Try Out ChatGPT’s New Agent Builder!

I think it's a very interesting [music] update. I believe like they made this part of the developer announcements and it's kind of for theirs because like you know uh you can actually create like chat bots build agents connect them to chat bots but still like it's kind of hard to use them if you're not a developer. They have some really cool stuffs. Maybe others can build on this stuff and so it's pretty [music] cool stuff. Yeah. >> Hi, my name is Dmitri Bonichi and I'm a content creator, agency owner and AI enthusiast. You're listening to the AI Agents podcast brought to you by Jot Form and featuring our very own CEO and founder Idkin 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 today's episode, Idkin and I are going to break down a new update, a brand new update from OpenAI. We're really excited to chat about it. It is a part of their agent toolkit, their agent kit that they released a couple days ago. It is the AI agent builder. How you doing? Are you excited to talk about this? >> Yeah. Yeah. Uh I think it's it's it's a very interesting um update. So they have a Dave Dave I believe like um they they made this part of the developer announcements and it's it's kind of for theirs because like you know uh you can actually create like uh chat bots uh build agents connect them to chatbots but still like it's kind of uh hard to use them if you're not a developer but I think they they have some really cool stuff

so um maybe others can build on this stuff and uh and just makes it so much easier to build uh AI agents. Uh they have like AI agent builder that has like a canvas where you can actually build AI agents. So it's pretty cool stuff. Yeah, >> absolutely. Um when you saw this, what was your first what was your initial thoughts on on its impact and what it was going to mean for uh the industry? >> Yeah. Um I felt like this is um and let me share what they have uh so we can take a look while we talk about it. >> So let's take a look at the agent builder. Um right uh so this is the agent builder. Um maybe I will show you this one instead. So this is the agent builder and um and basically you can like in your

agent this is kind of like an um similar to interfaces you where you see like on uh N8 and um >> make.com >> make or zapier like you can actually build um like workflows and um and you can build these workflows based on steps and here I actually built this like singlestep agent uh and this agent and I just put the prompt here. Search online about AI news summarize what happens in the last two days. Uh basically I could just do so much more but uh the agent builder is like um you know you can you can add an agent. Agent is basically uh it's kind of like a single worker in a workflow. And uh and let me show you from their announcement. Um so this is kind of their announcement post and he here they have a customer service example and this customer

service example uh like this customer support agent actually there are like multiple agents return agent retention agent information agent. So for example if the customer is asking about an information u it's redirected to this agent. If the customer is asking about like you know if they are like trying to cancel their account or something you know maybe it will try to retain those customers and the return is more like you know if they want to get refund or like return their whatever they bought it's this agents. So um right in this case there are like multiple agents going it's like you can think of these agents like workers here you have this agent that can classify >> the upcoming question like whatever the question is asking it can look at the question and understand like what the user is trying to do kind of like

when you would call a line like uh call on the phone and then you you you hear things like okay uh you know enter one if you want to you know uh cancel your subscription enter two to do so it's kind of like that but you don't have to do like enter one you can just read what the customer is asking and then here you can decide like uh exactly which agent should take care of this and each agent has their own prompt has their own data it can they can use tools so if we go back to the um agent builder. I can basically enter the prompt here provide instructions. I can I can decide on like how much time it should spend like how what kind of model it should use. But I can decide like which tools it it should use in

this case because I wanted to search online. Basically I added the web search tool and I can click on here. >> Uh for example, I could use MCP and and we did an episode about that. I could use MCP to connect to any other provider. Um I could do like file search probably within a computer if this agent is being run inside like a environment like a you know computer environment where it can actually go and uh do this or maybe it can just look at the um uploaded files or things like that and it can actually run code as well. So basically you can actually put your code here as well and then you decide on the output format. the auto format kind of decide depends on like what do you want to do next like in this case I have like two steps

and like just just like you know uh basically it's probably in the news but if I wanted to get the outputs and put it in like a JSON format so that I can actually pass it to like another agent then I would use JSON basically it's structured data it's like glorified CSV uh and Widget is kind of widget is kind of interesting. Basically, you can actually create a widget and show it on that. >> Uh, for example, in this news, I could actually um let me share the tab. I I just enter this prompt that says like, okay, create me a news widget. >> And basically, it can list the news here. It can show it within the news. I was able to get get it to work like connected to it. But basically if it was connected it shows here. Um and like I

just and this is also fine like uh let me share the tab. Uh basically when I run that uh it was able to give me like the here here are the most important addon in this last two days. US clears large NVIDIA chip. Um EU launches 1 billion apply AI. why it matters uh on ongoing fallout between open air amd and so on. So basically um uh when I run in reasoning agent uh it was able to search the web find the most important things it used reasoning and then it created this list. Um and I was mentioning how like you could also use a widget and I will show you uh an example widget here. Basically uh this is a chatbot. Uh no this is more like a chat uh example. In this case um in this case they are able to like you

you you can use this chatkit uh to create like chat interfaces. So this is all very like developer like so. So I don't think if you are like someone who is just um you know who just wants to use it, you can use this. But um it's more oriented towards developers. But just just to show you like what's possible, they may they have this example chatkit. Um and uh let's say I want if I want to go to Japan, I click I enter Japan. It's going to do uh a search like so uh they have an they probably created this using the agent builder to showcase this but basically it went to Japan. Now it's showing the country. Now I have like you know I completed one of the words. So basically if I'm traveling all these worlds uh or all these countries uh basically

it can just give me information but it's showing like what kind of things it can do. Basically you can customize the page. So you are basically building um uh building apps with chat interfaces. U that's I think that's the main use case at this point with this one. If we go back to this page here. >> Yeah. Uh I was showing this examples but um let me see if there are any other screenshots but here there's a HubSpot example. Hubspot actually built this um built this chatbot and then here you can see like different uh different things. These probably were built using the uh widget uh builder uh product as well. So um this is definitely towards uh more towards uh developer uh audiences more like if you're an enterprise if you want to build a you know product on on this um that's fine

uh but this is like you know I still wouldn't consider this um you know a competitor to Jatform AI agents for example because in JF AI just you just create widgets uh and then um you So you chatbot and customize train it. Obviously you don't have this much ability there. You can also connect it to um if you are a developer you if you build your own tools you can connect to those uh tools but uh it's made more towards the like uh you know audiences no code audiences that can actually build uh the stuff. Let me let me share from AI just just as a comparison as well. >> For sure. Yeah. Um yeah uh so here um so yeah J form AI agents um you can create like these uh agents um and we just released the WordPress agent so you can create

like a phone agent uh chat bots uh voice agent uh WhatsApp messenger Instagram like all kinds of places Shopify Gmail right Uh but I think the main difference is that you don't need to like develop uh develop your own UI or like uh like all this stuff. It's already ready with just from AI agents. Uh and let me show you the um basically the uh the agent builder. This is the Jatform agent builder. Uh in the Jatform agent builder like you you just pick like which uh which of the uh you know UI you want to. So you don't have to build your own UI and you but you can still customize your UI if you don't have an offer voice option you can just disable if you don't want to have forms right uh you can just remove it here as well like it's

highly customizable you can customize the design as well like just change the avatar and you know all kinds of uh style changes uh like this. So, and then all you have to do is like you can either link to your um chatbot or you can just copy this code and on to your web page and you're ready to go. Uh or like if you if you're going to use the phone agent, you just, you know, buy a phone number and uh buy a phone number and start using that um uh that phone number so people can connect to your agents. So I would say uh you know there is no more options uh and just you know OpenAI agent builder is like on one end where you have all these like different developer tools and you can write code to kind of build these like

more advanced level uh integrated uh like chat bots or agent products. But there are like products like J from AI agents where you can actually you have all these readymate uh and I would consider like N8 or like Zapier or make in that end like in the end of the Jot form AI agents end as well uh because it's they're already ready like they are ready to be used like you just click on uh click on buttons give authentication right if you're using NA10 what do you do like you just uh connect it to your let's Say if you wanted to look at your uh data in Gmail or if you wanted to like uh you know uh connect to Hopspot or something, you just point and click make a connection and you're ready to go. Uh so I think this is just creating like

so much more opportunities and it's a good thing and I I'm really excited about this. >> Yeah. Yeah. No, I'm really excited about it, too. And I'll just kind of walk you through a couple things that I'm I'm seeing as well that's kind of interesting. Um, first of all, just, you know, obviously we're talking about the builder mainly in this episode, but I do think it's fair to kind of showcase um the full uh announcement that they did make real quick. So, and how it maybe is priced because people usually care about that. So, you know, agent kit is the main announcement where builder was a part of it. So, it's that visual canvas builder that Idken was showing. Um then there's the registry which is a central place for admins to manage how data and tools connect across OpenAI products. The chat kit is

also a toolkit for embedding uh customizable chatbased experiences in your product. And this is interesting because chat GPT as well as claws released an agents SDK right and what this has allowed for people to do and by the way for those that don't know um it's kind of funny. I was talking about this on an interview we did. Most people don't know the acronyms. They just know, sorry, they don't know what the words are in the acronym. So, just so you know, software development kit is what SDK is cuz I was using the term like a like a person who doesn't know what they're talking about and then not knowing that it meant software development kit, which I found funny. Um, but alto together, how does this kind of work um, cohesively with things like MCPs? Because I think you showcase something that's pretty interesting.

Building these agents out that have the ability to interact with MCPs is is really powerful. Um, there's pre-built connectors for sure, but I'm I'm going to give some examples that I think could be be really helpful um that I'm going to try to get my hands on to make my customer experience better for my my clients. So um just taking a look at the bottom right quick cuz I do want to mention pricing. So starting today, chat kit and the new eval capabilities are generally available to all developers. Agent builder is available in beta and connector registry is beginning its beta roll out with some API. Um it seems like basically all these tools are included with standard API model pricing. So it's pay as you go, right? So it's going to be pay as you go for anyone wondering because that I feel like

it's not that obvious. Um, but just to kind of showcase some of the higher level powers of this stuff. Um, let's take a look at something like this customer uh service bot, right? >> Um, as you did I do it again? Did I switch accounts again? I did switch accounts again. Give me a second. [laughter] >> Um, here it is. So, this is the customer service bot build the builder the the pre-default run one, right? you can have little notes for yourself of how it works, but this is an example of a kind of way that it um will function. So, we have this classification agent, right? [snorts] I think it's nice. You can change between the models. That's cool. So, you can adjust the pricing and and whatnot. They actually just released something in this that was a little bit under the radar, but

it's GPT5 Pro. >> Okay, so that's a little bit of bit of piece of news. And what does that mean? Um, >> who knows what that means? Let's actually give an answer to the question. Um because I am cur I am curious uh versus other models what that means. Um >> verse other GPT models like it's announced and they really it was kind of under the radar those was mentioned but nobody seems to care. Um [clears throat] that it's in the API now I think is the main thing. I guess it's you know what actually I take this back. So, it wasn't really in it wasn't in the API, but it was available in the um uh it was available in app. >> So, this guy right here, you see how it says now available in the API October 6th, a couple days ago. So,

it was not available previously is what I'm getting at. So you guys know this tool is now um it's now available in the API um from that standpoint cuz I don't know why I actually am kind of confused why they would release a model like for a month and a half or two months and it not be available in the API because GBT5 across the board was but regardless >> go they they have done that before as well they they they usually do that because >> I think they have a culture of shipping quickly as quickly as possible. So sometimes they will just cut out features while they are shipping um so that you know uh and then they handle the rest later and I think that I think that's great because like we get all those updates like if you try to make everything

perfect and just everything like just you will not ship anything. Sometimes you just have to go with what you have and uh you know improve things uh after the launch and we also do that a lot and um it's just um never perfect. Yeah. >> Yeah, you're absolutely right. So just getting back to this for a second uh talking about how this agent workflow can work like what's I think really amazing about the different tools that it has um obviously web search is something but I really and that's what I was mentioning before with chatkit how you can make your own um different tools uh how it in chat it's a toolkit um for embedding customizable chatbased agent experiences in your product so that was what it was referencing earlier um but I think the MCP stuff's really interesting, right? And the reason for that

is >> we have an agent here that is a bunch of conditional logic. And say for example, you're making an agent that's trying to solve a problem. seems like you're determining whether there's a issue here or not. And then from there, there's multiple different agents that can solve problems. So you're an information agent for answeringformational queries. You're a customer retention conversational agent whose goal is to prevent subscription cancellations etc. Right? So like for example with the information agent let's say most people think about this stuff at like a website level but I think what's going to be cool in general is if you're able to have the triggers for this product, right? the starting point for this product come from something like a a question that your client asks you in Slack or something like that. With MCPS now, you can do such interesting things

where you could like sure it's got its default ones and maybe it's not here like the ones that are here are the ones that it's like pre-built. But, you know, there's really cool to there's really cool tools out there um that have made MCPs like I think of Slack. I think as well of a lot of different things here from an MCP standpoint where maybe you have a client that's confused about an invoice and you stripe you invoice via Stripe. Maybe you have a client that's confused about an invoice and you you know you use PayPal. Maybe um they're confused about where a file is that they're asking about and you manage files in Google Drive. Maybe they're confused about your next booking the next day that you're available to book and you use Google calendar. Maybe they're confused about a previous conversation you had

and you can utilize just the server base, right? And you can add like so like for example, Javform's not here. Yeah. Right. Um, but you could if they have a question about like, hey, I want to fill out this form to make some content like because how I work is like people if they have a request, they'll fill out a form. You know, I can access a list of different forms we have and determine which one makes sense based off of the request. These are the types of little things that I think the MC MCP capability um I've opened my eyes to recently that it's actually a lot deeper and better than um I had considered cuz it's like oh it's like the AI sorry it's like yes it's like the AI's ability I would say for MCPs practically speaking it's like giving your AI

access to the entire list of capabilities in a um API endpoint list and having it be able to access everything all at once and and parse things out for itself. So if you just say connect it to it's like what I think we probably always wanted integrations to be. You know what I mean? Like that's what I've I've always wanted to be like I give it access to the thing and it figures it out. Not me have to map thing. You know what I mean? So >> yeah, you figured out you figured out like what to do >> here. You got access. It's like, you know, if you have like a real assistant. Hey, here's my calendar. Do what you want to do. Uh, go set up appointments for me. Do all that. >> Exactly. Thank you. Yeah, that's what I'm getting at. It's the

f that's why it's agentic because you can it can it can reason and it can you give it complete access to something rather than like I want you to be able to respond to emails coming in. No, I want you to be able to figure out if an email is coming in. What does that mean? What should I do? And then this gives you the conditional >> pallet because MCP kind of knows like uh what's available how to do those things. MCP is kind of uh gives your u gives the AI ability to talk to other services. Um and it's just much better than APIs because APIs usually don't have those kind of documentations. They miss documentations but MCP already includes the documentation. Um so it's really exciting stuff and um and in that example for example the customer service right it needs to provide

refund right uh how does it provide refund if it's like stripe right it it could just use a stripe MCP connection to just uh make the refund uh the agent can handle that as well like just uh because it has the access to be able to do those do those things as well um so just MCP tools tools. Um, they just give agents so much ability, not just answer questions. And this also reminds me like this agent builder is also a big competitor to products like NA10, make or zapier. Um, and I know that you use them heavily. Uh, I know that like your your whole world is uh like you know kind of automated using those products. So I have a question for you. Would you actually use the um the agent kit or agent builder um instead of these products for your automations?

>> Well, I'll tell you what. I think I'm I as you're asking me the question, I was trying to think of examples of what I could do to make it happen because like I I showcased there's a lot of I >> do you have right now like uh could you replace it with this? Yeah, I mean this is a good question because if I take a look at just like the capabilities of tools, um, yeah, anything that's kind of there in regards to the MCP server capability, I would think is is a reasonable thing to replace it with. Like a big problem I run into sometimes is its specifics of things. Like if I could connect the notion MCP to this, I think I would use this very heavily for things like um >> yeah, go to server, type mcp.notion.com. >> Yeah, if I go

to server mc Yeah, exactly. I'd get mcp notion. I know we went through this, >> but yeah, but exactly like I would be able to in my opinion. >> Mhm. >> To make some changes to workflows I have where I feel like I'm wasting a lot of automation credits. You know what I'm saying? Like there's it's pretty annoying doing things that are very humanlike at scale with the content calendar management type stuff where like I feel like I could get an MCP access and be like tell an agent like hey I want you to reorganize the list of content and make sure that the content number order cuz like we number every piece of content like one first thing we make to the last thing and little things like that are actually really hard to automate, you know, in like a publish date moves changing

the content number relative to what's around it, right? And I have implemented workflows in make.com that have chat GPT in there, for example, that do things like um parse out the information so that it then can make the change. Now, what's interesting about this is that MCP itself isn't going to cost me money. Well, the only thing that's going to cost me money here is the tokens from uh Chetchup BT. And if essentially I'm already using that level of tokens in my automation process because of the fact that I have to put OpenAI modules in there to solve complex questions, I might as well just use this. So, >> um yeah. No, seriously. You're paying for NA10 or uh make and then you're also paying for uh open AI chip too. >> Uh why pay twice? So just uh use this. >> Exactly. I agree.

>> And this one also has one other advantage >> uh and that's the chat interface. So basically um you have NA10 but in addition to NA10 you also have a chat interface where you can actually talk to uh your automations and uh try your automations improve them. Um so there are some advantages but I think the main problem with this one is like uh all those integrations uh they're probably not there yet. And um and the other thing is like it's never like when a company is doing everything like if you think about it open their priorities chip right because they earn uh you know their living from chip or like you know investors as well but uh but a company like Zapr or make or nan they are they earn their bread from uh just that product. So the product is like much more

higher level uh quality. Uh and this one like I don't expect this to be really as that high quality as the other ones because like you know they they can only put so much resources in this product. Um you know similarly like Google actually has like a forms product right Google forms. >> Yeah it's pretty bad. Japro has like 700 employees like we are all working on a single product right and Google actually has like I mean we have multiple products a bit like but Google Google doesn't care like you know Google uh you know uh you know headquarters of Google I don't think they talk about Google forms and at Jatform we are always talking about Jatform forms because that's our main product and because like we are constantly thinking about it improving it advancing it it So um the same way I would

expect these uh these companies to evolve and and competition is good uh with higher competition they have to do more. Uh so I don't expect them to go away. I think they they're they're great. They're also using taking advantage of MCP. So they are their integrations are getting even better with MCP. Um so I think it's it's win for everyone. Um, >> well, this kind of goes back to the theory that a lot of the people I'm interviewing seem to have, which is that there is a list of different tools that you're using right now that OpenAI is essentially and and Claude and Perplexi are some attempting to make it the hub and like the source of everything and then you can branch out to the different micro tools that do a great job. Like I mentioned, Slack, notion, etc. So they're not going to

replace like a lot of the different tools that exist, but they might be the knowledge hub or the ecosystem base, right? And this is another step in that direction, I think, when you're making little agents that are able to connect to connect everything, you know, like if I I'm thinking right now, the second I saw this, I was like, I I've been considering when somebody asks a question of me like, hey, what's the status of XYZ topic client work-wise? Um, you know, notion MCP it I think is is a fair response just like, hey, you know, just have the agent kind of sort out when there's things that need to get escalated. Um, and escalated to who. And the nice thing about MCPS is it just makes it a lot easier than having to manually map that stuff out. And this would make it an

integral part of my this would be very that'd be agentic essentially because it would be replacing a a person who's you know important in the fact that they're keeping clients happy and having quick responses. But by basically making an agent ready to respond and all I need is a a human in the loop yes no approval, I'd be putting myself in a way better spot for success than like always relying on a person's ability to respond quickly and their ability to look stuff up accurately. Because honestly, if MCPs continue to be good like this with high levels of reasoning, um being able to find the facts of the situation is going to be a lot better done by AI anyways, I think, rather than people. >> Yeah, I I think I was thinking about another reason why you shouldn't just abandon, you know, NAN or

Jatform AI agents for this. uh just you know the same way how like you know like the Google you know Google forms is not priority like open AI has been released in all these different kinds of uh things like do you remember custom GPDS uh do you remember tasks >> all right uh there just like so many things and none of them they actually like they were big when they released it and we covered them as well >> yeah but I feel like they've kind of >> they going to invest further on them because it's just like it's like kind of like u they want to make news. Uh it's great. They probably have like dedicated teams but they probably don't put like that that much resources in those uh products. Uh maybe a single small dedicated team working on them but they're not advanced

at all. like we don't see any improvements on them over a long time and then they just release this not ne shiny thing and then my worry would be like okay I moved from uh you know NA10 to >> or make Yeah. >> Yeah. Right. Open AAI agent builder and then like they don't improve it over time. >> They don't care about it at all. Yeah. >> Yeah. Because that's that's a definite concern for me. No, that's that's a definite concern for me. I think that's fair. Like I would I would respect Like I use relevance a lot, right? And like my initial thought for this is like well I can save on make.com operations by utilizing a relevance like uh module inside of make. So it just runs the once and that has connection to MCPS and stuff and that to me is more

safe than the big company because I'm like well I know relevance is trying to become yeah >> the premium a AI agent company. Um so it's >> it's a fair point you bring up. It's it's kind of concerning as like a user of pro of this product like what do I what do I know that they're going to care about because the model is the only thing I've seen consistently that they have attempted to improve. >> Yeah. And uh still it's it's pretty cool. Uh yeah, you should definitely check it out. You should play with it. Uh and it's just available for free. We live in such a great age. Uh everything is available for free. Go and play with it. And uh it's it's so much fun. Definitely. I enjoy uh you know playing with it. >> All right. Well, with that being said,

I think we pretty much covered everything. Um do you have anything else to say about it? >> Yeah. Uh it's exciting times uh always with AI. >> Yes, it is. All right. With that being said, thank you so much for being on the show and thank you everyone for listening to this episode or watching it. If you did like it, make sure to leave us a like and subscribe on YouTube. And do not forget to leave us podcast reviews on Apple's podcast and Spotify. We just reached a new weekly high um and monthly high on those platforms. So kudos to everybody for listening, watching, and enjoying us gab on about these different tools and interviewing people that um are really really cool. So thanks for watching and we'll see you in the next one. Bye.