ChatGPT Workspace Agents: Automating Workflows with AI Integrations and Memory | EP 140

Speaker A: It's just like there's just so much you can do and the power actually comes from the integrations and the memory and the skills and just amazing. It's like you have someone that actually does the work for you and then you just tell them, share some connections with them, like just your products you use, and then tell them instructions and then they do it for you. But this time, it's an AI agent.

Speaker B: Hi, my name is Dimitri Panici and I'm a content creator, agency owner and AI enthusiast. You're listening to the AI Agents Podcast brought to you by Jotform and featuring our very own CEO and founder, Idekin Tank. This is the show where artificial intelligence meets innovation, productivity, and the tools shaping the future of work. Enjoy the show. Hello and welcome back to another episode of the AI Agents Podcast. In this episode we're going to be talking about ChatGPT workspace agent, new thing that, uh, ChatGPT has recently released. It's cloud hosted agents inside your ChatGPT and are a big upgrade from what custom GPTs were before. How are you doing today, Adikun?

Speaker A: Hey, Dmitry. How are you doing?

Speaker B: Doing good.

Speaker A: Yeah, I've been, uh, busy. Uh, we are very busy at Jotform building all these AI tools like every day and. Right. We are building all these like, workspace features. Uh, ChatGPT came out with this Workspace Agents feature and it's like, I think it's really good. Um, it's really buggy. Like when I tried, like, you know, uh, uh, there are a lot of buggy things, but uh, it's also really good, like the potential is there, like to be able to do so much, um, especially like, you know, um, if you want to, like, if you are doing something every day, every hour, uh, like it's just so easy to automate network, uh, with the ChatGPT workspace agents. And that's why we wanted to cover that today. And um. Yeah, go ahead, start.

Speaker B: All right, I'll kick things off by just kind of showcasing what um, they've uh, given to us as an example of um, good video to kind of kick things off as to what it is. And they're kind of opening one page. Here it is. Give me a sec. Sa. It's m. Yeah. So this is basically something that I think has probably been in the works for a while. You know, it's. It's um, cloud side agents that can actually like, do things and interact with things and use different MCP connectors and whatnot. Where we were kind of Alluding to this a while ago with mcps, how there's just going to be so much of a difference and connectability. Whereas now uh, or in the beginning it was very active, you had to like kind of chat with stuff and interact with the mcps. But now it's a little bit more um, active on its own. Um, so you know it's powered by Codex which is one of the big things that ChatGPT has recently improved on and um, they can take on a lot of tasks that people already do for work. As you saw different things like preparing reports to responding to messages, writing code and they all run in the cloud. Um so there are a lot of really good examples Adikin's got here. But some interesting examples you could think of would be software reviewer, product, uh, feedback router, weekly, ah, metrics reporter, lead outreach agent, third party risk manager and um, much more. So you know you'll see a lot of the examples that people um, use. Um, like it's just basically what you do for your daily tasks that you don't want to end up doing. I mean anyone who's ever put together a weekly report, you know a lot goes into what makes a weekly report but practically an AI can do that report very easily for you. Um, I think what's interesting to me about this is they only actually released it to their like kind of business plan. Um, whereas I even pay for ChatGPT Pro, um, which is a personal thing and I don't have access to it. So it's definitely like more of a team based thing. Uh and it's very interesting to see uh, them release this because there have been some other products that have come out recently which I can compare and contrast um, in a little bit. But I did want to first just state how does this different than what the product was prior with custom GPTs. So like custom GPTs it couldn't run when you were logged off. There's no scheduled capabilities, you couldn't interact with Slack, which is a big one. Um, there was limited code execution with Codex, there was no multi step workflows with memory, there was no approval gates on rights, there was no shared with workspace capabilities and the skills, the skills standards were a lot worse across the board. It's a big improvement from when custom GPTs.

Speaker A: I think what uh, makes this so powerful is three things. Uh, the first one is integrations, right? ChatGPT added this app store recently and uh, there's just like so many great apps there, like Gmail is there, Google Drive is there, you Know, Slack is there, like JotForm is there as well, right? Like you have all these great products that are ready to integrate with like ChatGPT can now, like the agents can now use these products, right? That's the first one, the integrations, uh, or the apps, whatever you want to call them. The second one is these skills. You can actually create skills now. And skills are like kind of like bunch of uh, files, basically. Like you just every time, like, let's say you do something every day, right? Uh, uh, whatever you do, if you just describe what you do and then if you try to get the ChatGPT to do it and if you see that it makes a mistake and then you update that file and even like ChatGPT can actually update those files as well. Like you are creating a skill. Skill is not like something magical. It's just a file or it's like combinations of files, right? Three files or five files that describe your workflow or whatever. So you have the integrations, you have the skills. And the third one is the memory. It can actually remember stuff. Uh, you can give it a list or database, like a spreadsheet or something. It will remember all the things from the past, so it will use that information. So and combination of these three things makes it so powerful. And everyone is talking about how like it's, it's SaaS is over. But I feel like this is just making SaaS so much more powerful because now you can easily like, it's like someone is using your, you know, going to your Gmail, your Google Drive, your Slack. Obviously that someone is like, you know, chatgpt, so you can, you know, it's not some person that you don't know. It's, you know, you can trust. Uh, and that AI is basically doing the task for you. And now you are basically, when you give access to your calendar or your, you know, Slack, now it's able to like, you know, send you a Slack message or like, you know, put things on your calendar, right? It's just able to do the work for you. And that's what makes this so much powerful. And what's great about this is it's actually running on the cloud, which means basically, uh, uh, you can say, okay, hey, uh, I created this agent. It should run every hour. And if you shut down your computer, it's still going to run every hour. It's because it's just running somewhere else, uh, on ChatGPT server, so you don't have to keep your computer open. So that also makes this so much Powerful and yeah, we have great examples. But um, do we want to compare to Claude's uh, many agents? Um, yeah, for sure.

Speaker B: There's another product that Claude recently released called which is very similar to the what they have here.

Speaker C: Claude Managed Agents is a suite of APIs for building and deploying agents at scale. You define agents with specific tools, Personas and capabilities. You configure sandbox environments with the right packages and network controls. You fire off sessions from your own application and then Claude does the work inside an isolated container with full file system access, bash execution and web search. I have right here a Kanban board sitting on top of Managed Agents. I drag one over to the in progress and then that fires off a session automatically. Now the ticket says optimize website performance. So my backend creates a session. It points to an environment that I configured with Lighthouse and Puppeteer pre installed and mount my GitHub repo into that container. Claude has the code base, but it's

Speaker A: definitely not made for normal people.

Speaker B: No, it's a little bit more advanced, a little bit more advanced.

Speaker A: And uh, you know Puppeteer, it's made

Speaker B: for developers, it's a little bit more advanced.

Speaker A: But uh, the ChatGPT workspace agents are definitely made for normal people. There is no like technical jargon like this. Yeah, no, I think that's the main difference. Like uh, it's a little bit more,

Speaker B: it's a little bit more advanced. Everything seems to be a little bit more like developer oriented with Claude versus ChatGPT.

Speaker A: And my examples are also like for normal people, not for technical people. Should I go ahead with my demo?

Speaker B: Yeah, for sure.

Speaker A: All right, uh, let's see. Here I am. So here we are, uh, I'm on chatgpt, right? Uh, it says developer mode, but uh, I guess I enabled some stuff. Uh, but here's the thing, the agents are not available on regular uh chpt. You have to have like a business account, education account or enterprise account. And here I am, I'm showing from our enterprise jotform, uh, enterprise account. Um, so like you know, I mean if you think about it you, you are either a business uh to, to use these kind of automations. My guess is you're a business and uh, like you can get like two person business account and it's still you know, cheaper uh than like this max plan or something. So it's not that unreachable but it's like little bit expand, more expensive compared to the like regular chatgpt or the free chatgpt. But um, anyways so it starts showing these um, agents here. Now. Um, so basically uh, I'm going to create like two examples. I already created these examples because it's takes a long time. So uh, I create them but I will show how it creates as well. So let's start with the J for uh, like uptime minute monitor account. Let's say, you know, you're a small business, you have, you have a website, but you want to make sure that if your website goes down you are aware of it. Um, so create an uptime monitor agent that checks a list of URLs every hour and email me if there any of the pages are down. So basically this will notify you ah, if your website is down. And uh, you want to make sure that it's always up. Um, so it's very similar to like newchat and then you just type it here. Um, so did I write it on chat? Uh, I think I, you know, I need to write it on the. Okay. On the agent. I need to create it as an agent. Okay, sorry about that. Let me start from the agent part. Create an agent. Okay, so I basically here I click on this plus sign. Um, so this is the agents page. Uh, basically you can pick from these templates. There are like templates here. Uh, and the, the agents you built, they are listed here or also in your business or like you know, your organization, if other people share their agents, you can actually reuse them, you can duplicate them. Uh, for example, this uptime manager agent I just created today, someone else can actually, uh, because I shared with my, with whole Jotform, some other employee in Jotform can actually use this agent to track some other URL. Um, so I'm just going to create a new agent here. Let's click on Create Agent. So the create agents say like uh, this stuff is uh, really cool. The icon, the eyes are always same. Um, and I'm just going to type it here. Create an uptime monitor agent that checks the list of URLs every hour and email me if any of these pages are down. Okay, that looks good. So I'm going to submit this and basically it's going to start like. So they have this agent builder. But um, you don't have to use the agent builder directly. Basically you have this uh, like ChatGPT, uh, builder that actually builds the um, agent for you. But once you build this agent you can actually go to the like the Agent builder page and then you can customize it manually. It's very similar to how we do a jotform like when you need to create a form, uh, you just tell jotform, uh, uh, AI like what kind of form you want and it creates it for you. But once it's created you can always use the form builder to edit it manually. It's very similar to that uh, process. So it's saying I'm checking whether the available app can support emails. Okay. It says this agent will check list of URLs. It uses Gmail, run every hour in ChatGPT. Okay. And check website uptime, detect outages, email incident alerts. Looks good. I like everything. I'll just hit start building and it will start building the agent. So here's the cool part. It's building the agent inside the agent builder. So the this is the chat part and this is the agent builder. And now this is kind of similar to Atlas. Uh, Basically the agent ChatGPT is actually building the agent for me and I can actually see how it's making changes. For example here you see there's no instructions here but it will come here and put instructions as well. So it will just do this thing uh, to create the agent for me. And I love this. Uh, it's following Riley Curses.

Speaker B: AI making agents for us. We don't even need to, we just

Speaker A: tell it the ideas and AI is creating the AI agent for us and we can see the reasoning here. Write an agent instructions. So it's going to write the instructions here, uh, for URL monitoring and email Alerts, set the ChatGPT description, add the earlier schedules for it. Okay. Did you see that there was no instructions and now it's actually putting these instruction here. And while it's working I think I can also edit this right. I can put stuff like you know, updated myself as well. All right, so both the uh, AI and I can actually work on it at the same time. Looks uh, like it's almost finished. I still need URL list. Okay. Uh, it needs the URL to check if it's up. So I'm going to give the URL and this is the URL. So um, I'm giving it the URL and basically it's just going to add this to the URL. So I was mentioned that there are like three things that important things that the agents have. The first one is integrations. It integrates with uh, chatgpt apps like Gmail, Slack, you know, Jotform, whatever you want to use. And the skills are basically files that describe similar to these instructions describe how to and they're reusable. Right. If you create a skill, let's say you create like a writing skill, SEO skill or you know, uh, like how to respond to customer skills. And once you create that document, you can actually use it on all of your agents or whatever you are building with AI, you can reuse them and usually these skills uh, can uh, be transferred From Cloud to ChatGPT to Gemini. You can actually reuse them as well. So uh, basically, um, it looks like. So the schedule is here. It also updated here. It added this schedule every hour. ChatGPT. Check this URL uptime. Let's take a look at this hourly schedule run Every hour channel ChatGPT and instructions. Check this URL appears if it's a very down uh, report. So uh, I can also manually edit whatever it is doing. So this is amazing. Like they didn't just build an agent builder, but they also build an agent that builds the agents. So that's pretty cool stuff. And agent is ready. Try and chat. Let's try it. Uh, so basically, and it was pretty fast. So I don't have to show the one that I already did previously. Uh, yeah, check if it's. Check if it is up. Uh, Someone in aged activating computer. So because these are kind of running on the cloud. Ah, they are not running on this computer. It needs to activate something every hour, I guess making um, a plan. Uh, so let me show you the, the one I created previously. I need to URL check example. Oh, every time. I guess it doesn't write it in the memory. Okay, I will give it but I will show the one that I did previously. Uh, this is, yeah, this is the one I made. And then uh, like we can actually see like analytics when it runs. We can actually see it. Okay, this one I created actually it run like uh, multiple times so far. Uh, run run on ChatGPT. Like some of them I run manually. Some of them run on schedule. And like I can actually see when this one run on schedule. Let's see if I can click on this and see it. But I can actually see it somewhere else. Um, I should be able to also see it on like there was a place where I could actually see the runs. Not this one. Uh, I guess I'm also learning this. But there is a way to actually see the output of those um, runs. Uh, analytics. Okay. I couldn't find it. Okay, let's take a look at this one. Okay, it checked jotform.com, it appears up. I was able to load the homepage it says so that's good. Um, let's see. I can um, basically I can go back here and I can see the activity and I can click on this. Okay, this is where I can see the previous rounds. So basically if I go to the agent page and if I go to like activity, uh, and I can do this for the one that I already created as well. I can go to the activity and then I can actually see uh, what happened here. So that one was one agent I created. The second agent I'm going to create, it's going to be, um, basically, uh, create an agent. Um, basically I'm going to create it here. Build an agent, create an agent that checks for Google Docs file every day and reports any changes by email. Uh, to like my demo email address, I put it here. Let's see. I need the Gmail integration to be able to send emails. I also need Google Docs integration to uh, check the file or just provided the URLs usually enough. So while this is working, I'm going to show the one that I already created. Okay, here we can see the one I created before. And uh, basically it's in the integrations. I added Google Drive and Gmail integration and then it has the memory. So I provided this Google Docs file. Uh, so this is my Google Docs file. This is the first version and I will add something here. Let's see if it will be able to catch. This is the second version. Let's see if it will be able to catch this new addition here. Okay, I'm going to go back here and run it. Okay, let's run it and see if it's going to be able to catch it. Um, I wonder if it will work because I didn't run it before. So maybe it doesn't have the second version. But what I'm going to do is I will just do the third version. Uh, run it now. Bring an agent online. Booting up computer. So one thing is, uh, this is going to use credits, I guess. Um, I think they were free until like yesterday. So it's. This one is probably charging us credits. Uh, uh, but I doubt that it's very expensive. Um, but yeah, you want to be, you know, mindful of uh, like if you just run it too many times, it might cost a lot. I'm king of the document. Check nav, load in the monitor, read the current docs file snapshot, send the baseline change email if needed. Let's see if it will be. It will be able to catch the changes. Okay. Because there was no previous version. It didn't. Okay, um, it's going to send an email now. It asked for approved. Approved probably because it was the first time. Monitoring started the test Google Docs file. Okay so it's going to send me this email here. Applied uh, page to creating files. Save a memory, change the baseline saved locally. So it put like you saw that memory section, right? There was this memory uh section on the app. Right. Memory. It's actually saving it here once it runs it. So title the test Google Docs file content. Okay, so uh, okay, let's take a look at if I received an email monitoring started. Um, my test Google Docs file monitoring has started for Google Docs file. No priority saved versions. Okay, here's what I'm going to do. I'm going to change this file, the third version and then I'm going to run it again. Um, was it here? Okay, run again. Let's see if it's going to discover the changes and report it to me. So as you can see, this is like the agent. Basically it's showing the agent here. So I can always disable the agent by clicking on here. So basically, um, this is a new UI that shows you if you are talking to an agent, not just ChatGPT planning, um, monitoring Google Doc compared to previous version. Always approve it for this chit approve. So I guess if you create an agent, uh, you want to uh, first run it a few times yourself, perfect it and once it's ready you can go uh, ahead and uh, run it hourly or daily, whatever you prefer. Changes detected. Okay, that's great. Okay, it's going to send me an email. Let's take a look at if it sent me an email. Uh, where's my Gmail? Okay, uh, did I close it? Okay here. So let's take a look if there's a new email refresh. Okay. All right. Changes detected. Edit third version. It even like it was able to discover what changes what was made. So this is very useful. Let's say you shared a Google Docs file with someone and if you know if they, you want to be triggered, uh, whenever someone makes a change. I think Google Google Drive already has this feature. But uh, this is just an example and you can receive an email, you can send it to Slack. I was just working on the Slack integration. I had some problems so I'm not going to be able to show that demo. But basically, uh, this is amazing. Um, like just you can do so much with this.

Speaker B: I would just say. Yeah, no, I think it's interesting because ChatGPT and a lot of these new platforms, it went from you having to really put a lot of work into prompting and making these different agents that would then have to be manually worked with. Whereas now it's literally gotten to the point to where you can tell it, hey, I wanted to do something, and then it can do the research on what it means to do that. Well, then do it for you. We're really working in a totally different landscape than we six months ago.

Speaker A: It's amazing. Uh, it's just like there's just so much you can do. Uh, and the power actually comes from the integrations and the memory and the skills. Uh, and just amazing. It's like you have someone that actually does the work for you and then you just tell them, share some um, uh, uh, connections with them, like just um, your products you use and then tell them instructions and then they do it for you. Um, but this time it's an AI agent and you could do this, you could do this six months ago, you could do this one year ago, but you could only do this with uh, either cloud code or like, you know, uh, cursor or something like that. Right? You, you had to write code, you had to like wipe code it. But today like, you don't have to wipe code it. Like you can just any normal person, they can just go like get a business account, education account and then just uh, start building this kind of stuff. Yeah, amazing.

Speaker B: I'm curious by the way, just for what you're doing with Jotform, what is kind of the long term vision that you see for different AI agents and how you're trying to implement stuff like this, I guess in what you're doing with Jamaform agents. And I know maybe it's different or just how does it give you inspiration for what to do to improve our own AI suite?

Speaker A: Yeah, um, so these days, uh, we are working on like for example, Jotform's own uh, workflow features. And uh, we have very similar features. Um, so the question is like, you know, this one already has like so much. Even Jatform integration, it can do many things. Um, I think the main difference is like, uh, I think it's about really um, nailing down things, right? So you can actually build all the stuff that we are building with JotForm AI, but you have to spend a lot of time here, try to get things right. But we usually provide these already optimized features, uh, and products, uh, as opposed to building, building them from uh, building them, you know, building like this, like, um. So I think that's the main difference. The other thing is like, you know, it's just like you have to learn this stuff, right? You also have to have like a chatgpt, uh, business account, which in Jotform, like you don't actually need that. Uh, you know, uh, even with Jotform free account, you have all these features. Uh, there are some limits, but basically all these features are available. We don't even ask for API Keys or ChatGPT Connect, uh, or anything like that. You can use all these AI features. They are just coming with Jotform, uh, already optimized, opening, uh, correctly executed. Uh, so I think that's the main difference. Um, but these are amazing, uh, so this is just inspiration, uh, for us. Just being able to do all these things, uh, within our own product is uh, you know, that could be like really exciting. Um, so, yeah, definitely. I think there is just like, um, there's this worry that like SaaS products like Jazz form, right, uh, or other products, you can actually build these products using know, Codex or you know, uh, cloud code or like build agents that can also do the same things. And if you have a very simple product, maybe you can. But, but the difference is like uh, products, like established products, they already, you know, they have been building these products for a long time and they, you know, we made all the mistakes, right? And when you try to build, if you try to build your own sets, you're going to make the same mistakes and you're going to. And it might actually cost you even more to try to host this, uh, run this every hour, uh, just doing so many things yourself. So I think there is a balance here, right? If you want to do um, stuff that been really sold by SaaS products, I think it's better to cheaper, easier, uh, stable, uh, less buggy, more secure to let the, you know, the managed products like SaaS products to handle those things for you. But if you are building, like if you're integrating all these SaaS products to each other, if you are building all these workflows, all these custom workflows, then you know, use products like uh, ChatGPT, um, you know, like ChatGPT agents. What was this called? ChatGPT workspace agents. Workspace Agents, yeah. Yes, Workspace agents to kind of uh, like duct tape them together, like build these uh, things together. And uh, I think it's like just you became so much more product. Right? Right. In the past you had products like Zapier make, uh, right. Na, 10 things like that that can actually like manually integrate them to each other. But now these products are actually much more flexible, much more powerful, smarter to kind of do the work between these products. Right? Uh, let the QuickBooks handle the. Really like, you don't want to learn. You don't want to learn accounting. Like you don't want to learn the laws, uh, of accounting and like just mess with taxes and stuff like that. Let QuickBooks handle that, right? Uh, and you want to let Jotform handle your emails, right? Because we send, you know, 10 million emails every day. We know how to send emails. Right? Uh, and just, you know, don't put, you know, uh, don't lose emails in spam, things like that, right? Like you want to make sure your forms are stable. They have like the uh, uptime, like, you know, almost 100% uptime, things like that, right? And secure forms. Right? You want to have Jotform handle your forms or like those form email emails that come from forms or like your, you know, the data or if you're like a healthcare company, uh, like the hipaa, you know, you can sign a BA with just form, things like that, right? So you want to make sure that like you're not messing with those things, right? You want to make sure. But when you are integrating with QuickBooks and JotForm, right? If you don't have integration with QuickBooks, I believe we have integration by the way. But if you don't have an, if you didn't have an integration, you could just go to uh, chatgpt, uh, agents, like workspace agents, and then build that integration. Now you can actually build the whole workflow, uh, and duct tape them together with AI and just, you know, uh, automate your work and um, you know, uh, automating your work will save you so much time that you can put that time into, you know, making things better. Save, uh, that time to uh, to work on important things as opposed to like, you know, repetitive tasks.

Speaker B: All right, well, I think just kind of closing things out here. I uh, just say thanks for the insight. Thanks for the showcase. Sorry my uh, things I don't have, uh, the business edition, but I really appreciate you holding down the fourth and uh, fourth and showing things and um, let's uh, keep covering cool stuff.

Speaker A: Yeah. See you the next time.

Speaker B: Thank you so much for watching and we'll see you in the next one.