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 going to show you workspace agents in chatt. So if this is your first build hour, um build hours is all about helping teams get more practical value from open a products with real examples and tips you can use after the session. So today is a special session focused on building workspace agents in chat and workspace agents are available in research preview today to chat business enterprise edu and chatupt for teachers plans. So here's what we're going to cover first. Christina is going to start off with a quick intro to workspace agents. Then Hojun is going to 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 GBTs, and go into some detail on 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&A chat box where you can submit questions at any time during the 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 codeex powered agents in chatbt. They're built to handle complex longrunning 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 chatbt 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 chat GPT sidebar, describe a workflow that your team already does, and chat GBT 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 want to 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. Um, 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 going to walk through building two of these examples. Um the meeting prep agent and the software reviewer. So I'll hand it over to him. >> Yeah, thanks Christina. So uh we'll be walking through how to build these agents. Um and we're going to start off with this meeting prep agent. Um now I'm sure folks who are customerf facing on this call um you have just a lot of uh customer meetings to get to every single day. Uh so this is an agent that I built to help with the manual multi-system work of customer meeting prep. Uh so as you can see on the right side here every single morning I get a email from my agent uh named auto which checks my calendar. uh it 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 uh docs that I want to share with my team members so uh we're aligned on the agenda that's all done for me behind the scenes. Um and the great part of this is that this agent uh can be shared um so that others can customize it for their specific workflow. um and get the similar value that I'm seeing. So, let's jump in. Uh what we'll do is actually we'll go through this cookbook uh which um actually walks you through how to create the agent that I just described. Uh 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 um 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. Uh don't worry, these are just mock events. uh with some fake customers. But I just wanted to show you a real live desk run uh towards the end of this demo section. Uh so here we're going to go through um my calendar and then we'll also have uh some you know sample assets uh to show you how it all comes together. Uh so some account notes um other company details um and then a sales meeting prep to bring it all together. So let's dive into chatbt and kick off this agent build process. Workspace agents um is going to allow you to build agents uh just using natural language. So you can start from a blank slate or you can utilize some of our friendly templates down below uh to get started with your agent. Um but I want to show everyone how easy it is to build these agents from scratch. So let's get started with a prompt uh that just outlines what we discussed uh in terms of the use case and will get me started uh building and working with chat GPT. So here my prompt just calls out that it's going to help me with sales meeting prep. It calls out the various tools and apps that I might need. Um in this case it also says here I have a template for you to use. Uh 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 uh what you'll see is that chat GPT is going to guide you throughout the build process. um is going to create an outline for the agent plan that I can review before it gets started building. Um but I really can just work with chat uh GPT here. Um just use natural language, give it feedback, iterate together and we can build a really powerful complex agent. So here I have um you know the agent plan everything I need the capabilities look good. So let's get started with the build process. Um, so what you'll notice in a couple of seconds here, um, is that the agent is going to or sorry, chat GPT is going to continue to live in the left side pane here, um, while it works on the right side to, uh, wire up all of the, you know, scheduling, the apps, uh, the tools, and also create a fine-tune instruction set. Uh so this is a really great noode way to build agents um that still as Christina mentioned uh work like our codeex uh agent so it can run across longunning tasks uh behind the scenes access data and take action across systems. So pretty shortly here uh we're going to see uh chat GPT finish up wiring up the agent and we'll be able to actually do a test run and show you all of this live. All right. So, we'll give it um a little more time here and I'll start to poke around as it um finishes configuring some of the different applications and tools and steps. All right. So, you can see here um it's finished up the instructions. Um it's going to um you know continue to build out the agent. At this point, I can go in here and make modifications directly. Um or again I can provide that feedback to chat on the left side and it'll take care of all of that configuration. All right so almost done there. Um let me actually wire in some tools here. Uh 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 um you have the ability to you know lock down what the agent actually has access to. Uh, so in this case, we're looking at Google calendar. Um, and I just need the agent to reference my calendar. I don't need it to take any write actions or anything like that. Uh, so I'm actually going to disable um, you know, those actions uh, for the agent to take those actions. And I can know confidently that anytime it's accessing my calendar, it's not going to um, you know, adjust or change any events. Um, it's just going to go about the workflow that I've described. I'm also going to quickly add in Google Drive um as well as Gmail um actually seems like okay there we go. So I'm going to add in Gmail here um you know similar configuration um in this case I actually wanted to run an email to me uh 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. uh providing them a playbook um 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. Um so we have the apps and tools that the agent can use. Um now let's actually give it some additional direction on how to best support uh my meeting prep creation. So the great thing about skills within chat GBT 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 uh you can bring them in and have your agent use the same uh so that it can scale the expertise um and consistency across your organization. You can also choose popular skills within your organization that already exists. Or one of my favorite um kind of tools here is actually being able to create a new skill and have chat GPT 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 um this agent playbook. It'll kind of keep going through this. Um so we'll give it a few seconds here. Um, but I also have um a kind of refined agent that we can skip ahead to um and be able to jump into a preview run. U but I always like to kind of observe uh what chat GBT and the agent is doing as a part of the build process. So uh we'll take a look uh uh at what it's doing for a little bit longer here. All right. So, it's finding that template within my skill or sorry, finding that template within my Google Docs um and adding it in with some formatting instructions so that um you know, and I'll show you the final product um coming up here, but it's going to beautifully format these meeting briefs for me with tables, um 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. All right. So, this looks um like it's working out well. Um I'm actually going to skip ahead to an agent version uh that's maybe refined a bit more. You know, I've had some time to work with it. Um and I want to show you what it could look like to easily run a preview test as well. So, here um I'm going to start up a new session. Um again, the same agent maybe with just a little more refinement. I've given um chat GPT some feedback and we've done some preview runs. Uh, so I feel pretty good about this agent and I'm at that step where I can schedule and distribute it. U, but let's let's actually do a quick preview test run. So I'm going to ask chat to do that. Can you run a preview with my calendar for tomorrow? So much like that experience of um or the no code experience of building these agents, you can also work with chat GBT uh to actually, you know, run tests for you. Like I said, you can give feedback once uh these runs are executed and you can essentially, you know, continue to work with it in this way. Um so you don't need IT and engineering support to build out these workflows. We really want to make sure that these subject matter experts um are the ones that can build out these flows and again distribute it to teams and share them um as needed. So cool. Right on quue. We're going to get this preview test uh kicked off. Um, and the other great part of these preview runs is that you can see exactly what the agent is doing. You can see, um, you know, how it's pulling information from other systems, um, what its chain of thought is, um, and be able to follow along. Now, this isn't going to be a view that you babysit or that an end user will necessarily see every time, uh, but it's great for those preview runs and also show you how can how you can look at previous um, historic runs of your agent. Um, but this is of course especially great as you first build out this agent and um you do some tests before you put it into a production setting. All right, so as you can see we're following along here. Uh much like the agent outline and uh the other ways that uh chat tpt uh provides workspace agents functionality. uh we're able to see the highle um kind of plan for this agent as it executes this workflow. Um 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. Um, and then we're going to get a final result pretty shortly here, um, that we can review. Um, and you know, I'll actually then show you how we can schedule this agent, share it, um, and distribute it to your team for the for them to remix and utilize for their own workflows as well. All right. Well, we'll let this run a little longer, but um in the demo industry, as like to call it, uh we have this concept of cooking demos. So, much like on TV when someone's u putting together a meal and they pull something out of the oven right away, um let's actually skip ahead because uh want to be able to get to your questions and the second uh demo agent as well. So, if I go into my inbox, um I have a previous run um that I used with this agent. Uh, so you can see here, you know, really great formatted email with all the briefs that I need. Um, 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. Um, and I want to be able to go in and prep for Blossomart, Pedal Pay, um, as well as Nectar Works. Um, 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 uh information. Um, 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, uh, 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 um and what our shared goals and outcomes uh for the meeting will be with the customer. So um as you can imagine this is saving me, you know, hours on a daily basis prepping for all of my customer meetings. Um 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 um 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 um of the agent and its execution. So what you can do after you know running through tests is start to um share and distribute this agent. So um you can obviously add a schedule for yourself and we would want this to run daily for my workflow. Uh but let's say this starts to work really well for you and everyone else wants to have the same uh type of you know efficiency and hours saving that you're seeing with this agent. Well, you're able to then enable uh 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 Chat GBT to reconfigure it. Um or maybe there are additional steps you want to include like having chat GPT also create a slide deck for you after uh putting together briefs. You can um hopefully tell that there's a lot of customiz um customization and flexibility that workspace agents provides to you. Um so hopefully you're inspired uh to build out an agent that's similar. Um again you can refer back to this cookbook uh to be able to go through uh the agent um after this session as well. All right. So, that was a fun one. Uh, what we did for the meeting prep agent is we built an agent using natural language. Uh, we went into the configuration of the app parameters and access. Uh, we enabled skills. Um, and also memories. I'll include that in the uh second agent. Um, and then we ran an agent preview. Um, so we can check out the agent behavior. Um, and once that looked good, uh, we decide to share it um, share the agent for team use. So, um, really helpful agent, uh, in my day-to-day. Hopefully folks, um, 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, uh, 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 um, is modeled off a real agent that we have um, living in one of our Slack channels. I'm sure everyone uh at your company you have a channel like this where high volume high sensit or time-sensitive requests are made in Slack to get a software tool by a employee you know when they need it. Um and typically those channels um 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. Uh, versus having an agent take that action for you where it will research, um, the capabilities of your requested vendor. It'll compare options against the approved stack, you know, reason across uh, details like utilization of existing tools and then it'll provide guidance to the user or even take action. So, um, you know, in the cases that a human in the loop is required, it 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 uh this example. Um so let me move into this experience here. So, um I'm going to start off from scratch here once again just to show you that even if we're building an agent that um maybe takes on a little more work uh maybe is a little more complex than a personal kind of meeting prep agent uh that you can still start from the st starting point. So here uh we have you know a prompt and in this case I actually have a skill that I want to bring in. Um, so I'm going to actually uh bring in a skill that our product um or sorry, excuse me, procurement team built um to codify their workflow and best practices when it comes to these uh sort of decisions and evaluation. But much like the previous agent, I'm starting off with the systems and um tools that the agent will utilize and then I'm giving this skill and we'll have chatbt wire it all together. Um I'll skip ahead a bit. We won't go through the full build process because we took a look at the previous one. Um, but again, uh, 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 uh, 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, um, things look good. So, we'll get started here. And maybe while this uh wires together, I'll take the opportunity to showcase uh the mock data set that this agent is using um you know in our production agent actually leverages a software kind of management platform and a custom MCP. Uh here I brought that into Google Sheets just for simplicity of the demo. Um but I pulled this up to showcase that you know the agent's actually going to look across all of the approved software. um you know the utilization of uh the tool as it exists today um and the functionality and the capabilities of each tool that we have so that when a user is requesting a tool uh for a very specific purpose um the agent's able to reason and know that if it's making a recommendation from our approved stack um it's actually going to meet the needs of that user. So that's what we have provided to the agent. Um, 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, uh, the interface where these requests are being made. So, if I go back here into Slate, um, you know, obviously there's a lot of configuration going on here. Um, so maybe I'll skip ahead to, uh, previously built version, but I encourage everyone to try this out. Uh, especially when you go through it for the first time. It's one of my favorite kind of parts of this product. Uh watching Chat GBT put everything together and allowing you to follow along as well. All right, run on Q. We have uh the instructions. Um but yeah, I'm going to skip ahead to that sort of refined agent um and walk you through a couple of setup steps before we go into Slack and show a live test run. So, uh, first and foremost, um, what we have here is, uh, the ability to actually add Slack. Um, so I'll showcase that by adding, actually, I'll I'll pull this up, and you'll be able to see, um, that you can specify what channel that an agent should operate in. Um, as well as, you know, whether it should respond when mentioned or for any, um, you know, relevant messages, uh, that it can handle in the channel. Uh, you can also add additional instructions. Um so how to handle requests, how to respond in thread um things like that. Uh so there's a lot of customization that you could do um when it comes to integrating this agent into Slack um as well. So here um we have the agent. One thing I forgot to call out on the previous agent, so I'll bring up here is this concept of memory. Uh so memory allows your agent to uh save notes, you know, context, outputs, um other things that will make it uh better at its workflow over time. Uh so you can think of this as a persistent set of uh contexts that the agent will keep um and be able to execute on in its workflow much like it does with skills and tools as well. All right. Well, everything looks good. Um so let me show you one more thing in the platform. will actually go into Slack. Um, so on each agent 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, uh, this is where for each agent that you create, you also have this centralized place uh, to view all of the previous agent runs. Um, and you can go in here and check the agent trace down to that level of detail. So um this is especially handy of course when you have an agent like this that is autonomous and runs on its own uh so that you can go back to any requests um and everything is centrally audited and um and captured for you. Um and this of course is also exportable via our API. So just wanted to show that before we dive into the live example. Uh so here I have my Slack channel and some previous test runs. Um, I'm just going to show you what it looks like to kick off one of these uh cases and then I'll show you and jump ahead to the agent response uh because it does take a little more time uh than of course a chatbot. Uh 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. Um but the agent is going to go through and um you know do some research on screen studio. is going to uh reason against um you know the policy the scale um as well as the approved software vendor list uh 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 um you know for the same exact request uh 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 uh versus me being able to self-s serve Screen Studio. Um it clearly lists out the reasoning why the rationale um you know we actually have an approved tool called Bloom um that would work well for my needs of needing a high quality uh demo recording tool. U but it's blocked right now because it's overutilized. There are actually no available licenses. So that's why um 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 um in to use screen studio in this case. Um but you can see that the agent is responding with the sources checked with all of the reasoning and um 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 agents done that work for me and as the final step here um 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 Jura as a task uh for them to review and quickly unblock me from uh getting access to a tool that I need. So, um, you know, this has been a really powerful agent. Like I noted, it's actually handling all of these requests for our team today. Um, and saving our IT and procurement team a lot of time. Um, but also, you know, making sure that users when they have these time-sensitive requests because they're typically made last minute, uh, they have the tools that they need. [gasps] All right. So, with that, uh, we went into the IT agent. um how to build an agent with existing skills, how to set up Slack so that the agent can serve teams within other interfaces. Um 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. Um but with that, I will pass it back over to Christina to talk through getting started. Great. Um, 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. Um, 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 Chatbt 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. Um, 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. Um, and we'll have an automatic converter from GPTs to workspace agents um, coming soon. Um so I also want to take a quick second to talk about um GPTs and what um this means for our GBT 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 GBT. Um but when we first launched GBTS 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 GBTs. shared agents that can run multi-step workflows across tools on schedules with approvals and follow-through. And so if your team already has GBTs, 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. Um, now we have some time to answer some of the questions you've been submitting. Um, so the first one that came in, quite a few Microsoft suite users out there. So maybe Hojan, can you help explain um do agents work with Microsoft tools and a bit about like >> Yeah, absolutely. Yeah. So um I did use the Google Drive uh mostly because it's easier to set up for my demos um and I'm familiar with them, but you can use Microsoft tools like SharePoint, Outlook, and Teams. Um, and as you saw, you know, what uh 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'll be able to replicate what I built uh with those suite of products as well. Uh the last kind of distinction I'll make um when it comes to the Slack example, agents can run in chatgpt and slack today. Not quite yet for teams or these tools, but we are actively investing and adding more soon. >> Awesome. Thank you. >> All right, the next question is about memory. Maybe Christina, do you want to talk about how agent memory is different? >> Yeah, so Hoin touched a bit on this in the demo. Um, 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 um future runs. Um, and you can prompt for it to add it explicitly. It can also choose to add it automatically. And these um this file system is specific to every channel that the agent is in. So if you are using this agent in chatgbt or you're sharing it with someone else um in chat GBT, uh every user has their own file system for their own um instance of the agent. Um 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. >> Very cool. >> Yeah. I see. And the improvement over time is an interesting bit. >> Um, 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? >> I can take this one. So, um, agents can kept be kept private. They can be, uh, specific to your workflows like I showed with the first agent. Um 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 um you know these are agents that can only be shared with members of your chat GPT workspace. Um but within those uh you're able to distribute your agents. Uh you're able to allow others to duplicate and remix those agents like I was talking about with the sales meeting prep. Um, and ultimately the admins are able to control who's able to build, publish, and share those agents as well. Um, and so for enterprise and edu, um, that's, uh, within our admin settings, uh, and our role-based access controls. >> For now, we only, um, allow the owner of the agent to actually make edits in that agent, but very soon um, we'll allow multiple people to edit an agent as well. >> Cool. >> Awesome. And can you create agents with different roles? Like any kind of best practices for more specialized agents? >> Yeah. Um I would say we just showed you a couple. Christina alluded to the fact that uh here at OpenAI, we have agents uh across every single business unit. Um 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 uh with all of the apps, tools, files, skills, and instructions that they need to excel in that role. So, um 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, uh they can be guided or corrected over time. Uh so, they can really be improved. Um, 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 chatbt um to, you know, change and improve the uh agent behavior as well. >> Awesome. Thank you. Um, and Christina, do you have to use the desktop app? >> No. Um, so I mean, as we as we demoed, you can um create, run, share these all directly from the web. Um, you can use them in chat, you can use them in Slack. Um, and they run in the cloud, so they keep working even when you're away. Um, 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 um aren't actually running on the app or locally. They're both in the cloud. Uh, one's running on the schedule um and giving me notes for my commute. And then the second one obviously is living in Slack. >> Awesome. Cool. And Hojan, could you answer questions about pricing? >> Yeah, absolutely. So, um, pricing, uh, right now, Workspace agents are in research preview and they're, uh, free, uh, until May 6. So, you know, I'm sound like a broken record, but uh, when I say to when I'm encouraging you all to try it, you know, really put it through its paces and try it out during that period. Um, after that usage will move to credit based pricing. Um so the number of credits that an agent run consumes um will obviously depend on the complexity of the agent and the tests that it performs. Um but you can think of it similar to how we um or more complex codeex tests like you that use more credits. Um because again these are agents that have the same harness and the ability to perform longunning complex tests. Uh we'll be sharing more specific guidance uh through admin and building channels regarding pricing as we get closer to um that May 6 date. >> Awesome. And then we actually had a couple more questions come in. Um so since you mentioned codecs, Christina, maybe you could talk a little bit about what makes workspace agents different than working with codeex. >> Yeah, definitely. So I think um first of all, they're they're meant for teams and so they can be shared with um people in different functions at your company. Um they also run in the cloud and so they run even when your when your computer is closed. Um and then yeah again they can be kind of deployed in these different environments as well. So used directly from Slack with um many other triggers coming soon. >> Awesome. And then quite a few questions around skills. Um 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 um 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 codify or maybe you're already using in different uh platforms and tools. Um, and the agent is the worker that actually leverages that skill and other, you know, instructions that you have uh to do the real work. So, um, an agent can have multiple skills um, and be able to utilize them for the specific, you know, task that that it's working on. Um so you can think of it in that way. Once again skills are going to be like your best practices. Sometimes there are scripts, there are other uh uh kind of direction and guidance and then the agent will actually take on the skill. It'll use also the apps and tools and execute on that work uh as a whole. >> Awesome. >> Yeah, >> thank you so much for taking these questions. Um yeah, I think we're going to wrap up and thank you everyone so much for joining today. So to wrap up, here are just a few helpful resources that we will share after this session. Um so like Cojin mentioned the cookbook for building the sales meeting prep as well as some other um resources with additional demos. And then also want to 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 uh to sign up for those sessions and also watch past build hours. Thank you so much for attending and thank you so much to Christian Hojun again. >> Yeah, thanks all. >> Thank you. Thank you.