Discover how the combination of AI and no-code automation is not just a trend but a revolutionary leap forward in business efficiency.
This episode shows the complex world of AI and no-code automation, presenting it as an accessible tool for every business. Whether you're a seasoned automation specialist or an entrepreneur looking to streamline operations, the insights shared here will illuminate the path to enhanced productivity and innovation.
In this session, you'll discover:
Reid Robinson, Lead Product Manager for AI at Zapier, joins us to share his expertise and experiences. Reid's journey through AI and automation provides a unique perspective on harnessing technology to elevate business operations.
About Leveraging AI
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Discover how the combination of AI and no-code automation is not just a trend but a revolutionary leap forward in business efficiency.
This episode shows the complex world of AI and no-code automation, presenting it as an accessible tool for every business. Whether you're a seasoned automation specialist or an entrepreneur looking to streamline operations, the insights shared here will illuminate the path to enhanced productivity and innovation.
In this session, you'll discover:
Reid Robinson, Lead Product Manager for AI at Zapier, joins us to share his expertise and experiences. Reid's journey through AI and automation provides a unique perspective on harnessing technology to elevate business operations.
About Leveraging AI
If you’ve enjoyed or benefited from some of the insights of this episode, leave us a five-star review on your favorite podcast platform, and let us know what you learned, found helpful, or liked most about this show!
Hello and welcome to Leveraging AI, the podcast that shares practical, ethical ways to leverage AI to improve efficiency, grow your business, and advance your career. I know I said, I say that every single episode, I'm really excited, but today I'm really excited. And the reason I'm really excited about today's episode is because today we're going Literally, every company in the world right now is trying to do one of the following things, either understand how to build efficiencies with AI or already building efficiencies with AI, and in both cases, they're looking for the most easy and efficient way to do it to do that. But what the hell does that mean? Easy way to deploy efficiencies with AI. So that's a question again, a lot of people are asking. it first of all means that you have to map your existing work processes and at least understand the processes, but it happens to your business. But it also means that you have to somehow connect multiple steps and multiple tools and multiple processes that you have in your company in a way that will be more automated than it is today and leverage AI to do this. Now, the reality is that most companies in the world today are not even doing basic automation. That does not even require AI. And suddenly everybody's chasing about AI automation. Now, the truth is, There are several companies who have been providing no code automation to the world for many years now. The most notable one is Zapier. So if you've been living under a rock and haven't been using Zapier or at least heard about it, then you probably haven't done any kind of automation, in the last, I don't know how many years, but if we could talk to somebody in Zapier about automation in general, and then maybe if you're lucky, get the lead product manager for AI in Zapier, that will be really amazing to look at what's possible today. So I'm glad to tell you, and hence why I'm very excited that Reid Robinson, who is the Lead Product Manager for AI in Zapier is here with us today. And we're going to dive into actual specific use cases that will show you the power of combining No code automation together with AI capabilities and how you can use it in various aspects of the business. I'm personally really excited about this. Reid and I was trying to schedule this several times. We both have crazy, busy schedules and kids. And so it wasn't easy, but we're now here. Reid, welcome to
Reid Robinson:Leveraging AI. Thanks. Excited to be here. I think that was a phenomenal intro and yeah, I've been working at Zapier for a number of years. I actually left to start up my own business and rejoined as I was really excited about everything going on with generative AI and think that for this technology to really get into the hands of a lot of people, Zapier is very well positioned to do that. And we've been seeing tens of thousands of users leveraging AI in their workflows doing millions of tasks through our platform in the last. year and a half. And it's been incredible. And so excited to share some of these cases and hopefully be very, I think I'm excited for today was I'm a, let's say technical tinkerer. and I love teaching people how to do things with platforms. And so excited today to dive in and hopefully inspire some people, show some people and help you actually see how do you do this in not only, a very simple way, but also like, how can you take that to the future state and really start to do things really powerful within your business.
Isar Meitis:Amazing. It's exactly what we want to do. I think the biggest thing about this podcast is we talk a lot about the how, and not just about the what, which is a lot of other podcasts are doing. But before we dive into that, I want to ask you more of a high level question. And then we'll dive into all the different use cases. And it's a question that I can tell you when I teach my courses, I want to tell you what I'm saying, but then I want to hear you, as I mentioned, a lot of people don't even do basic automation, right? Which as as somebody who has worked with Zapier for many years, is Amazing. Like it's magical as far as the stuff that it can do. The question is, when do you need AI, right? When do you cross the gap between, Oh, I can do just basic automation and move the stuff from here to here, and I'll tell you what I say, and then I really want to hear what you say. So what I teach, both the companies that I consult to, or when I teach courses is when some human input is required. And I'll explain in one sentence what does that mean. If no human input is required, if all you need is to move that file from here to here and then save it to that and then change it to this format and then send an email about it, I don't need any human input and I can do it old school Zapier without any AI. If I need a lot of human input, meaning I'm going to rewrite my business strategy, it's probably not the right tool, but if I need some business, some human input, meaning I need to analyze the data and give feedback about it, or I need to summarize data that I have, or I need to invest time in actually writing the email itself that requires some thinking process. I think that's where AI enhances traditional automation, but I really want to hear what you think. From your perspective, from a Zapier point of view, where is, and I'm sure it's not a clear line, but that gray area between, Oh, you just use old school automation to actually go and infuse AI into the process.
Reid Robinson:Yeah, it's a very good question. I would, I like your phrasing. I would say what we've observed from users is that's not always the case. I think there are a lot where the particularly LLMs, one of the nice capabilities of them is the fact that they can produce what feels like something new, right? Like it's not an ML model that is predicting what it's seen in the past. it is able to produce like a net new content or a net new output. And that is very cool. we have seen a lot of users leveraging AI and workflows to help parse structured data, right? And I think that's a very, it's not really human input to side of it. I can talk about a use case I had for, my wife's receipts. she's does a lot of photography, gets a lot of receipts, particularly on the go. And in the past, I had been horrifically taking that pile of receipts at the end of the year, putting them into Google Sheet, and then getting it over to her accountant, which was not fun. And I actually solved part of this through a combination of Glide, which is a no code mobile app builder phenomenal platform for getting things into like sheets. And then on the Zapier side, I actually built out a workflow that took like a new Thing in a Google sheet. I took the image. I use the tool called PDF code to get the text from the image. And then I used the chat to extract structure data capability to pull out all the information from the receipt and even categorize it and put it back. I'm saying all this and realizing you did that is technically human input, but you, I was trying to show more of an extract structure data use case that is when things happen someplace, there might be structured data, you need to move from one place to another, but. The fields might not match very nicely. That's been a really big one. so
Isar Meitis:basically again, you, I don't know if it's human input, but it's understanding data that is not. Yeah. structured and or that is not numbers, right? If it's numbers, so qualitative data, just use Google Sheets, like you don't need anything else. Sorry, quantitative data, but it's qualitative data. It's written text that is not structured in any clear way. Then analyzing it is a huge aspect of this. And I agree with you a hundred percent.
Reid Robinson:Exactly.
Isar Meitis:Awesome. Okay. So the stage is yours. I know you prepared a few cool use cases. Let's just dive into the first one and we'll take it from there. Let's do it. Let's do
Reid Robinson:it. All right. Let me share my screen here. Let's do, let's just
Isar Meitis:do everything. those of you are listening to this as a podcast, A, don't be terrified. We will explain everything that's on the screen. And B, there is a YouTube channel for this as well by Multiply. The link is in every single show notes of this show. So if you are driving, listen, stay with us. It will be fascinating to listen to. If you're at home and you can open a computer and watch this thing, then there's a link in the show notes and you can find
Reid Robinson:it as well. All right. I'll try to be for those listening. I'll do my best. Uh, vision model, understanding, telling you what you're seeing, on the screen here. so yeah, right now we're looking at the, what's called the Zapier editor, which is your environment where you actually build workflows in Zapier. And for those not familiar with Zapier, essentially it is a platform where you are creating automations that take data from point A to point B and C and D and E and so forth. And so one of the earliest use cases I ever played around with, what was then the DaVinci three model from open AI was around helping me add things to my to do list. And I'm not a very neat or organized person to be clear, but one of the things that was driving me always a bit nuts for a while was I had a workflow in Zapier for many years, which was when I, someone messaged me in Slack to do something, I want to make a note to do that. And traditionally that workflow was, I added an emoji to that message that basically said to do, and that would add it to my to do list, which I use tick. nobody asked me why, I don't use to do list or any millions of the other ones. This one just did what I needed it to do. So I, I've continued to use it. From there, the problem is. Again, if you think automation, you're taking some element of the previous thing and putting it somewhere else. And when it comes to a Slack message, there's not a lot of great ways to do that because you have the message text, which is like the actual content of the message, the channel name, who sent you the message, and what do you really do to make a title of a to do list item that gives you any context of what it is you're meant to do. And that's when, Gen AI is a great example of taking information, applying some sort of reasoning to it. And giving you a better output. And in this case, I take my new messages from Slack. I threw in a very simple prompt. I think this prompt hasn't changed in a year and a half, so I'm sure it could be better. but essentially I'm just telling it, I'm adding this following message to my to do list. Give me something less than a hundred characters. And don't use quotation marks. That was a thing the models used to do. It doesn't even do that anymore. but I don't want to risk changing things, so I tell it to give me a title. And it does a great job! Every time! And what's great about this is now in my to do list I have these really nice things. line items of what I'm meant to be doing for this. It's a very simple step, but I think it's a great one that I love starting with just to articulate what is possible and what are you doing about taking some form of text and doing something different with it. and yeah,
Isar Meitis:I think it's an awesome use case. I want to, again, generalize this to show you how incredibly powerful this is. So think about, okay, this is redoing his thing, right? But think about how many people in Zapier or in your company are connected to Slack. If you're a Slack user, if not, it could be teams or whatever other platform you're using to communicate between people in your organization. And in some of these channels, there are. Requests to do something. What amazing capability would be if every single time any person in the organization is asked to do something, it creates a clear, definitive task for him in their task management platform. This could be, Monday JIRA, ClickUp, whatever, doesn't matter, right? Wherever it is, or TickTick, wherever it is, you're managing tasks, that these tasks get created automatically are very clear what they are. It gives you some context of where it came from, and it just. Automatically shows up, meaning you're not going to miss anything you're requested to do, not just you, anybody in the organization is not going to miss anything on the place that you actually track task performance and assignment and so on. And so tools like this. It sounds very simplistic. Think about how many tasks that lost are not done are not clear enough. So it's not clear enough. Now somebody has to go back and ask a question and find the person and he just answers the next day. And then it's not relevant that there's so many problems that this little simple thing solves. And again, in Zapier, it solves it literally in three simple steps. You connect your Slack channel, you connect it to Chachipiti, you connect it to TickTick and you tell it. In between what it needs to do, which in this particular case is just idle. What is it that I need to do based on the message? I think it's amazing.
Reid Robinson:I think so too. And I've actually talked to, I think you gave a good overview. I have talked to it managers, for instance, who use a very similar workflow to get things into Jira or, any other tool, Where they're keeping track. I've done this. I have special emojis. For that, that aligned to different product areas. for a very similar thing of adding things to the backlog and you want some sort of context of what on earth are you adding to the backlog? something like that. So awesome. I'll show quickly the one I was mentioning, earlier, which is on the receipts. Cause I think this is a great one of diving into more structured data. And for this, there, there's some components I can probably skip over, but essentially, when a new receipt is logged, By my wife, which is when she takes a picture of a receipt. what I do here is I upload that file to Google Drive. I then use a tool called PDFCo, which converts a image file into a text, right? So it's extracting the text from a receipt. And then I'm using this fancy step here. It's not actually fancy. Actually, let me switch views here so you can actually see the details of this. Called the extract structure data in ChatGPT action that we have. It's a very fun one because it takes the some amount of text and helps you extract again. But in this case here, what it actually does is you can tell it what values it can extract and you can create many different criteria to do this and actually be very descriptive of what each of these fields should be. So if you see here, I added all of these fields for every single one of these. I have these areas here where I can say, is this a required one or not? Should this be text? Should it be an email? Should it be numbers? some sort of guidance to the model of what this likely could be. even giving it like, if it's a pick list type of thing, you can tell it what options there are. And you'll see, so for date I told it, this is the date the transaction took place, and yes, it's required, and so on and so forth. And probably the most fun one for me in terms of time savings, is even just the categories. So I gave from the accountant list of all of the expense categories that she has, and asked it to pick the best category as a field. And of course I review this in the sheet, I'm not just blindly agreeing to everything it says, but it's phenomenal.
Isar Meitis:I want to pause you just for that, just one second, but the people are not watching this. So the fields that he's talking about could be, okay, who's the merchant? What was the amount? What was the date? what was this for? Address if there is like whatever fields they are usually on receipts, right? So that's one aspect, but the other aspect doesn't actually come from the receipt, it comes from her company or work or policy or whatever, which are the categories of the expenses that you want to drop this into, which could be a million different things. it could be travel, could be food, could be, market, like whatever the receipt was for, and then he will try to understand based on the merchant or what he was or where it is, or what's the time of day in the best to the best of its knowledge to categorize these into different buckets, which again just takes a huge amount of time when you need to do this at scale. Exactly.
Reid Robinson:Even, it takes a huge amount of time, even for my wife who has maybe 200 receipts a year, and it's saving me, what would have been, I don't know, probably, this would take me five minutes per receipt. right? That's a thousand minutes of my time. Yeah, which is very nice. Yeah. so yeah, especially
Isar Meitis:that she's not paying you for that time. I assume I have a feeling. Yeah, I'm
Reid Robinson:not getting the accountant or bookkeeper rate for sure.
Isar Meitis:I want to jump on what you just said showed because it reminded me something that I'm doing. and once I'll tell you what it is, I can guarantee you after you put your kids to bed or kid to bed, you're going to try to create this. So I'm not doing it within ChatGPT itself natively, but I think what you just showed could be amazing. And I apologize. I'm hijacking your examples, but I really think you're going to love it. What are the biggest problems in the business world for decades? Is what do you do after trade shows? So you come back from a trade show, you have 200 business cards and you remember about three of them what did they relate to? And the whole point of going to a trade show is grow the business and have the connection. So you need to know what these 200 business cards are and it, every company and every person have their little trick, like they take pictures. They're right on the back of the card. They have this folder where they staple stuff and write notes. Every there's these digital scanners, like they all have. What I'm doing now, like for the past, I don't know, six months, I'm literally using ChatGPT as is on the phone. And I take a picture of the card and it knows how to quote unquote OCR the card. So you're telling her, okay, I just need the data from the card. It will do it. So I will take a picture of the business card and then I hit submit. And then I click on the record button and I say, listen, I don't need you to do anything with information right now. Just extract the information from this business card. I spoke to Reid,. This is what we talked about. I want you to remember this. I will tell you later on what to do with this. And that's it. And it says, okay, got it. That's the response you're going to get or something a little longer than that. And that's it. But at the end of the day, I have one chat that has 37 business cards. OCRed with what I said I need to do about them. And then I ask you to create a table for me, name, last name, company name, date, all the stuff that's on the business card and what I said. And now I have a CSV file that I can upload to my CRM. And I know everything with about, Six seconds of investment after I get a business card for a job. So you can build stuff like that with exactly the processes that you just showed, right? Because you know how to extract information, how to attach images, you know how to do all these kinds of things. In your case, it's also going to save the step of actually loading into the CRM because you can connect it to the CRM. I hope challenge is accepted and you're going to try to make it happen.
Reid Robinson:Oh, I think you'll appreciate. what we'll get to now, because I can tell you, you can already do that with Zapier. And if you have access to ChatGPT, there's the capability to create a GPT, right? With custom actions. You can actually connect custom actions to run zaps, as all example later. But essentially, if you just had one called log to CRM, or log trade show. Whatever you want to go business card, forgot the word for business card. then that would be as simple as selling it to just every time you upload a picture, log it to your CRM with whatever other data you give it to. And GPTs do work with mobile now with actions, which is really exciting. So you can even use your speech to yeah, exactly. Interesting. So you can do that. I've played around, I actually just dropped a tutorial like two weeks ago on how to run a full zap from a GPT, because it does get fun for what it enables in that sort of environment. Yes. You can definitely do that. I'll pause
Isar Meitis:for just one second. For those of you who don't know what a zap is, this is what GPT is. So zaps are the different steps within Zapier. So every step you do every function, everything you call is a zap. actually no, the whole structure. if you're trying to do a thing, it's a zap that will have multiple steps in it to actually get to do it. And the GPT is like a use case specific instruction based small version of ChatGPT that anybody can create if you have the paid account. So you can create your own mini processes within the open AI platform. But then based on what we told us, there is a way to call Zapier from within the GPT to do things that the GPT does not know how to do. And Zapier does such as connect to your CRM. Exactly. Okay. Next example. I'm awesome.
Reid Robinson:We can dive into that. Actually. I had, I'll show you literally that actually funny enough, I had that one ready to go. So I had one slightly different, but in this instance, I wanted a GPT, which again is a custom version of chatGPT that lives in ChatGPT. And in this instance, I had one as a demo of a customer success manager assistant. So for those following along in their cars, we're looking at ChatGPT right now. We have something called CSM assistant on the screen. And all I'm going to tell it is, this one is trained to help me do account plans. And so I've asked it to help me with an account plan. It wants to know the customer. I'm going to ask it to do a customer called block, which is the, I think formerly known as square. And at this point, you were being asked by the GPT to send some information over to Zapier. And you can see here, we're sending company name lock, as that is what I've told it to do. Now, what's cool is at this point in time, I'll switch screens briefly to show Zapier. What I'm doing is we're getting it to do what's called a new request in GPT actions. Now, this is a Private internal app at the moment. I actually shared a link for this. I can share the link with you as well to include with the audience. so this is not public. Don't go wild building things, with this just yet. Or if you do, let me know so I can say, we need to make this real. And what this does is it allows you to basically do a form for a GPT. And you, in Zapier, you just tell what fields your form should have. And then the GPT attempts to fill in the forms, and then based on what data it gets back, you can do future steps in your workflow. And so in this, I look up the company in Salesforce, I then, based on the company, look up open opportunities, I look up for contacts, I look up activities, and then I send all of that data back to the GPT. And you can see on this screen here, I'm showing I've mapped things like the employee count, the open opportunity names, the description, what stage we're in, who the main contacts are, everything you would normally find by an annoying amount of clicking around in Salesforce. And now I'm back in the GPT. It just told me everything. It told me everything about that account. It told me the open opportunities. It told me who those open opportunities are with. As well as, now, ideas on what I should be doing to strengthen the plan within that account as their CSM. so yeah, this is, you could do the same thing for, upload a, you could create a workflow, for instance. I'll give you a fun one for, and we'll dive into something similar. I have a business card from a trade show. Here are some details that I met. Your form might include Your typical CRM fields, like first name, last name, email, company name, right? all of those different fields. And it might have another one for like notes, right? what additional notes do you have on this contact? Your GPT might be trained to then add that to the CRM. And then in Zapier, you can actually have that step. Not only add to the CRM, you can enrich with tools like clear bit, which would look up other information about the account, such as, what size company they are, you can use other steps that will actually go out and browse their website to pull in the content from their website to then use in another. ChatGPT step in the zap. So it gets really meta here of having multiple chats, UBT things going on and ask it, Hey, based on this website and based on what my company does. What are personalized offerings you think we could provide to this company? Take into account these additional notes about this individual I met with, right? And that all is coming in from that initial business card and the additional notes. And then you output that to the GPT, potentially. You could have it just say, hey, log this into Serum. You could even say, draft an email for me and queue it up so that when I'm back at the end of the day after my trade show, That's I'm just reviewing 25, 30 emails that all now have enriched personalized emails to whoever you just met with.
Isar Meitis:All I can say, those of you can see me, I'm doing like my head exploding kind of a thing with my hands. But it's literally mind blowing what you just said. And again, I want to generalize that because my head is now going 250 miles an hour. But the. I'll try to generalize it without going too crazy in generalization, but the connectivity between a GPT, which is a set of instructions within open AI that really understands data and the ability to connect it to Zapier to go and collect that data from any source that Zapier is connected to, which is any source because it's connected. I don't know the number now, like 6, different kinds of applications, right? That it's connected to. So basically, if you're using it, it's probably already connected to Zapier, at least to one extent or another. you can literally go and start a process within open AI to research something, anything that relates to your business, have Zapier go and collect that additional information from multiple sources and bring it back to the GPT to analyze that information to give you the outcome that you're trying to get, which could be anything. Now, even the example you just gave, which again, is mind blowing connected with the example that I gave. You can do this in real time. You can meet somebody, scan their card, and within about 20 seconds, get a summary about their company, their position, who's their boss, what have they been doing, what kind of money they spent in which areas that relate to your industry, et cetera to your phone. This is, ah, this is insane.
Reid Robinson:Yeah. It starts to get fun. If you have if you could pretend to have a little air pod in your ear, you can use the text to speech stuff and it'll start having it, go into your ear at the same time, be cool, but yeah, it is quite powerful. I'll take this quick moment to say that as much as I love Zapier, and we are a no code platform, I will acknowledge Zapier takes time to learn to become powerful and using AI and automation together and really just using automation together, I will caution, it does take a little bit of getting used to, and probably the most critical thing that you need to start to understand that I think users mostly struggle with, Is the whole concept of data flowing from one point to another and being reused in different places. And in this world of Zapier, you look at, there's what we call a trigger. And this is when something happens. And in this case, the something happening is essentially you're getting information from your GPT session. And here you'll see the four elements that we can use as. Future details, right? Like future information into future steps that expands as you now do a lookup in, like we used one of those elements to look up the company name in Salesforce. And from that we got. Don't want me to test that we got a lot more option and a lot more fields to use, right? all of a sudden we went from that one. Now we have Salesforce data, and now we have a lot of things. And that keeps going as you go further. And you can use, ultimately in step six here, we're using data from steps two, three, four, and five all together. And so it's just, it's a really important concept to think about and understand. And it's probably the most critical one to get. I hope that helps a lot of people beginning their journey in this. I've helped a lot of people, get started with Zapier. And I'd say that, yeah, the, as you become proficient, it's one of those types of tools. And it's one of the reasons I've actually left Zapier and then rejoined. yeah. That when you start to become proficient with it, it is extremely powerful for your business.
Isar Meitis:I, I want to add something to what you said, because I think it's very important. I appreciate you saying this. And I want to ask you a very specific technical question. So the thing I want to add is people like, okay, this sounds really complicated, even the people watching it as a screen or never seen this before, like all these fields and all these parameters and okay. Let's say you need to invest, I'll go really extreme, a week to learn how to use this. And it's not like in a couple of hours, you'll get the idea. And in another day of playing with this, you'll be good. Maybe not as good as read, but you'll be good. let's, but let's say it's a week. Let's say it's a week. If you do what we just said, that week is going to save people in your business months. Of work within this coming year. So the ROI of learning these tools. And again, I don't want to, I don't want to say anything bad about Zapier. I think it's an amazing platform, but there's other platforms like it, right? It doesn't have to be Zapier. there's make there's anything, there's like multiple tools that do similar stuff. Zapier is definitely the most user friendly of them all. Still going back to what Reid said, there's a learning curve. Like you have to learn how to use these tools, but the ROI is Literally insane. Like it's going to save you full positions of people. You don't have to fire these people. They can just do other stuff. They don't have to now go and figure out copy, paste stuff from the CRM to write a report, to give to the salesperson, to talk to the customer. Like it's just all that tedious work goes away. At scale. Like it doesn't matter how many people want to use it from that moment on it. We'll just do it again and again. by the way, my joke always when people, when we start talking about Zapier and say, wow, what if this breaks? I'm like, okay, Zapier breaks. There's a big issue. It's like the internet breaks. there's so many companies, especially in the marketing world that are completely dependent on Zapier working that God forbid, if this thing ever breaks, we have serious problems. So I think that's, there's not a lot of companies. I can say that with confidence that I think it will run well. I think Zapier is one of them because they've proven to do this for many years now without any major Catastrophes happening. The questions that I want to ask is how is it actually connected? what did you add on the GPT side in the customer instructions of the GPT to connect the two together?
Reid Robinson:Good question. Now I'll take the time to break that down for folks listening. when you create a GPT in ChatGPT, you are essentially creating, it's like different version of ChatGPT. And within that you have various components, right? There is the GPT. In this case, I've done CSM assistant, there's the instructions, which are the, instructions, you can upload files as knowledge. You can give it access to OpenAI's pre baked tools, which are like web browsing, Dall-e image generation, and code interpreter, aka for some folks, data analysis. and then there's this funky concept called actions. now I will throw one fun story in here. For those who follow along, and for anybody who doesn't, OpenAI's Dev Day, which is when they announced the launch of GPTs, which was this big fanfare with Sam Altman on stage doing, and they did one demo of a GPT. And that one demo of a GPT was one running Zapier in a custom action. And I got to work with the team on that and be actually in the first time I've done a partnership launch where I was actually in the audience physically live as something that's being demoed. And that was a live demo of a very new product for both sides, which was extremely exciting and extremely nerve wracking. I cannot express the sense of relief when that thing did its job in front of, I think it was about 70, 000 people watching live. I was one of them. There you go. and it's funny. It was post people didn't realize that was live. I've had a lot of conversation. They're like, why did they do that live demo? And I was like, that's what they do. And, but anyway, suffice to say, I love, I have a great admiration for what GPTs do. Yeah. And so in this. This is the way you are giving the GPT access to everything else in the world, right? Like the three capabilities they give is more like you can upload files that they can reference, browsing, Dall-e, code. And actions. are essentially your way of giving it anything else. Now, as I mentioned, this is a private one. We do have something for AI actions. You can find information on that on Zapier's website. But this one in particular is a new private app that we're playing around with called GPT Actions, at least for now. And essentially, we've tried to make it as simple as possible. Here, as you're seeing, we made it like, almost a form fill is what we had in mind. So you're describing the field names. And then all you need to do is take, if you go to the test step, you'll see this kind of thing called a import URL. I will be clear. I do not love that. This experience is very developer language y for cheaply creation, but it is what it is for this point in time. it's probably my number one request. I literally just met with a bunch of the product managers from chatgpt to again, advocate. Please, for the love, make this easier for less technical users. Yeah. so soon hopefully. But you'll see on this area here, there's an import from URL button. You paste that URL and that is really all you need to do. That's amazing.
Isar Meitis:Okay. So I want to explain something to those who are watching and those who are not watching. The actions within the GPTs. And I know we're getting way more technical than I wanted in this particular podcast, but the actions in GPT allows you to connect as Reid said, to any application out there. And the way he does it is run API. Those of you don't know what an API, it's a language for one software to talk to another software. That's basically what it is. And the way you explain to any machine, how an API works is what's called a schema, which is a recipe. It basically tells us, this is how you talk to this other person. And instead of knowing how to do that, all you have to do here is paste a URL, which I assume then contains the rest of the stuff. And then, so you really don't need to know any programming. You don't even need to know what the acronyms of API means, and you definitely do not know how to write a schema. All you have to know is to copy the URL from one place, place it in the GP. Copy the URL from Zapier, placed it in the GPT, click import, and then the two are connected. So it's not as hard or scary or as technical as it sounds.
Reid Robinson:Yes. That's true. And I hope it gets a lot easier. we've actually played around internally with a GPT to help people create GPTs. It's really wild. That's mad.
Isar Meitis:it's like they may put a hole in the universe, the time space continuum may collapse, but okay.
Reid Robinson:All right. So the other Probably one of the other use cases I want to touch on, and I know we have some other topics we want to get to, but one of the cool things too is, especially while we're on the topic of GPTs, at the same day that OpenAI announced GPTs, they also announced something cool called the Assistance API. Now the Assistance API is essentially the the fraternal twin of GPTs. and people are like, are you twins? And then they're like, you could tell they're not identical twins type of situation here. Because while GPTs allow you to create custom versions of chat, GPT with tools and capabilities and whatnot, they live in ChatGPT. You cannot use a GPT outside. Of ChatGPT, whereas with the assistance API, you can create an assistant. that has tools, that has documents, that has instructions, and now you can use it via an API anywhere, including Zapier. Now before I, I'll, I'm going to dive into a use case that I think is a really cool one because it's essentially, you get a new email in Gmail. Then you ask your prebuilt ChatGPT assistant to, based on your instructions, which I'll dive into, to give you a response. And you're then going to do a draft reply in the same thread in Gmail. And I want to show quickly how you actually make one of these assistants. So you're in the OpenAI Playground. Now, you, you can build this in Zappy, I'll be clear, but I will tell you, don't do it. It is better to build your assistant on the OpenAI Playground. It is far simpler and you can literally pick it from a dropdown list in Zapier. So you'll save yourself a lot of hassle by just doing it in the OpenAI Playground. If you don't know what the OpenAI Playground is, it is essentially their environment for playing with things built on their APIs, which the assistant. API is. And so when you, if you don't know more, like literally just Google open AI playground, you'll get there. And once you're there, I actually think it defaults to the assistance tab at the moment. If not click the assistance tab, you'll be asked to create an assistant and just like GPTs, they have names, they have instructions. They have, you can give them code interpreter and files for retrieval, but then they start to get a bit funky. They don't have Dall-e. Most people don't care. They also don't have, web browsing. Browsing. That one kind of sucks. But you can choose your model, and this is because you're paying for API usage at this point. And give it the model. Now, what's cool about this is for businesses, you can start to give it knowledge documentation you have. So if you want to build a support bot, if you want to build an accounting policy bot, if you want to build something that's going to live in a workflow, you can now do that very easily. And if you give it access to these tools, I'll break this down. Code Interpreter, again, more like data analysis. It does really well with CSVs. Retrieval is the ability to read and understand what's in a document, like PDF, docs, that type of thing. So you do want to choose which one or both that you want to give them and attach your files. Alright, so we've done that. In this case, I built one for replying to emails as An insurance sales agent. And I even told it, you know, write it, but essentially try to ask for a meeting and here's my Calendly link to include. Then, in Zapier, you'll see I've now selected the conversation with assistant action. And now here's the fun part. If you start to build workflows where you want to give an assistant new files and have it reply to information based on those new files, let's say you did one for, reading business cards and extracting the data. You could do it here, right? You would just include the file of the business card and do that there. and then when I run this, you'll see, it wrote up a nice little email saying, dear Alex and Emma, congratulations on the arrival of your daughter, Lily, right? Blah, blah, blah. And then it ultimately asks for a meeting. Good job. But fun part comes when you start to get this back to Gmail, right? I don't want this to send. Right away. I want to have it give me an opportunity to review what it said. Make sure that's accurate, right? you, if if you trust it, you can do it right away. So I just ran this and now when I go to Gmail, I can see in my, I'll refresh the screen here. I have the initial email and then below that, which is populated, is the actual email. That chat TPT wrote in that it personalized based on information. I gave it about the fake life insurance business I have, as well as the real Calendly link to book time with me. And you can see, there are a couple of things that I probably would want to actually edit before hitting send. but really cool ways of doing that in a nice workflow there.
Isar Meitis:So I want to touch on a few things. First of all, I want to simplify again for the people listening or even the people following the video, three very simple steps, right? Step one, you get an email to your inbox. Step two, you send that email through a tool that, again, we discussed the details of it to a ChatGPT to analyze what's in the email and what should probably be a good response to that email, at least initial response. And then step three is it literally puts it as a draft in your email platform. In this case, Gmail for you to review and then decide if it's good or not. If not make small changes and hit submit. The whole process takes about, I don't know, five seconds, where if you as a person had to do this, it would have taken you probably five minutes. And if it's somebody that you have to do this 30 times a day, it 30 times five minutes, which is now an hour and a half that you just saved. And so all you have to do is now review this, make sure that it makes sense and hit send. But now I'm going to go real crazy with this and combine this with the previous conversation we've had. Let's say that instead of just looking at the email as the resource of information, you actually go and query your CRM. So you're getting an email from a client. You go to query the CRM and you say, Oh, what were the issues? What were the previous communications? What did we have on them on Slack? what's the, get more context to the large language model to be able to draft an even better email. What's going to happen. And you're going to get. Less work because the email is going to be more accurate to the actual situation. And so going back to what I said in a very beginning, and the conversation we had of where AI brings value to just old school automation is exactly this. And I think the examples that you showed are literally mind blowing when you start to connect the dots of what they can do. Again, even just the very last thing that we said, anyone comes in, go check the CRM, Go check Slack. See what else was there, connect all the dots together, now write a respond, put it in draft. So all I have to say is review it, make a few small changes, hit send. That process is not five minutes. That process is an hour because you have to go to the CRM and you have to go and find all the stuff that happened before. It may not have been you, may have been other agents or people in your company dealing with that company. And then Slack is like just that work. And then analyzing all the information you found to write the draft email is a lot of work, but it's a work that a lot of people have to do every single day. And now this can give you, and you can even do a step in between that say, okay, on top of the email in bold letters, tell me why you wrote the email. what were the inputs that you found? So now you don't have to do the research because you'll understand why the AI decided to write the email the way he decided to write it. I think it's incredible.
Reid Robinson:I think so too. And I'll say, yeah, it's fun. There's a lot of fun. Then how to elegate when you start to. Get this far. I have a lot of fun. I have a lot of fun in the final, like tinkering stages of it as well. You'll see for those really paying attention, what's on the screen here. the copy is verbose. really give these things examples to be better. there's, it's, there's a lot of fun in not just getting it to that good stage of the workflow. I have a lot of fun tinkering with this and getting it to that final stage,
Isar Meitis:Let's do a quick recap of that and then talk a little bit about Central. And then I think, we'll be done because I think people have a lot to process from this conversation. I know I do, and I actually know this stuff. So if my head is exploding, I'm sure people listening is even harder to follow. So few sentences about what Central is and then we'll be
Reid Robinson:good. Yeah, absolutely. pull up the actual screen. I'm going to realize I'm on the old internal version. So central is a brand new experiment, like preview, I would say preview product that we just released. And it is essentially our first take at a more like AI at the core Zapier experience. And if you try central today, it's free to try. So yeah, go ahead, go ham, try it as much as you want. Really. is quite cool. And in it, you create what we're now calling in this environment, behaviors. so it's a more friendly term of saying zaps in the past world. but what's cool about this is you can do this in all, a chat experience. and there's two elements that really live here. There's behaviors, which are your, We've been talking about like triggers and action style stuff. And then there's data sources. So which is a new concept that we've introduced here, which is essentially a way to get knowledge from a third party application into the experience of a bot. So if you want to add details from your Google docs, Google sheets, air table, or notion. You can do so today. If you additional apps, you see in there, let us know. And for behaviors, once again, fun is it's very chat, natural language based. So I could take the other example I had, which is, when I get new emails for life insurance inquiries, right up a draft reply, People know about our services and I'm like, probably say something about your services and maybe spell a word inquiries. would help. And you'll see, it really starts to try to help you here, which is cool. and you'll see before I did, in fact, do new email matching search. And it's going to walk you through kind of everything that you would want to be able to do here. So in this one, I might just do, I'll just give it what I had for, which was subject life insurance. I added that as a trigger. And what's fascinating here is you're adding all these actions without really any of the mapping that you previously did in Zapier. So when I choose this thing to do, we now have this concept called have AI select a value. And what's fun about this is you don't have to think as much. You can have it actually determine what, who it should go to, what the body of that email should be, what your signature should be, or you can say, you know what, I actually want to tell you what to do, whether it be multiple choices to choose from or set a specific choice. it's a lot of fun things that you can start to play around with here in just getting a lot more ease of use in creating a workflow and also in particular creating workflows that have AI running at the core of every single time. So they're no longer in this instance, like what we would call deterministic, meaning they're going to do the same thing every time it now has a model running behind the scenes, helping you figure out how all of that should be connected together.
Isar Meitis:Phenomenal. So again, in quick summary, it's a much more. user friendly version of Zapier because you don't need to understand all the terminology and steps and triggers and all of that, because it basically understand from a chat, what you're trying to do. And at the same time, it's more powerful because it knows how to analyze what's actually happening and like maneuver as needed because it understands that the steps and the information and so on.
Reid Robinson:So yeah, fun slide. Yeah. Recommend people check it out. again, it's in free preview right now, so you can't go wrong. you're not going to waste your billing limits on it, to play around.
Isar Meitis:Awesome. Central. zapier. com for those of you who are looking for that. Reid, this has been fantastic. this has really been everything I expected and beyond. I really appreciate you taking the time. I think what you guys are doing is amazing. as I mentioned in the beginning, it wasn't just to hype you. I did plan to hype you. But in addition to that, I really think Zapier is amazingly well positioned to lead this charge on AI based automation, just because you already have the user base and the community and the connectivity do whatever X number of applications. And I think. This is a very exciting time. And those of you who are looking to learn more, just follow the right people. Just, this podcast is a good start, but there's many people doing these really cool, Applications and like Reid said, tinkering and finding out really cool use cases for that. Just find those people on YouTube or LinkedIn or tech talk on wherever it is that you're following people. And you'll find amazing examples that are literally everyday tasks that happen in every single company in any size that can save any hours and hours of work to your team or yourself. So check it out. And again, Reid, thank you so much. This was
Reid Robinson:absolutely awesome. Awesome. Thanks for having me.