How To Create AI Agents That Do Your Job with Zapier
Let's build with Zapier and AI (100+ videos)📅 2024-03-11
Transcript
[00:00] in today's video you're going to create your new assistant don't worry this new assistant doesn't require a W2 or employee Insurance we're going to learn how to leverage zap your central with creating behaviors and adding data sources so we can start actually understanding how this can apply to our business and the labor associated with our business therefore in order to do so correctly we're going to have to use an AI provider so we're going to use open Ai and its assistance feature we're going to create a data set that is our business profile so we can get better and more effective answers and to Showcase his power we're going to go ahead and automatically respond to this email let's jump into today's video to
[00:30] see the rest of the video click the link down below it's a $50 pay no I'm just joking so in this video this is going to be probably one of the more powerful videos I do on this channel especially since basically no one's doing any videos on zap your central it almost like I don't think any even know maybe a lot of people don't know about it right now but we're going to go ahead and I'm just going to show you a really cool way so that when you start actually leveraging this feature it gives you more effective outputs because one thing as we know when dealing with J gbt any AI model is a huge thing that happens is that it's not it doesn't understand context in the sense of like why you're
[01:02] using it in the specific type of output you're looking for as we know custom instructions can help with Chad gbt there's different ways we can go about it so that the outputs are more effective in this video we're going to learn how to get more effective outputs of Zap your Central and more specifically how this would apply to our automations this is my Twitter description down below let's jump in today's video right off the bat go ahead and check out that video right there if you don't even have an idea what zamb your central is it's going to show you how to create behaviors add data sources everything that this can do up to this point then come back to this video on this video though we're going to just go over how to make these things even more
[01:33] effective so we got our assistant your new assistant maybe it's Jim maybe it's Mary whatever it may be we're going to go ahead and train this new assistant with a data source data source is going to be very simple create a Google doc Microsoft doc and let's proceed the way I want you to think of this Google doc is it's going to be basically the context that we provide these Bots we get more effective outputs so provide all relevant information that your assistant would know in the cont context that this was an employee such as company name location maybe stuff like
[02:04] founded CEO industry isn't as relevant but for me I'm just going to leave it there mission statement services offered this is probably pretty important so for example we got solar panel installation Energy Efficiency Consulting sustainable urban planning Corbin do you run this company no I do not this is all a chat gbt output I also want to notate that the actual formatting of this document is irrelevant for I mean it's relevant if you want to see like it look cooler or more pretty so if I bold this for my eyes to make it like more clear to me
[02:34] you can do that but when a AI model reads this they don't care they're just reading the data right so let's keep on coming down here if this is relevant for your use case you would provide key clients maybe an award probably super important would be contact information such as your phone number email website now we can kind of get into more of the proctoring of how we want to structure our responses depending on the context so maybe your assistant is handling email maybe your assistant handling Excel sheets this is where you're going to give the kind of the structuring of how you want to input data how you want
[03:06] to respond to emails you know everything above the board here is our email response guidelines to ensure consistency and professionalism in our communication follow these guidelines when responding emails promptness probably not too relevant honestly so we can delete that tone relevant Clarity relevant accuracy relevant escalation possibly relevant probably not too much in this context so you know tone Clarity accuracy maybe you want to add another part here that says you know reference this email at
[03:37] the end and then provide the email whatever it may be leave it at that so this is like what we're really looking for here is that you are operating a business so you know how you operate with clients you know from past data and past conversations past experiences what typically gets the better responses and if you don't then keep it General AB test proceed in that manner then we got frequently asked questions which can be relevant for the context of your business so such as how do I get a quote for a solar panel installation what
[04:08] warranty do you offer on your renewable energy sources can your Energy Solutions be implemented in existing buildings you know stuff that comes up very very often within your business once you kind of created a docet like this and it has a sufficient amount of data this is aerial at 11 font I would say the max amount of data you probably want to provide a bot in this context is between 5 to 10 pages Reed let's go ahead and then export this as a PDF so I'm going to come up here to file download PDF so we have the information about our business Corbin
[04:39] let's just jump over here and just start going no no no we got to actually set it up so that we can reference this information in the context of artificial intelligence so we're going to use open aai you actually probably already know what open a is if you have no clue about zappier and open Ai and how they work together you can check out the playlist at the end here the first two videos are like combined or over an hour and 10 minutes so you can get everything you need to know about this industry but hit create also side note I've done a ton of videos when it comes to assistance API in the specific context of zapier no code friendly solutions that sounded like a slogan
[05:09] let's get the name here I'm going to go ahead and name it the fake business here instructions are going to be relevant but let's just come down here real quick to the files this is what we care about so we're going to upload our data the data we're uploading is that PDF we just created together in Google doc couldn't save don't worry uh we're going to go ahead and add some instructions here we're going to say follow the and then I like to usually just give like copy this and I'll add to quotation marks for context on how to respond we'll leave
[05:39] you at that for now now depending on your use case this is going to give you discretion on what model to use if you're building out an API honestly I would try both depending on how effective the outputs and inputs are just as a quick shorthand 3.5 is cheaper than gbt 4 gbt 4 is for more the way I like to phrase it is basically use gbt 4 outputs if a human's going to look at it if a human's not going to look at it and it's more of a data manipulation data input output you'd use 3.5 for example I can see a context here where you would fine-tune a API assistant like this for
[06:12] Excel sheets and data inputting and outputting 3.5 could probably work in that context but this matter we're going to go ahead and just jump up to gbt 4 Turbo keep in mind as well in the industry with apis and accessing AI providers there is a downtrend when it comes to cost other than like Cloud 3 that is more expensive than gbt 4 but for the most part a downtrend in November of last year when it came to accessing an endpoint of gbt 4 it's actually 66% cheaper now super cool let's go ahead and hit save side note in
[06:42] order to access files we do need to have code interpreter and retrieval on okay all right so we have our assistant here reason we use this assistant is for the fact that now we can reference this assistant in zapier coming over to zapier here let's create our Behavior we're going to hit create behavior and here we go let's do an event and this could be any context but I usually like to do the event of basically we can command it I think that's probably the most powerful point of this feature and we're going to say the command is
[07:14] respond email I think that's good add trigger so anytime I hit or anytime I type in respond email this is going to run this following flow so we're going to add an action we're going to do an action of Gmail we're going to say find email for reference this is the example data point coming in here this dude I don't know who this guy is but uh he looks pretty cool I like that sweater maybe we could be friends is basically asking how do I get a quote on solar pan panel installation I'm going over here let going and proceed so the search thing
[07:45] we're going to do is typically when it comes to receiving inquiries over email most of the times you'll have a piece of fix text in there so are you receiving from Yelp Google Business whatever it may be you'll have a piece of fixed text in the subject l line therefore making this entire process a lot easier when it comes to making sure that we only find specific emails that are inquiry emails knowing this I'm going to reference this tutorial right here this tutorial right here is going to show you how to basically grab emails that maybe aren't perfectly structured that maybe come in
[08:16] like this and you can go ahead and learn how to use AI in order to identify between a you know MailChimp subscription and a actual lead email so use that tutorial right there for the purposes of this tutorial I'm just going to show you how to respond and you know use the chat bot so I'm going to just go ahead and do this for right now reference that tutorial up there if you want to see how to use AI to identify or alternatively use filters add action and the next action here is going to be a Chad gbt block so do chat gbt we'll do
[08:47] conversation or we'll do conversation with assistant that's the whole point of what we just did there so we'll do conversation with assistant we're going to go ahead and have it generate the message the assistant is going to be a specific field here we you might have to give the exact ID here unless it shows up does show up we're going to click our assistant so we're going to have to have ai generate the value for this field and everything else we're going to have ai do except the model should be very specific we should have that up to gbt 4 now here's the first thing that you'll realize not Vision we don't need this to
[09:19] see anything here's the first thing you'll realize when leveraging Central is it doesn't necessarily allow you to reference data points and chain it that is where this comes into play so this can this can get a little confusing but you have to make sure your instructions are very specific on the way it handles this kind of information that's why we're basically leaving the message as have ai generate the field but it has context from the assistant and the instructions we set up within the assistant's API okay let's go ahead and add an action here we're going to say send email or create a draft send email
[09:51] will automatically send the email to the receive to the basically the person that sent the email to us purposes this video so we can kind of look at it we're going to say create a draft and then add action all right so this is where the fun comes into play this may take a couple more times to go can I update this name I cannot update this name we're going to go ahead and begin here when I say uh apostrophe or quotation marks find the email and understand the type subject line and body once you understand both
[10:26] use we say conversation we'll just type it out conversation with assistant action to write to generate a subject line and body then once generated output place in the create draft action two things first thing this is my first time actually proctoring in this context so we'll see if this even works second thing I may
[10:57] actually have to add more information to the assistant API to make it more specialized for email responsing so it can't it maybe not be able to be a broad version of the data let's find out though let's test test this Behavior what we're expecting here is we're expecting a draft to show up that responds to this inquiry and let's see what happens I want to also point out as well if this works this is actually going to be pretty crazy because alternatively this would required a formatter block uh in the context of an automation it would have required a lot more blocks a lot
[11:28] more work in order to get this kind of output so we got the conversation with the assistants complete let's see if it can actually set up a draft here and we can gut check it with our data source to make sure it's actually the correct answer okay I'm going to be genuinely impressed if this works let's see reload this got a draft to get a quote for the solar pan installation fromation we have two options contact them through the website call their customer service line let's see if that's what we put into our draft
[12:00] or data scrolling down here over here y'all okay initial inquiry this the first time I'm doing this y'all all right this is like so we got how do I get a quote for solar panel installation how do I get a quote for solar panal installation you can request it quote by contacting through contacting us through their website or calling our customer service line draft contact through the website call customer service line okay okay okay it worked now to be honest with
[12:31] y'all maybe we need to add a little bit more information like hey when responding to inquiries provide the customer service line like the number or provide the website but holy smokes y'all this actually just worked okay this worked it was able to reference the data yeah no this worked y'all so now that this works which is like extremely cool if I want to go go ahead and just refresh a chat here and and if you referenced that other video I referenced earlier now what you can do
[13:02] is we can set up logic where okay first off I want to point out something as well which is kind of insane so you know how I said earlier that you would need a formatter or block like if you're familiar with this channel you know what I'm talking about you would need a formatter block to split the data points between the subject line and the body and how to like structure everything with this conversational with Central what they're doing is they're adding like a when I reference a Chad gbt block and conversation of assistant they're adding another layer of artificial intelligence to like basically smooth it over to make sure
[13:33] that everything comes out more correct in a way like this is big y'all if you have little to no knowledge of how to talk to these models this seems like a very easy way to start doing this okay right bot command let's go ahead whatever we can give it bot command the way we references it though we're going to respond email so I may need to do more videos on this because for example what I could do is I'm starting my workday I have 10 unread emails I'm like you I don't want to respond to any of that already lasered in through the API assistant who I am purpose everything
[14:05] done some work on it all I would need to do now and I could do a loop function to go through the emails is simply go to my central zampier click respond to email and then we can set up a flow where it's checking every single email has context and then responds to every single email with drafts or automatically okay yeah yeah this is it it seems like they'll as well you may need to actually Proctor and say do AC respond email and then it will keep
[14:36] going of everything we just set up that's impressive y'all uh yeah this is this is huge this basically I mean imagine two years from now this if we can do this now I mean we could basically what I just described that workflow we could do as well which would be actually pretty crazy good let me know if you want to see more videos like this make sure you leave a like it let me know on this channel if you want to see more stuff like this it's completely free I'm going to leave a playlist at the end here showing everything we can do with zapier I've kind of like moved it around more some more recent videos are on the top you get all the context of how to start leveraging artificial
[15:06] intelligence and automations in your business but today's video we just created your new assistant and with basically probably a four or five hours of just going at it you could probably create something very effective for your business this is also free as of now I have no clue how they're going to charge in the future I'll see you in the next video that's a playist I was referring to when it comes to zapier and artificial intelligence and how to leverage it that's a random video that's my face make sure to click subscribe click notification Bell leave a comment all that good stuff bye-bye