building these agents, you can also work with ChatGPT, to actually, you know, run tests for you. Like I said, you can give feedback once, these runs are executed, and you can essentially, you know, continue to work with it in this way. So you don't need IT and engineering support to build out these workflows. We really wanna make sure that these subject matter experts, are the ones that can build up these flows and, again, distribute it to teams and share them, as needed. So, cool. Right on cue. We're gonna get this preview test, kicked off. And the other great part of these preview runs is that you can see exactly what the agent is doing. You can see, you know, how it's pulling information from other systems, what its chain of thought is, and be able to follow along. Now this isn't gonna be a view that you babysit or that an end user will necessarily see every time, but it's great for those preview runs and also show you how can how you can look at previous, historic runs of your agent. But this is, of course, especially great as you first build out this agent and, you do some tests before you put it into a production setting. Alright. So as you can see, we're following along here, much like the agent outline and, the other ways that, Chat TpT, provides workspace agents functionality, we're able to see the high level, kind of plan for this agent as it executes this workflow. And we can see exactly for each step, you know, what it's doing, what files it's accessing, you know, how it's going about utilizing the skills. And then we're going to get a final result pretty shortly here, that we can review. And, you know, I'll actually then show you how we can schedule this agent, share it, and distribute it to your team for the for them to remix and utilize for their own workflows as well. Alright. Well, we'll let this run a little longer, but, in the demo industry, as I like to call it, we have this concept of cooking demos. So much like on TV when someone's, putting together a meal and they pull something out of the oven right away, let's actually skip ahead because, wanna be able to get to your questions and the second, demo agent as well. So if I go into my inbox, I have a previous run, that I used with this agent. So you can see here, you know, really great formatted email with all the briefs that I need. And this is based off of my calendar and all of the customer conversations that I have for tomorrow. So it's a busy day, and I wanna be able to go in and prep for Blossom Mart, PedalPay, as well as Nectarworks, you know, just, at with a lot of simplicity. So what the agent did is it looked across my Google Drive for all of my customer contacts, you know, maybe did some research on the web to pull in additional, information. And as you can see here, it used that skill really well to give me a highly formatted and styled document with, you know, the executive readout, customer snapshot, meeting objective. These are all things within my template that I typically look for when I'm putting together these meeting briefs. And the great part of this is I can also share this with my team members, so they know exactly what the agenda will be, and what our shared goals and outcomes, for the meeting will be with the customer. So, as you can imagine, this is saving me, you know, hours on a daily basis, prepping for all of my customer meetings. I can look at this on the way to work or in my free time. I don't have to do all of the manual work, because auto, my agent is going to do that for me on a scheduled basis. So let's go back to the agent, and I'll show the last sort of step here, of the agent and its execution. So what you can do after, you know, running through tests is start to, share and distribute this agent. So, you can obviously add a schedule for yourself, and we would want this to run daily for my workflow. But let's say this starts to work really well for you, and everyone else wants to have the same, type of, you know, efficiency and hour saving that you're seeing with this agent. Well, you're able to then enable, sharing for this agent. So what it'll mean is that anyone in your workspace will be able to utilize this agent or duplicate and remix it. Let's say, you know, someone uses SharePoint instead of Google Drive, they can make that change to this agent and use ChatGPT to reconfigure it. Or maybe there are additional steps you wanna include like having ChatGPT also create a slide deck for you after, putting together briefs. You can, hopefully tell that there's a lot of customization and flexibility that Workspace agents provides to you. So hopefully you're inspired, to build out an agent that's similar. Again, you can refer back to this cookbook, to be able to go through, the agent, after this session as well. Alright. So that was a fun one. What we did for the meeting prep agent is we built an agent using natural language. We went into the configuration of the app parameters and access. We enabled skills, and also memories. I'll include that in the second agent. And then we ran an agent preview, so we can check out the agent behavior. And once that looked good, we decided to share it, share the agent for team use. So, really helpful agent, in my day to day. Hopefully, folks, build out your own version, but let's go into the second example for today. So this one is going to be, right off the bat, more focused on serving the needs of your organization or team members. So this is a software review agent, and much like the previous agent, this is actually an example that, is modeled off a real agent that we have, living in one of our Slack channels. I'm sure everyone, at your company, you have a channel like this where high volume, high sense or time sensitive requests are made in Slack to get a software tool by a employee, you know, when they need it. And typically, those channels, you know, I sometimes feel like the experience is more like a chatbot. I just get links. I have to do my own, you know, discovery, versus having an agent take that action for you, where it will research, the capabilities of your requested vendor. It'll compare options against the approved stack, you know, reason across, details like utilization of existing tools, and then it'll provide guidance to the user or even take action. So, you know, in the cases that a human, the loop is required, you can even escalate into Jira. So this has been an amazing agent that internally has saved a lot of time for our IT team, obviously, but also has helped reduce our duplicate tool spend and sprawl. So really excited to go into, this example. So let me move into this experience here. So, I'm gonna start off from scratch here once again just to show you that even if we're building an agent that, maybe takes on a little more work, maybe is a little more complex than a personal kind of meeting prep agent, that you can still start from the starting point. So here, we have, you know, a prompt. And in this case, I actually have a skill that I want to bring in. So I'm going to actually, bring in a skill that our products, or sorry, excuse me, procurement team built, to codify their workflow and best practices when it comes to these, sort of decisions and evaluation. But much like the previous agent, I'm starting off with the systems and, tools that the agent will utilize, and then I'm giving the skill, and we'll have ChatGPT wire it all together. I'll skip ahead a bit. We won't go through the full build process because we took a look at the previous one. But again, I hope this inspires you to start from scratch, and maybe bring a process or workflow that you're an expert on, and be able to build that, agentic team member that can either assist or take over that work for you. So Awesome. Much like the previous agent, we have the agent plan, you know, things look good, so we'll get started here. And maybe while this, wires together, I'll take the opportunity to showcase, the mock dataset that this agent is using. You know, in our production agent, it actually leverages a software kind of management platform and a custom MCP. Here I brought that into Google Sheets just for simplicity of the demo. But I pull this up to showcase that, you know, the agent's actually going to look across all of the approved software, you know, the utilization of, the tool as it is today, and the functionality and the capabilities of each tool that we have so that when a user is requesting a tool, for a very specific purpose, the agent is able to reason and know that if it's making a recommendation from our approved stack, it's actually going to meet the needs of that user. So that's what we have provided to the agent. And what we also have, and I'll go into, is a Slack channel that I'll show how we can, you know, bring that agent into Slack, the interface where these requests are being made. So if I go back here into Slate, you know, obviously, there's a lot of configuration going on here. So maybe I'll skip ahead to, previously built version, but I encourage everyone to try this out, especially when you go through it for the first time. It's one of my favorite kind of parts of this product, watching ChatGPT put everything together and allowing you to follow along as well. Alright. Right on queue, we have, the instructions. But, yeah, I'm gonna skip ahead to that sort of refined agent, and walk you through a couple of setup steps before we go into Slack and show a lab test run. So, first and foremost, what we have here is, the ability to actually add Slack. So I'll showcase that by adding actually, I'll I'll pull this up and you'll be able to see, that you can specify what channel that an agent should operate in, as well as, you know, whether it should respond when mentioned or for any, you know, relevant messages, that it can handle in the channel. You can also add additional instructions, so how to handle requests, how to respond in thread, things like that. So there's a lot of customization that you could do, when it comes to integrating this agent into Slack, as well. So here, we have the agent. One thing I forgot to call on the previous agent, so I'll bring up here is this concept of memory. So memory allows your agent to, save notes, you know, context, outputs, other things that will make it, better at its workflow over time. So you can think of this as a persistent set of, context that the agent will keep, and be able to execute on in its workflow much like it does with skills and tools as well. Alright. Well, everything looks good. So let me show you one more thing in the platform. We'll actually go into Slack. So on each agent's screen, you're also going to be able to see all of its activity. So that preview run I showed you with the previous agent where you're able to see everything that the agent did, this is where for each agent that you create, you also have this centralized place, to view all of the previous agent runs, and you can go in here and check the agent trace down to that level of detail. So, this is especially handy, of course, when you have an agent like this that is autonomous and runs on its own, so that you can go back to any requests, and everything is centrally audited and, and captured for you. And this, of course, is also exportable via our API. So just wanted to show that before we dive into the live example. So here I have my Slack channel and some previous test runs. I'm just gonna show you what it looks like to kick off one of these, cases, and then I'll show you and jump ahead to the agent response, because it does take a little more time, than, of course, a chatbot. The agent, like I outlined, is actually going through, and you can see here the dialogue popped up that it's working. It's received my message. But the agent is going to go through and, you know, do some research on Screen Studio. It's going to, reason against, you know, the policy, the skill, as well as the approved software vendor list, before it gets back to the user with a recommendation or action that it took. So, you know, while it's working, let's actually go back to this one here, and I'll show you all, you know, for the same exact request, what the agent was able to help me out with. So Slate responds and actually says that its decision is that this request needs IT review, versus me being able to self serve Screen Studio. It clearly lists out the reasoning why, the rationale. You know, we actually have a proof tool called Bloom, that would work well for my needs of needing a high quality, demo recording tool. But it is blocked right now because it's overutilized. There are actually no available licenses. So that's why, you know, the agent in this case has actually chosen to escalate to IT as IT will need to either provision additional licenses or maybe they'll make an exception for me, and to use Screen Studio in this case. But you can see that the agent is responding with the sources checked with all of the reasoning and, guidance for the user. So I don't have to have this sort of experience where I now have to look up the tool, maybe provide my rationale, the agent's done that work for me. And as the final step here, you know, instead of having requiring me to open up my own ticket for IT review or requiring IT to do that, it's also created this nicely in Jira as a task, for them to review and quickly unblock me from, getting access to a tool that I need. So, you know, this has been a really powerful agent like I noted. It's handling all of these requests for our team today, and saving our IT and procurement team a lot of time, but also, you know, making sure that users, when they have these time sensitive requests because they're typically made last minute, they have the tools that they need. Alright. So with that, we went into the IT agent, how to build an agent with existing skills, how to set up Slack so that the agent can serve teams within other interfaces, how to interact with the agent in Slack, you know, viewing the agent run history and traces, and then last of all, how you can even handle escalations in Jira. So much like the previous agent, hopefully this inspires folks to build out agents that help your team members. But with that, I will pass it back over to Christina to talk through getting started. Great. Okay. So now that we've walked through some demos, I'll cover a few few tips on how to how to build with Workspace agents. So there's really four ways to build an agent. First, you can build one in conversation as as we saw in the demo. So just describe a workflow your team already does often and ChatGPT will help you with the setup. Second, you can start with one of our templates. And so this is a great option if you want a faster starting point with built in skills and tools and then go on and customize from there. Third, you can bring in an existing workflow by importing skills and apps from other platforms so that you don't need to start from scratch. And finally, if your team is already using custom GPTs, you can start to test those workflows in Workspace agents, and we'll have an automatic converter from GPTs to Workspace agents, coming soon. So I also wanna take a quick second to talk about, GPTs and what, this means for our GPT users. So custom GPTs were our first step towards lightweight process automation and many teams have already found them helpful for creating shared templates to be used in Chat GPT. But when we first launched GPTs, we didn't yet have the right models or platform primitives to make them truly powerful and extensible. And so workspace agents are the next stage of GPTs. Shared agents that can run multi step workflows across tools, on schedules, with approvals and follow through. And so if your team already has GPTs, I would actually start by testing some of those workflows out with agents. And finally, let's talk about permissions and admin controls. So as the builder, you always stay in control like Hojun demoed, you decide what tools and data your agent can use and when approvals may be required for more sensitive tasks. And for enterprise and EDU plans, admins can also use role based access controls to define who can use agents and which apps are available. And finally, the compliance API then gives admins the ability to monitor and manage usage over time. So these agents are very powerful, but they're also built to operate within the controls and governance that your organization needs. Amazing. Thank you so much, Christine and Hojun. Now we have some time to answer some of the questions you've been submitting. So the first one that came in, quite a few Microsoft suite users out there. So maybe, Hojun, can you help explain, do agents work with Microsoft tools and a bit about, like Yeah. Absolutely. Yeah. Yeah. So, I did use the Google Drive, mostly because it's easier to set up for my demos, and I'm familiar with them. But you can use Microsoft tools like SharePoint, Outlook, and Teams. And as you saw, you know, what, you can read or write or the agent can read or write will depend on the specific app and your Microsoft permissions, as well as your workspace admin settings, but you'd be able to replicate what I built, with those suite of products as well. The last kind of distinction I'll make, when it comes to the Slack example, agents can run-in Chat GPT and Slack today. Not quite yet for Teams or these tools, but we are actively investing in those services and adding more soon. Awesome. Thank you. Alright. The next question is about memory. Maybe, Christina, do you wanna talk about how agent memory is different? Yeah. So Hochin touched a bit on this in the demo. But whenever you build a workspace agent, you can enable memory for that agent as well. And so this is a persistent file system where the agent can store files and notes to be reused across, future runs, and you can prompt for it to add it explicitly. It can also choose to add it automatically. And these, this file system is specific to every channel that the agent is in. So if you are using this agent in ChatGPT or you're sharing it with someone else, in ChatGPT, every user has their own file system for their own, instance of the agent. Every Slack channel also has its own memory. So if you're in kind of a shared Slack channel, all of the memories within that Slack channel will be shared across all the different messages that are that are sent. K. Cool. Yeah. Nice. And then improvement over time is an interesting bit. Cool. The other question there have been some questions about sharing. So do you get access automatically to all the agents in your workspace? How do you share them across your company? Yeah. I can take this one. So, agents can kept be kept private. They can be, specific to your workflows like I showed with the first agent, but they can also be shared with anyone in your organization with a link, or they could also be listed within your company's workspace agent directory. So in that case, you know, these are agents that can only be shared with members of your ChatGPT workspace. But within those, you're able to distribute your agents. You're able to allow others to duplicate and remix those agents like I was talking about with the sales meeting prep. And ultimately, the admins are able to control who's able to build, publish, and share those agents as well. And so for enterprise and EDU, that's within our admin settings, and our role based access controls. For now, we only, allow the owner of the agent to actually make edits in that agent. But very soon, we'll allow multiple people to edit an agent as well. Nice. Cool. Awesome. And can you create agents with different roles, like, any kind of best practices for more specialized agents? Yeah. I would say we just showed you a couple. Christina alluded to the fact that, here at OpenAI, we have agents, across every single business unit. And so you can have agents with different roles. In fact, they work best when you have a specialized role in mind with them or for them, with all of the apps, tools, files, skills, and instructions that they need to excel in that role. So, you can think of them as, you know, really capable team members that you build and you provide everything they need to be successful along with that guidance. So given that they also have memory like we just discussed, they can be guided or corrected over time. So they can really be improved. And so like I said, you know, get started, build these agents because it's really easy to refine and provide feedback, and you can continue to use ChatGPT, to, you know, change and improve the, agent behavior as well. Awesome. Thank you. And Christina, do you have to use the desktop app? No. So, I mean, as we as we demoed, you can, create, run, share these all directly from the web. You can use them in Chat. You can use them in Slack. And they run-in the cloud, so they keep working even when you're away. You can also use them from the desktop app. You can use them from mobile as well. So really kind of across our different surfaces. Yeah. Yeah. I'll just add to that that both the agents that we demonstrated today, aren't actually running on the app or locally. They're both in the cloud. One's running on the schedule, and giving me notes for my commute. And then the second one, obviously, is living in Slack. Awesome. Cool. And, Hojun, could you answer questions about pricing? Yeah. Absolutely. So, pricing, right now, workspace agents are in research preview and they're, free, until May 6. So, you know, I sound like a broken record, but, when I say to when I'm encouraging all to try it, you know, really put it through its paces and try it out during that period. After that usage, we'll move to credit based pricing. So the number of credits that an agent run consumes, will obviously depend on the complexity of the agent and the test that it performs, but you can think of it similar to how we, or more complex codex tests, like, you that use more credits, because, again, these are agents that have the same harness and the ability to perform, long running complex tests. We'll be sharing more specific guidance, through admin and billing channels regarding pricing as we get closer to, that May 6 date. Awesome. And then we actually had a couple more questions come in. So since you mentioned codecs, Christina, maybe you could talk a little bit about what makes Workspace agents different than working with codecs. Yeah. Definitely. So I think, first of all, they're they're meant for teams, and so they can be shared with, people in different functions at your company. They also run-in the cloud, and so they run even when your when your computer is closed. And then yeah. Again, they can be kinda deployed in these different environments as well. So used directly from Slack with, any other triggers coming soon. Awesome. And then quite a few questions around skills. Can you help folks understand, you know, what's the difference between skills and agents? Can agents run multiple skills? And how do you kind of determine, when you should add a skill versus defining the agent instructions? Yeah. Absolutely. So I would think about skills as being sort of, you know, playbook, best practices, and processes that you have that you could cut a far or maybe you're already using in different, platforms and tools. And the agent is the worker that actually leverages that skill and other, you know, instructions that you have, to do the real work. So, an agent can have multiple skills, and be able to utilize them for the specific, you know, tasks that that it's working on. So you can think of it in that way. Once again, skills are gonna be, like, your best practices. Sometimes there are scripts. There are other, kind of direction and guidance, and then the agent will actually take on the skill. Or use also the apps and tools and execute on that work, as a whole. Awesome. Yeah. Thank you so much for taking these questions. Of course. Yeah. I think we're gonna wrap up, and thank you everyone so much for joining today. So to wrap up here, just a few helpful resources that we will share after this session. So like Ko Jin mentioned, the cookbook for building the sales meeting prep as well as some other, resources with additional demos. And then also wanna do a quick plug on upcoming build hours. We have two upcoming sessions that are focused on our API. So you can just follow the link here to watch, to sign up for those sessions and also watch past build hours. Thank you so much for attending, and thank you so much to Christina Hojun again. Yeah. Thanks all. Thank you. Thank you.