I can’t believe we built multiple AI agents in 100 min CrewAI, Zapier, Cursor AI

all right we got Joe MOA on the Pod co-founder CEO of crew AI uh probably one of the leading if not the leading uh AI agent platforms what are we learning today Joe well first of all thank you so much for having me uh I think it's going to be a lot of fun so we're going to be building a few different things I was thinking about your community and I was like what if if we use AI agents to do things like data enrichment so given that you're t respect a lead it finds information

about them automatically writes an email sent to them but then we take it like it up a notch and we kind of like produce a real time report and kind of like generate a PDF I don't know just trying to have some fun and if we get the time we can do more fancy stuff towards the end so I hopefully hopefully have some spare time at the [Music] end and okay that sounds good and I think could you explain you know explain by the end of this episode what are the tangible skills people are going

to get yes so by the end of this you all should be able to BU Cru uh fully understand what AI agents are how to orchestrate them using crew Ai and even deploy them into a production environment if you want to use or free tier on the Enterprise and you're going to be able to do it there as well so you'll be able to go now from this and start building AI agents for yourself and how to make processes and to be clear there's like I'm not sponsored by Joe at all like my incentive is

to get you people to build agents and make them work for you because I know a lot of you are Solo entrepreneurs solo preneurs have a couple of employees or even you have a business that has employees but you're looking to have a army of Agents do more work for you that's why I brought Joe in um so without further Ado maybe let's share our screen and and and get to work I love that and it's good that you point that out because honestly with crew itself being like in startup now we have a bigger

team like there are more people to help out in so many fronts but from the GGO being able to use AI agents to scale an AI agent company was kind of like an interesting kind like Inception like experience so I I definitely can appreciate for folks out there trying can like build businesses how the idea of having an agents automating certain processes can be quite amazing actually um right all right let me share my screen let's get things going there we go all right so I T about we start with a simple use case and

then we grow up into more complex ones so let's Dy out what we're going to be trying to do this one I think this going to be an interesting one so in the past especially when you're thinking about using cases like about companies in general then start there's two things that you really want to nail once that you have a product or you're building a product you want to make sure that you do good marketing and you do good sales right because that's not only like Revenue but also an opportunity for you to learn so

you can learn what works what doesn't work like what tracks with your product and what doesn't so I thought why don't we do something that is able to do lead enrichment so let's say you got a lead you want to learn more about this person and this company and there's a few questions that you're trying to answer here once like is this my ICP my ideal customer profile also is this company a good match and like should I actually spend some time pursuing interviewing and trying to understand more about them but then that information alone

like you you can do only so much with it but could we also use agents to take one step further and make it actionable and say like all right can we have ideas for how to use our product given that we now learn about this person so the agents will not only do the enrichment and learn about the company and learn about the person they're going to come up with these ideas on how they can use our product and then afterwards can we just go ahead and draft an email and kind of like trying to

reach out to them like hey thank you for signing up um this is how you could be using your product three ideas for business business very excited about it like let's chat so honestly I think this kind of like a more straightforward kind of use case we can start building with no code but I want to take a step further after this and even further if we get the time but before we get there let's get to building so for this one because I want to start very simple I'm going to start with no code

if you go on crew AI Enterprise there's a free tier that you can create an account there's some setting up that you need to do but once you have done it you got to have access to crew Studio crew studio is basically a way for you to create AI agents by chatting like you can chat your way into an automation so for this one let's try let's try to create one ourselves now I have something in here for us to use and I'm going to read it out loud with you all so the idea is

I want to create a crew that will be able to research both a person and a company based on their name email domain of the company I wanted to make sure to learn most that I can about their business so their role their seniority the company level how big the company is what is their industry their culture and three ideas about how they can use AI agents and at end I want to produce an email that is friendly it doesn't sound like an AI offers to explain how they cuse agents on the business call and

basically describing a little more about how I want that email to sound like all right so now I have a plan right the outputs the fully crafted email ready to send including the insights there's going to be three inputs for this agents the name uh the email and the domain it's suggesting to use three tools a serer da um scrape website and GT3 let's remove this so we don't need we don't need gpt3 uh then three agents one for doing the researching one for do do analysis another one for doing the Mao Drafting and four

tasks so what you see here is that with AI agents and crew AI specifically you don't need to have one agent to one task it's not necessarily one to one so you can see the first one is doing by the research agent that is basically researching the thing the person the second one is is researching the company and it's also done by the research agent and then the third one is where the analysis agent comes in and then the emo agent comes in after that so this feels good let's create our crew plan quick break

in the Pod to tell you a little bit about startup Empire so startup Empire is my private membership where it's a bunch of people like me like you who want to build out their startup ideas now they're looking for content to help accelerate that they're looking for potential co-founders they're looking for uh tutorials from people like me to come in and tell them how do you do email marketing how do you build an audience how do you go viral on Twitter all these different things that's exactly what startup Empire is and it's for people who

want to start a startup but are looking for ideas or it's for people who have a startup but just they're not seeing the traction uh that they need so you can check out the link to Startup empire.co in the description and that's going to basically take this UI uh up a notch so we can actually see visually your tasks and agents so here you can see profile researcher IDE generation email content specialist and I can keep changing in here if I want to so let's say that I want to make this like a senior email

accountant specialist there you go um and let's check like how they're doing condu your research to together information about individual domain this that's their company domain all right this this feels good let's do it generate crew so this actually generates the code for you if you want to download it or allows you to deploy this right away if you really if you really want to do that so in here you can see it's very straightforward we have three tasks the first one second one third one we can change the tasks can change the agents uh

one thing that I do want to do is for example the research profile was using serer and scraping the generating ideas is I I think I want to also it to use ser and scraping just in case it wants to go a little like more broad on searching for ideas online and the third task uses no tools now the other thing that I want to do is for the idea generator let's do GPT 40 instead of 40 mini and for the email let's go a little crazy Let's Lose on ow and mini uh so that

we can get like some extra Fen in there so uh let's see if it's going to work honestly ow and mini in my experience for gentic behaviors sometimes falls short um just because it already has like chain off tot and B it on it it tries to basically like figure out everything itself without using any tools and kind of like outsmart a little bit what you're trying to do but hey there we go so we have our initial crew in here very simple what I'm going to do is I'm going to deploy this right away

so that we can go ahead and test it but we're going to also download the code so let me go back into that so that we can um so that we can start trying things out all right there we go this is do one yes so let's do download The Cone so that we can uh change a few things about this make it more complex go extra custom again you don't necessarily need to use this UI in the end of the day if you want to create this Cruise locally using only the open source you

can do crei create and name of your crew and that will basically ask ask you a few questions around like all right what is the provider oh sorry create crew there you go it's going to ask you a few questions about like what is our provider so you can say ah open AI what the model you want to use you can say well I want all my agents to use gp4 o Mei you can specify individual models per agent down the road then Jo I had a question I had a question about the models you

know I think a lot of people get to that point and they're like yeah I don't know which model to pick for this particular task how should how should people think about that that's a great question Greg so usually the way that I think about it is like smaller models like smaller or 7B 14b they they don't work as well you can get them to behave as agents but they usually take even longer because they go in so many blind alleys and keep like retrying things so you definitely want to go like my goto is

GPT 40 meaning and that usually works great that said you can evaluate different models locally we have a feature let me go into this new crew that we created name of your crew we have a feature called crew AI test and what that does is it allows you to test our agents using many different models so you can test like all right how does it does with GPT 4 how it does with GPT mini so you can get like how each model works and what are the score that each task gets so you get a

an actual quality score an actual hallucination score yes of course it's missing my uh my key let me added this in there make sure that we can get this all right I'm going to just come in here I get this nvar I have a key that I created just for us today and let me copy it there we go open it I API key and I'm going to put ours in here all right so what do you can do in the open source is you can run this command the create test and what it does

is basically run your agents but it's going to try to calculate um the task let me do just one interation by the way um oh I think maybe my key is wrong all right let me get another key one second classic classic yeah all right there we go as you're uh as you're fig putting that in I have a question around one of the things that I I noticed you changed was adding I think it was like a senior writing correspondent or whatever you added the word senior how how important is like does that actually

affect the outputs of your crew if you if you write senior to some extent yes I mean not necessarily senior alone right but because these agents they're kind of like impersonating roles you do get different behaviors depending on you do get different behaviors depending on how you give them like a different role so you can actually replicate this on chat GPT like if you ask jgpt to like give you an assessment on uh a stock it's going to give you kind of like a report but if you say uh behave as an FCC kind of

like a proven kind of like a St analyst it's going to give you a completely different answer even though all the remaining of the prompt is you the same so um it's one way that you can steer the model to behave in a certain way if you really want to go there so um here you can see like I just run it just for one execution just was one model but you can see what would be your task scores your crew over score your execution time so you can actually compare many models in this and

you can run this like many different times and see what happens and if you want to take this up a notch on the Enterprise side of things that again there is a free tier so you can try it out there is let me see one of the agents that I did run tests for maybe the pr review there is a way where you can compare yeah there you go you can compare many models so you can see here GPT 4 mini andring individual tasks score overall crew score and execution time so you can see that

overall son it was better on a lot of this but it tooks way way longer so probably I probably don't want to go there but I might replace a task or two to use sonin if I think this is useful uh so yeah again you can create all your agents with the CLI if you really want to go there and you can run all those things locally I just use the UI in case for the folks that are less tech savvy and they want to start with something simple totally and you know in terms of

output they're I would imagine they're the same right using the interface or or the CLI yeah the output will be the same um the the thing is once that you're using the Enterprise that automatically becomes an API right so that open UPS like opportunity for you to kind like interact with a few like different apps uh and we can integrate for example with like zaper on this one and I think this could be like an interesting one but uh it's essentially the same thing so for example I downloaded uh creating yes I downloaded the crew

that we just created on the UI and if you open it up you can actually get all your agents in here so you see like the profile researcher is in there you can see our senior one in there that you mentioned senior email content specialist right in there you can see that they're using o mini and JY 40 and 4 mini so so all the codes in here the code generator is also in here so you can kind of like inspect and take it like an a up a notch if you really want to uh

well all right so while this is uh deploying it just to be good um what we're going to do is I want to I want to do something uh a little fancy for us today so I want to make sure that we create a zapier to run this so by the way now that this is live I can actually again because it's an API I can see what are the inputs that it needs so we can see that need domain email and company name I can actually run this right here if I want to so

we can do a task real quick let's say ja crew ai.com uh let's see the domain crew.com and the company is crew AI let's do a kickoff and you can use web hooks like if you're more Technical and you want to like have progress bars on your app and you have to have other things um but you can also just do pooling as well if you want to just see like how are things going oh this one failed great so we're going to debug life that's going to be that's going to be fun but the

other thing that you can also do if you don't want to use the UI uh like the API you have a UI that you can use in here where you can basically create a card for your your agents and again it's going to use the same inputs and they're going to pick it up and they're going to work like directly from here or if you want to you can also export a UI you can export a react component right from here and that usually that said let's the bug let's see what what going wrong with

our crew uh we actually have it in here so we should be able to do create install to install dependences and create IR run just see what issues there we get create IR run and whatever broke it in there should break it in here as well the first run by the way you're going to notice is the first run takes a few extra seconds because it sets an environment um just the environment for you to actually run it there you go the agent is missing in the following description that's great all right so we know

how to fix this uh the thing is on the task there is yes there is no agent for or second task to generate ideas for that we want our idea generator so let's make sure that we get this in there and there we go um this should now be fixed if we run it again it should work and this might be an actually a good way for me to show how you can actually deploy this using G as well so let's say that I build this thing like uh I bu it manually right with code

I'm going to do grag test make this private boom so now what I want to do yeah there you go it tried to run and me start up in a I key that's great means that it's working so now let's do get init make this a repository let's do get ad so we add everything in let's just p out in here and there you have it yes so now we can deploy this grag test uh crew and we can do it directly in here there's a GitHub integration so again if you didn't do through the

UI and you generated things you can come in here and you can do grag test I can add my separate API key my open AI key that I have uh pre-registered and I can click deploy and um and there you go we're going to deploy this one uh similar thing I guess I I should have checked if all the agents were assigned on the UI there in the other one so let me delete this one and while that does its thing I want us to set up a webflow website that is going to use the

screw so um I'm just going to choose random template in here Joe's fantastic website that sounds that sounds right sounds right by the way Joe is there any particular reason why you chose webflow versus framer or another another product uh no really I'm just turning out to be a webflow customer so I really like use them um but this directly should work with anything out there the end of the day it's an API call so you should be good to go so let me delete all this grid and delete the spacer and I'm going to

add a form F block there we go so we know there's three feuds that we need for uh for our thing so I'm going to duplicate can I add a few directly I think I can so fud no I'm going to duplicate this once all right there we go so it's name uh email address and the other one was company domain if I recall correctly so let's do company domain so name should be good let's check settings yes name email address it's good should be email let's check how you're calling it on the C well

I guess it doesn't matter we can map it since we're using zapier and here I'm going to make sure that we do company domain and we can call this domain all right let's check if I did it all right email name all right this feels good let's publish it for now and we can start to set up our zaper so one thing that you will notice is that we have a few Integrations we're still beating out more again this is something that you can do manually as well if you're doing the open seource round but

if you're using Enterprise there's a few integration one of them is zapier another one's HubSpot another one is laack so that kind of can help you out with zapier on that you connected you can actually do you can actually use screw AI in here so let's say that we're going to start with a web hook because we're getting this request from somewhere um let's go for it now continue and we can test this URL I think I have Postman open here somewhere yes so there we go and it's uh domain email and name so let's

add name in here oh and let's do J MOA there you go send it boom request successful test trigger we got a request all the data is in there there you go and now we can open create I and we can actually choose to kick off a crew and our crew should show up in here assuming the deploy is done uh it is grag Task there you go I think deploy is two yeah it's a two happening there you go provisioning once that we deploy in we're going to see it load up exactly the same

feuds that we want in here so we're going to see the name the domain and um and the email show up in here so that we can map from what we're getting from this web hook and after this I'm going to use resend again no specific reason um just turn out to be a recent customer I really like zenoh haha I think he's doing a amazing job in there and we can send an email and uh we're going to use the same email that we are getting from our hook and oh that's not the two

yeah that's two from let's say no reply um reply at crew.com subject we can say um hey there let's chat by the way for people who don't know resend is is a way to send transactional emails um so meaning you know a transaction happens an email gets in and they've also got like a really beautiful um analytics platform um so it is a it is beautiful again I'm not like associated with with recent or crew or anything I just it's a cool product um and they're doing cool things yeah I agree it's funny when you

think about email like it's not kind of like a your your go-to kind of like startup if you're trying to think about starting a business feels like such a Consolidated Market but uh I guess there you go there do you have it all right quick ad break let me tell you about a business I invested in it's called boring marketing.com so a few years ago I met this group of people that were some of the best SEO experts in the world they were behind getting some of the biggest companies found on Google and the secret

sauce is they've got a set of technology and AI that could help you outrank your competition so for my own businesses I wanted that I didn't want to have to rely on Mark Zuckerberg I didn't want to depend on ads to drive customers to my businesses I wanted to rank high in Google that's why I like SEO and that's why I use boring marketing.com and that's why I invested in it they're so confident in their approach that they offer a 30-day Sprint with 100% money back guarantee who does that nowadays so check it out highly

recommend boring marketing.com so while this is deploying let me explain what we're going to be doing next after this okay and we get this working into to end well I find that an email is fun but it's kind of like a it's kind of like a demo right like it's what you see a lot of like other folks do as well I think we can do a custom report custom PDF report something that looks more fancy more drafted towards specifically their company with more specific like we already doing all this enrichment we're already like having

this uh agents kind of create these ideas on how they can use it so why not like go one step further do the craft report and then use that to be sent on the email so the custom PDF report this is something that you would hire typically a junior marketing person to actually put together take the data make it look nice maybe you want a designer there or or whatever but it's it's someone you're paying you know in the US at least kind of you know 50 to $80,000 a year um you're saying basically you

can just automate this whole thing yeah I mean what you can do is you can get like them to do maybe a 10 play right and then based off that 10 play you can like preview the information so you still get like that very nice high quality that you would get in terms of like capturing your ense and what you want to the PDF to look but then what you get is kind of like this very kind of like custom reports per for customer so as soon as someone sign up like as a lead they

get something and you can go a little crazy on this you can put their company logos you can put like a bunch of other things I usually I I love when I think about marketing and I'm going back to my CLEAR bit days here um one thing that I would think is I would love to send a lead materials that they would like to show in a meeting right like imagine saying something so beautiful that they are okay taking that and going back to their team and so like oh let me show you what we

should be doing and how we should be thinking about this so I think there's something special about that and just helping your champion in your like lead and your customer to actually kind of like uh get some props in there so uh for that for this custom PDF report we're going to need a template so let's figure out how we can get a template going uh eest way for me to generate templates now is using chat GPT uh I love it so so why don't we go to chat B here real quick and I have

a prompt in here write me a custom HTML that is pretty that I can export to make a beautiful report there's a bunch of typos in there but it should still works just fine so while chat GPT is writing in our deployment is done so we can go back into our zapier and let's go back into configure there we go refresh results I want to get GR test and refresh in here make sure that we got there we go so we can now use our a company it's not a company name so we can use

domain here I guess I missed the point on that one email and domain again there you go that should be okay all right this is good to go test St is going to try to kick off the screw in crei make that it works there you go and you can get you get a ky off ID in here you don't need to worry about that like the integration is going to figure itself out so it should just work and then as the HTML we basically want the crew result um into this one and we can

test the step make sure it's all fine uh c. oh I think there is a typo in here yes. there we go continue test stab and there we go so we can publish this one uh because we just trigger this you can actually get to see the execution in here uh we just run this 34 seconds ago we can go into the execution we can see what the email looks like he I hope this finds you well I have been follow cre impressive Journey truly inspired about the Innovation Solutions here are a few AI agent

ideas St for crei driven workflows remote teams automated compliance Dynamic customer engagement I I would use those ideas I think that fits into my alley right totally so then again we could optimize the prompt to make sure that no subject is included that doesn't have like this interpolation your name like we we can work around all that but let's call this a win for now let's good to go um there's a few extra matrics that we can talk about eventually but I think this should be good all right so now we have our zapier we

have our URL let's copy this one we can go into our webflow we can go into our form and we can add this I'm going to add this both as action and redirect just in case and publish in meanwhile I'm going to check my email just get it ready there we go I got an email and I want to make sure that we now get the email end to end on this one all right so we publish it let's test it boom by the way let me I think I lost reverse side there we go

is back so this is our landing page amazing landing page future is now give give our email in here so uh in here we're going to say jur email. my domain ci.com as well and I'm going to do subit um H I guess expecting this to be an email let me fix that in webflow real quick this is not an email Feud uh uh so style maybe let's check this out settings domain domain email form there we go there we go it's a plane publish it there we go nice all right so now we can

refresh this and we can do J J cre.com and the company domain is cre.com and I can click submit and there we go because I didn't add the proper redirect it took me to um the Json thing but it should now have a running execution in here stopped what was this C I stopped by this error missing company name company name is missing let's see what is the request that we got from the web hook uh data in data out so query string name oh I guess it's sending a get request so again post requests

publish publish like domains all right so let's go back refresh and let's do it again jaw mow H at crew.com and crew.com and do a submit there we go let's see if it's running now there we go it's running so while the crew runs and we get our email and everything's going let's check our PDF template uh there we go we have a PDF template I'm going to just copy this for now because we're ready downloaded our crew I'm going to do some of the work in here so I'm going to create a folder here

that I'm going to call template and in here I'm going to get a flyo that I'm going to call pdf. HTML I'm going to paste our PDF in there and let's see what that looks like just to make sure that it's nothing crazy s temp plates PDF let me open this up and bring it to screen all right it's okay for us to start again you can do like all sorts of crazy things in the end like because it's HTML and we're going to convert this in PDF you can just make sure that it looks

exactly how you want you could have images across the whole thing but that feels good enough now we want to make sure that this the content of the is completely Dynamic so what we're going to do in here is I'm going to basically ask it to update and I'm using csor here I read love I'm going to say replace the text content with variables we can interplate um using the their pattern boom there we go so now we're placing title intro all right this is this is good now we have a bunch of information for

us to interpolate in this PDF right but our crew is geared towards an email so we need to change that slightly first thing is with crew aai now for now our agents are working mainly with text text in text out it's email but this requires a more structured output we're talking about title sections so we want to make sure that these agents are outputting ad Jon or a penic object something that you can actually use programmatically so crew actually supports you to have like that crew like penic output or Jon output so let's ask it

to uh create a penic object um actually model that will hold out the data necessary to interpolate on the PDF file there you go by the way Greg do do you use cursor or you are a wind surf guy um well I'm actually more of a bolt guy than anything oo I love it Vats amazing yeah I I went to CS School dropped out but finished most of my classes so I'm somewhat technical but like a bad technical person and I found that Bol has been more geared towards non- technical people I hear you I

hear yes doesn't mean I don't like mess around with winser for cursor or whatever I I'm also in the stage where I I want to play with all the tools like I I've just been playing with lovable recently um which is another one and I just like to like I'm a learner you know so we don't know who's going to win yet so I might as we might as well play with a bunch of the tools I hear you and yeah a lot of people get to ask me this asking like hey like what is

going to happen with the market like how are things are going do you think like there's going to be a clear winner and what I'm telling everyone like it's very much early days still like no one knows how the things are going to play out and there's a few things that are not in your control for example you you don't truly know like how this is going to impact like companies or how this going to impact jobs and everything along the way but we don't have control over that but you do have control around like

what are you're going to do with your your time right what is happening right now so I usually recommend people to just learn these tools and as use them anyway because doesn't matter what happens in the future like what to wins or how AI impacts like everyone in the end of the day that will like that knowing these things are going to set you up for better like even independently of what happens totally all right let's go back into this so we just create this ptic object uh now that we have this ptic object we

actually can go to our final task in here and I'm actually saying like hey outputs this object so there's going to be a function calling at the end that's going to convert the output of our agents into this object that we need but now that I just noticed I'm going to actually remove this for a while because we need to update this task so let's go into tasks and say uh I need to update this tasks to now instead of coming up with an email to um get all the data necessary necessary to create this

object and I'm going to copy the object there this is the object that we are going after so I'm going to say there we go so let's see how it's going to go about things um seems to be keeping out the same agent the one thing that I would do is let's update the agents as well to factor this in we are now updating we're now updating this agent to generate to gather all data to generate genate this object instead of writing an email there you go it's going to probably rename our agents and change

a few things but that's fine yes there we go there we go update everything all right for Mini 4 o 4 mini let's do 4 on this one as well uh and in here now um update this to use the agent defined in agents there you go I gotta say once that you got used to this like new tools like cursor and such it's hard to go back totally it's so good it's so good oh funny enough they kept the same key so you can see it's email draft after even though they change the role

so maybe we maybe we do update like update the key update the agent Keys reflect the neurals there we [Music] go yeah go just changing this final one the PDF drafter so let's update this to use PDF drafter all right so now we Chang our agents we changed our tasks again I used cursor to do in AI so I'm not really like taging along and optimizing this prompts very much but if it works it will be it would be pretty good so now what we can do is we can ask it uh update this class

to use the new agent in tasks and it should it should do good job about that well there goes by the way we never checked our email uh I I probably got an email already about this and gu what I did get it got mail you got it I got a mail I got a mail yes let me let me move this into or view there you go I got our first one that we were testing on and I got the I got a new one that now seems to be doing great let me move

this into [Music] view all right email no reply move it here no no no move it move it here uh there we go there we go so you can see that I got a mail and hi Jo I hope this me finds Well my name is again we didn't like necessarily mess around with that I've been called impressive Jenny reach out for some innovative ideas about offers especially around strategic Partnerships customer experience Regulatory Compliance love to schedule a call to discuss would be available to check next week so again pretty cool you could make this

like so much much more advanced if you really want to and more precise and this is kind of like where you get more into kind of like what we're seeing companies doing where you're going to also integrate this with like a Google Calendar you're also going to integrate this with your CRM so now it's not only kind of like uh in like sending the email but actually scheduling the call checking if you have the times lot available and kind of like doing going out the way but this would be a good way for you to

get things started and again there's a few things that we could change in the prompt to not include the subject to not have interpolation and everything but just like a a quick example of how you can get some of these things kind of like going and um and how some companies go about this yeah I think you know tactically speaking if you're if you're building some of these things it's like that what I just saw that email is good enough to start and then spend the next few days just tweaking learning uh and making more

advanced but you don't I guess what I'm saying is you and I'm curious your thoughts like you don't need to go and build the the fanciest output from day one right don't you want to just try and get the the basics going first yeah a th% I think there's a few things that AI really shines in and and one of those is like it's as it's I think it's the easiest way for it to be a agents right now like if you want to be the more custom ones like ignoring like all like the the

more kind of like geared St kind like drag and drop only simpler use cases uh I think this is kind like the easiest way for you to inate because now like we if we spend like an actual couple hours in here going about our agents and our tasks and maybe doing all these Integrations we could get something pretty cool that we can just deploy right away and turn it on it's connected to our landing page and just have it going but I think you got to the point where you got a name in the head

where you want to make sure that you get something out there and once it's working then about inating on it so um all right so now let's just wrap this things up like we we can now pass an extra attribute C output penic again we're getting more technical in here because now you want your agents to Output actual structure data and this is the structure data that's going to Output so we can actually use it and what you want to make sure is that you are sorry what you want to make sure is is that

you are updating this PDF to now include all this data that your agents are getting to you so now we're getting to more Precision what we have done with crew so far is basically very fuzzy and there's a lot of agency like the agents are doing everything but then again as the use cases got more complex you start to need more precision and more like uh repeatability and consistency on their use cases so usually people start with before and after hooks I'm going to show that in a second uh but you can go into guard

rails and out the way into flows and flows are way more complex way more fun as well to build uh but they're way more complex and you can get like some crazy use cases once that you get into kind of like the flow level but um that is pretty interesting all right so for S before hooks we can basically do an after keyof um hook that means that that is going to run after the crew so you do The annotation and in here you can say I will create uh HTML and you get the crew

output as an attribute in here that you can use um that you can use down here so I actually have an example that we can use to ask it to do it for us so so let's say uh update this after key COV function to interpolate the variables in [Music] PDF here is a high level example all right this is an example of this again interpolating variables loading the template and then interpolating a bunch of variables and it's exactly what it's doing right now but very geared stores the exactly template that we have and we

can just make sure that we do the OS import so there you go it's loading the PDF and then uh interpolating all the variables so basically interpolating with everything that we have for the things that are arrays it's making lists out of them and then at the end it needs to return an actual crew output um object and This Crew output object needs to be there we go it needs to be and let me import This Crew object I think it's a from from crew AI is that it no I don't think that's it and

boom so that import the right class so now we're returning a crew output where we return the final template now completely interpolated so now there you have it this crew is able to not only do the things that it was doing before but generates the custon PDF for us so with these changes we can actually just give an update to our project so um get comment uh making it output a template plate and get Bush rain and now we're updating our crew to basically being able to generate this PDF so if I turned out Auto

deploy it will deploy right away if not we can just trigger a redeploy by clicking right in here so let's click on it and that's going to start redeploying this uh crew so getting back into this so we are updating our crew now and that crew is now going to be able to like to generate the PDF templates for us so that we can send them over email that seems good so while that is deploying I think we can take a brief look at flows again those are more technical so some of you might like

say hey all right maybe I don't want I don't want to Trad go that far but a lot of you might want to go that far uh and the cool interesting about flows is that that's where I kind of like a lot of the more complex use cases really come to life um and that's where we see like um like Integrations with sap Salesforce pulling out those different sources executing some python before after Cruise during Cruise things get a little uh a little all over the place uh but it starts simple it's just it can

get complex but it starts simple so this is this is how it works you can start a flow the same way that we created a crew you're going to do crei create flow and you're going to give your flow a name so I'm going to say educational content flow for example and the flow is an ability for you to have an even based automation that have crews in it so let me show you what this looks like so the flow has also a similar folder structure but you're going to see that it has a cruise

folder so you can have many crews inside the flow and the flow follows kind of like an EV like an event based system where you basically have an start function that you can an with this start annotation and then you can have other functions listening to functions before there might be a meeting event so for example it's going to generate a sentence count and then the function generate poem is going to be listening for once that one finishes and this one's going to kick off a crew and then save poem is going to be listening

to when generate poem finishes and then it's going to save that in a text file file so it starts simple but things can get a little fence in here where you can start something and then you kick off a crew and that crew you do some validation of the output and then depending on the result you call another crew if you really want to go there so a good example in a use case where this could be useful and and that's why I named educational content is for example generating long form documents so let's say

that he want to generate a crash course PDF again for the sake of another example where you go over like many different chapters you're talking about many different chapters like evolving into a topic this is a great example of our flow because you can have a function together inputs from the user for example what is the topic that this C is going to be about what are the like the like um the learning style for this uh person and what are some of the main interests that this person have so that we can use those

interests as examples and throughout the educational content and then you have a crew that generates a plan it plans out what are going to be the chapters that you're going to have like what they're going to look like and then you have another crew that writes a chapter so again you're going to have like a longterm content and like a fully blonde kind of like PDF at the end with like many many pages if you really want to incredible dude yeah this this is this is pretty interesting so again I'm going to I'm going to

real quick try to show like what this would look like I don't know if have time to put something end to end together uh grab keep me honest here yeah I'd love just to see what it would look like like to me I'm visual so I need I need to see that before you leave oh so if you're a vis you're going to absolutely love flows I'm going to show you one thing in a second so all right let's change this real quick so let's first um go into our cruise and rename this poem crew

to be um uh let's say content plan crew and rename this file to cont plan crew and rename this to content plan content plan crew there we go now again I'm going to be very like straight the point in here usually you would spend a lot of more time on thinking about your agents and asks but just because we're trying to put something Qui together I'm going to say I want a couple agents to help me create a Content plan for a crash course a crash course on a topic um these this agents will take

a few interpolations like topic [Music] uh learning style and uh interest inter interest there you go uh the topic is what the course is about the learning style should um advise how we plan the chapters in the interests should be used as for examples and such uh final output of this will be a plan with many chapters there we go let's see what they come up with all right it updated it added a couple oh sorry it added a couple in here uh course planner account expert and a learning specialist and yeah this feels this

feels very good now let's make sure that we also update our tasks so I want tasks to help me create a plan all right uh let's use the agents there you go so hopefully between the context of the agent ymo and what I just typed in there as prompt it is able to create me a good set of tasks all right so the final one is recommendation learning activities and engagement for each chapter I actually want def find all right so I'm going to say I want a final task that task that brings it out

together on a comprehensive list of our chapters and our information about them all right and there we have it the one thing that I'm going to say is I'm going to add this context for this agent all the other tasks so create the outline the develop and optimize so whatever the output of creating the outline task whatever the output of the develop chapter content and whatever the output of the optimizing learning all those outputs are going to be sent to this agent to help it to come up with this final final result all right so

now that we have this we just need to make sure that we are loading the right crew in here so this poem crew exists no longer these is our new content crew and instead of starting with a generate count I'm going to say um gather user input so what is the topic what is the learning style what is the interest that feels good now after gather the input I want to generate content plan so this going to be listening to that user input and it's going to call our crew and it's going to pass all

the inputs that we just gathered by the way it's using this state we want to probably update this to be another state there we go there we go so now we're using everything that we want in there so after the Cent plan is generated in this is where you can start getting things a little more fancy we can say like all right let's uh save content plan and we can save this as Jon by the way let's do Jon and in here we can say that we want to get the Jon output out of from

this and in here we want to make sure that we're updating this to use the content flow and the content flow and we have a plan the one thing that I will change though is because if you think about it the plan is a structured object right we're thinking about we have chapters and these chapters we have kind of like descriptions they're going to have titles so what we can do is we can say um create a pantic model that uses the output and I'm going to copy the final tasking here there we go the

output of this task um to hold the structure data of a list of chapters and there we go so we know that every chapter is going to have a title with description learning key Concepts all right a bunch of things and we are going to have a Content plan at the end so now we want to make sure that update is through to use the tasks and agents so it's going to update this to there we go use our new agents and our new tasks and we want this final task to do an output Jon

that is conted plan so we created a crew again we're like we did everything very fast in here I told you but call thing we got a little complex and we uh and we have our flow that we created in here so the cool thing is we could like now have another function in here after we save the plan where we can say generate um content and this generate content could be listening here let me do this could be listening to the same function to generate content plan so this is an interesting one you can

see that after you generate content you do two things you save the content and because that can happens synchronously we can go ahead and generate the content directly so we don't need to wait for the save to happen for us to generate the conty and this is where we could have another crew and This Crew will create uh each chapter in a loop for example and after we have created each chapter in a loop we could have a another one that is going to be saving content for example where we're actually saving the content so

we can do a key here there we go so all right so in here you can see We Gather the user results that's where thing starts then after getting user results we generate the content plan then after that we do two things we save it and we generate the content with another crew and then we save that content somewhere the cool thing is because Greg is a visual person you can do now crei flow plot and that will actually plot you a visual showing exactly what is happening with your flow so you can see what

are the crews that you have in there you can see what are the inputs like what are the different kind of like functions and events that are being emitted and that is something that people usually add to their Ry files and makes like parts of their documentation moving forward so let me open this up now it got generated so so I can show you what this looks like right there we go so we gather the user input we create the content plan and you can see this is a crew method so there's a crew in

this one then we have save a Content plan and generate content eventually generate content is going to be a crew as well so it's going to be similar to this node and then we save content and there's a bunch of more things that you can do you can do router triggers and and some other things but you got a visual representation dude I I had a smile on my face I don't know if people saw that because I'm like this this this is what I need you know I needed this I'm going to say it

does helps with like this more complex flows because like uh again you even if you crack it and you get the hangout of in you're still working on the team sometimes there's beaing this so the more visual that you can get for people to just like wrap their heads around what's happening the better and the easier it is for you to kind of like build the more complex flows like this one cool uh and then again like if you if you want to execute that is similar thing you can do c flow key cough and

I think we don't let me double check one thing I don't think we have environment variables on this one no we don't so let me get our environment variables on this one all right there we go one second let me move this in into screen yes so now we can do crew AI uh flow Ki off and that will kick off our agents again we're missing a crew in there the crew that actually generates the content but you get to hang out with it we could like do that now and do that in the loop

but it's asking for input so the topic will be a nuclear fusion I'm a very practical learner and the interest of the target audience would be games and Anime there we go and there you go our first uh our first crew is starting trying to create a high level curse outline so these first agents are going to do their thing you can see chapters introduction to nuclear fusion the science behind it Stellar Fusion process Fusion in technology current challenges the future and now the subject matter experts going to create an outline for each of those

chapters and things are going to keep going and going but you get the hang on how this is going to turn out to be now if we do add our second crew in here what we can do is because we pushed our plan into this and we know that this plan is basically this content plan that has chapters we can Loop that in into another crew so content plan here is actually content plan is actually not a string but instead there we go and what we can do is on this second place we do content

plan uh we actually do for there you go for chapter into each one of those chapters we can do content creator crew and what we're passing here at inputs we can we can pass those things individ ually if we want to or we can pass the whole thing um if we pass the whole thing it's going to be formatted weird but it might be okay so uh reg how are we doing on time can I keep going we yeah we we got we gota we got to keep going a little all right so let me

just copy this and letting me paste it in here and now this is going to be the content creator Creator cre crew so now we have this other crew in here let's rename this as well this is going to be cont Creator crew in here uh we don't need this anymore and we can go to our agents and we're going to say all right uh let's create new agents that are now responsible for creating a specific chapter they should research the topic online find interesting angles and write the chapter using the interpolated context day we

get with the following information and then and we can basically tell like this is what they're going to be getting Mission and let's see how things are going to turn out all right this seems to looks good let me approve this right away boom so we have three agents content researcher a technical writer an education content writer this feels good now let's update it our tasks so let's update the tasks to actually write to actually write the content now for for a given chapter using the agents defined in agent we can remove one of those

and I want to say they will have it access so this let's make sure to interpolate a chapter variable so tasks uh let's make sure we interpolate interpolate uh [Music] chapter variable that will include all the information boom there we go so while it rewrites that I want to make sure in here see it's trying to do some facing things research let's do not do any interpolations in this one this should be good oh the interest is good if we can add interest in here I think this could be interesting so learning style and interest

all right this feels this feels good all right what about the tasks in the tasks let's approve it right away as well uh I won't try to do anything uh fancy about this like trying to access the objects this I don't think would necessarily work so what I'm going to do instead is I'm going to basically say chapter chap chap chapter information information and we can do just interpolated the whole chapter thing consider how content and there we go I found this I found this way better engagement strategies boom and then we can do the

same thing for the other ones so interpolate only chapter equal equal in the same way we do above there we go all right so it's removing everything just keeping chapter well do one thing though it just did for this one so let me do everything below boom all right so chapter information is in here with view chapter content this is good optimizing for learning compin to final all right let's add chapter information all right this one get a little bit like a little bigger like we have a a few more tasks but that should not

be a problem now we want to make sure to um update this class to use the agents and tasks and there you have it Con of Creator crew all right updating everything this feels good all right now what we're going to do is by the way one thing that we don't mention is what format we want this I would say make this uh a markdown a markdown file and what I want to do on here is no I think this is good I think this is good we shouldn't need any of those as well so

now we can import our new crew so we have our content creator crew and we are generating this for each of the chapters and what we can do is for the sake of just keeping things a little simpler is I'm going to save here so save res. raw not raw to uh markdown file in um content slash there we go and we can create here here a new folder I'm going to say content for now and there you go we basically got the chapter name this is good we need to interpolate the other things as

well yes learning style interest yes yes yes all right like will this work I don't know brace for impact we did a lot in there you can see that when we run our crew the first time it did work like like we got all this plan so let's actually open because it saved that file that we asked it to um I hope so where is this yes content plan let's do Json beauy there we go so we have all the chapters introdution description learn objectives key Concepts technical content practical examples learning activity so you can

see the plan is in here for all the chapters so this looks good now now what we're going to do is make sure that after the plan is generated and we are saving to this file we didn't even need to save it we can drop this part eventually uh and you we actually call the thing that generates all the chapters so this should be good one thing that I'm missing though from uh This Crew is that they're not doing anything like they're using only the llm knowledge I would love to use for example um create

I tools and U maybe server so they can search line and maybe scraping like I think those are probably like the couple like easiest tools that we can use and then we can give these tools for some of these agents so they can find more relevant up to-date information so maybe the content reviewer get thems as well um and and that should be okay for these but for the ones they're planning I think they're even more important so the planner should actually um use these tools and and the outliner should probably also have like these

tools accessible to them and there we go because I'm adding the serer to I do need to add an environment variable but I think I have already done this before when I copy down my environment variable so that should be good um all right let's let's give it a try let's see what happens again we're doing we're doing a lot of things live here but I'm feeling optimistic are you feeling optimistic gr I'm I'm always feeling optimistic you're a better human than I am I will say like when you know obviously when you're building these

things you want to kind of limit the amount of variables right but we we we do have a lot going on uh but we do also have the CEO of crew AI so I'm feeling optimistic I love that all right so want to learned about lucar Fusion again uh learning style is practical uh games and anim Ma and AI as well let's see if like what a turn up in there all right our first crew we know it works so let's do its thing one thing that you can see now it's using our tools uh

searching the internet nuclear ficient curse outline educational examples and there's a few contents in there from like K12 to um Institute of physics and now it's actually reading some of those websites so it's going on and kind like trying to get information now searching for nuclear fusion curriculum development examples uh in has an isal outline so introduction the science of ficient real word fish and popular cuture and debates and perspectives this is good you see this is what happens when you give it like some extra tools the ability to search now it can go like

all right what about the debates like is this good is this bad I know a lot of people say like we have a huge nuclear fish in the sky the sun maybe should use that so you you don't know um and then subject matter now creates a description for each chapter talking about the key Concepts the important and look at this up engaging activity explore the use of nuclear fusion in popular enemy series like Attack on Titans so if there any enemy funds right there like as myself uh this this is very interesting I do

like that uh and for games it's mentioning about Fallout and the consequences of nuclear warfare so um very interesting one so it keep on going and going generating all the chap um this should be good I think this is the final agent now that actually creates the plan so if something is going to break it's going to break now between finishing the screw and kicking off the next one where it's going to like it's basically the the brand new piece that will be put together so cross fingers uh I think I think this is gonna

I hope this is gonna work I'm gonna say this is fun uh I don't I don't get to do much coding like uh nowadays just because of how the business are going so it's fun to spend some time doing this now especially live coding it's live coding yes I mean there we go of course we had a problem the all the things were good but then uh in the man to generate content a string has no attribute chapters all right so let's let's debug this real quick see if we can fix it is on this

form right here so if you look at this form is trying to Loop this variable here and this variable is a Json content so even though it is Json Jon for now in here is a is a string right so what I can do now is I can do penic um so it's an actual object on our planning crew instead of saying outp put Jason I'm going to say I want you to op penic for me and I think we we don't need to save this anymore into AJ five I mean we can still save

it but maybe now we need to do the the model dump Json uh but yeah let's let's not not just do that for now I'm for the sake of running this faster I'm going to remove our tools like we know that they work so let's remove them so it's basically more straightforward this agents run a little faster and let's rerun them and cross fingers let's see if it this one's going to work now uh [Music] um curse is going to be again uh nuclear fusion I'm a practical learner and uh anime games and AI all

right so crew now should be run faster it's not going to like spend so much time going over like the smaller detail so cross fingers okay now I'm really optimistic it's always like that right I mean for for the folks that have played around with like designs and everything it's like this is the point where you start to put underscore final underscore two and and you're wondering like is two the final version but I think this going to be a good one I'm I'm feeling optimist stick on this one as well all right uh the

one thing that I'm remembering now is um we did our our PDF thing I don't think we ever we ever test it out so maybe maybe it's worth it for me to uh try to see if that PDF deploy worked and we can check that out in a second as well just to make sure that everything is working fine on the PF generator side of things I make sure that we uh we don't drop the ball and leave people hanging on that one n all right so while this runs let me check that um all

right so our deployment is done so we should be able to update our zapier now so let's edit our zaper real quick all right so um there is a catch hook that works still we're still calling our same drag test that FS okay let's make sure that this is working that I didn't break in so I can re retest time uh all right seems to be working okay now remember that instead of sending the email now we getting an HTML at the end so there's a there's a service that I have used in the past

called PDF Co again I it's the only service that I could found that is able to trans like dynamically convert HTML to PDF so I'm going to use in here just for the sake of using it and what you can do is you can get your output of your crew so the result and basically output it in um as a pgf so I'm going to say I want a no margins uh this feel good all right so let's see if it runs yes there we go and on our resent now instead of like sending the

this content we can basically say here is your report and we can attach the uh URL continue publish publish and there we go we have updated it now so let's make sure that our crew is running accordingly we should be able to kick off a new one so I'm going to T is this the one that we I think this is a new one isn't it a new one let me make sure that we're using the right one in here so the kickoff I want to add this app real quick I want to make sure

that we are refreshing this list I'm going to choose another one real quick let's find our grag one grag test all right domain company name continue continue continue this is good continue continue continue and well we didn't change anything all right feel good let's try it out then domain is cre.com company domain is C.C I guess what happens is or cursor Dynamic updates kind of remove the name the person's name from the interpolation that's okay we just make sure this is running all right it is running so this works so this now should work all

the way from the getg go so let's let's do a request boom do we have a new run yes we have a new run going on here oh no that's not in there you go this is in all right let's see how that goes while that runs we go back ah another error now is error in the generate content content expert so it's not finding an agent that is called content expert I guess that's what we get from doing everything on um everything using curer so what happens is if you look at one of our

tasks it is looking for a contact expert agent but if you look at the agents we don't have a coner and agent so we can do a Content writer on this one and the other thing is you want to make sure make sure our agents in here exist on agents and there we go just to make sure that there you go there was another problem in there another one in here there you go now it's better and then in here let's say let's make sure we are using the r let sir we are defining the

r tasks and agents and then we can use the Creator and the Creator and there we have it so just to make sure because now we have like two Crews I might have tagged the wrong files I think that's what happening there yeah you see I think that was exactly what I did so we don't need this context in here for now all right let's run it again we go back to our zapier one so our zapier one it seems there's a few that are stuck maybe is that what's happening in here it's calling the

crew is delayed so maybe this crew is running let's double check that on crew AI get status failed cre attribute has no attributes benefit list again see me thinking that cursor is going to replace engineer so [Laughter] soon it's going to take quite a while so this is it we're trying to look for benefits list in what is there crew object has no benefits list in that's because we are looking crew output directly we want to do penic um output and we can do there we go identic and then our crew outputs we replace everything

with identic output and F so nuclear fish practical anime games and Ai and then for our other one we just update oh we just update the pr GS Comon fix push aring mang and then let's do let's trigger redeploy in [Music] there redeploy there we go all right so now we fix we fix both hopefully let's make sure that the things runs and that we got def the res I guess third ch's a thirdd time is a charm totally happens to the best of us man yeah I gotta say this is live coding right I

think like to everyone's point on like how AI is evolving coding is still a very interactive process right like uh definitely Ai and agents and things like Davin now and like other solutions they help a lot but um it's still like a lot of of like just making sure that from going going from zero to one is very fast and getting faster and faster going one to 10 is still very much like um what traditionally has been yeah that's that has been my experience so far all right a similar here but now on the task

side probably same thing I forgot to reference something somewhere so on the tasks I'm saying create a Content out line I'm actually saying create line and I don't have that task in here so all this context should actually be research chapter uh should be right chapter C and compile final plan I think this is it let me just double check this is the right one this is the Creator all right so compile defined curse plan complete structure plan now this this looks odd if you think about it that's the content creator plan the plan is

the one where the tasks are about compiling the outputs in the compressive plan this should be actually right in the chapter so if you look at this we can say inad inad stand off a plan we want the final actual chapter on this [Music] one right a complete polish mark down all right this feels more this is right in content reviewer we have that agent yes we do do we have those tasks yes so we're using the research we're using the writer and we're using the review chapter content all right this feels good and this

is our task so now let me make sure that we get our tasks from the content creator I'll properly defined in here um all right so research research WR wrer chapal review review and then optimize optimize this feels good and then compile find course plan we can keep that name for now even though what's doing is actually writing the markdown and the context in here we don't need to pass it given that we already passing in the tasks and let's kick it off again I think that's it only one way to find out only one

way to find out all right topic do you have another topic by the way other than nuclear SE um sure what about seo seo what is your learning style you know me I'm visual Visual and any interests I like games I think yeah all right games there we go all right let's see what what is going to be the plan for SEO for a visual person that like games yeah uh now I'm curious to see what the plan is going to being all right introdiction to SEO it's relevance in gaming oh there you go uh

keyword research and optimization onpage SEO fundamentals content creator marketing strategies technical Essentials tracking performance how you feel about that s plan yeah I think it it makes sense you know makes a lot of sense I wasn't yeah I'm I'm I think it looks good all right well introducing Concepts a couple use cases how fortnite effectiv use SEO to rank highly on gaming content and analyze Legend of daa breath of the wide I mean that's a killer game right there totally um then keyword researches using things like Keyword Planner Sam Rush a arffs that feels good

on page SEO again OverWatch call of study that's is interesting Destin 2 Hollow nine another great game then technical results and SEO performance directly all right s s the track then there's the final plan right there the whole Json thing and it works generating content so now our content researcher specialist actually is getting all that objects so you can see all the capture chapter information is in here the title the objectives like everything is in here it's always matched together because kind of like we Jem the entire chapter like object in there we could make

a little prettier and now because this crew we didn't remove the tools they're searching the internet so searching the internet finding information scraping websites there you go a bunch of website content that's kind like just reading and researching further and coming up with the actual content so now is the actual content writer we can see an initial writing of what the topic would be over view key concept the technical content the Practical examples learning activities there we go now I think we have uh the editor kind of like doing a review and again researching online

scrapping AB Pages like that the models are getting so fast now that it's even hard to keep up with everything that they're doing so there you go researching how technical visual Learners like in Practical and fun tips so making sure that it gets like some of like that uh that Insight in there yeah learning gaming for Education yeah it's going it's going out loud all out baby oh there we go all right so we should be able to see like and by the way this we're seeing all these logs because we're running on verbos mode

so you can disable all that if you don't want to get like everything and kind like the the stream of cches of the models and everything that they're doing but H if everything works out we should start to see uh files there we go populated right in here y so it wrote the first chapter Let's uh let's see let me actually remove those initial markdown codes here that it added there we go and then let's do markdown preview so introdution to se's relevance in gaming the learning objectives the key Concepts the C the Practical examples

what they have done learning activities we create infographic illustrating how search Eng GES Works wow that's going to be tough one um assets methods success metrics this feels good all right well I guess we what we could do and we have done like this like um this kind of like flow before where you use for like high level content creation is you can go like way deeper on some of the pred for the planning that's usually where most of the kind of like the optimization lies is making sure that the plans reflect like what exactly

what you're trying to kind like go after like the the fact that he's trying to not use game as examples but actually focus on gaming SEO is a little odd and it's not what we want so we could change some of the tasks and agents around that it's funny because I saw I saw that originally and I was like that's not what I want but then I was like actually that's probably a huge opportunity in like a niche um like you know I've seen a million SEO uh you know courses and content guides and stuff

like that but I haven't seen anything specifically spoken to to to video games yeah I gotta say definitely definitely sounds like a nich something that like U I think like people would really like U do and you can see a difference in there for like specific Industries on how they're playing some of this but um yeah so now what what is what is happening there behind the scenes again we have a new chapter now now talking about uh the keyword research optimization so we can double check what this one looks like uh where is [Music]

preview uh there we go so keyword research and optimization in gaming so introduction learn objectives the key Concepts explaining what is the keyword plan there what is the seam Rush uh and going over a few others about search volume comp competition relevance um and then examples of flow charts activities engagement strategist success metrics how does this trct with you someone that actually like uh that actually does care a lot about like SEO I mean it's it's directionally right you know of course it needs to be optimized yeah it would have been insane literally if it

didn't need to be optimized but I think like this is this is a this is a legit output yeah I think honestly a lot of like the optimization like happens especially on the firstness parts like on the actual agents is on these agents and tasks like emo files it's where like you're going to be like uh spending a lot of time like to make sure that you get the behavior that you want yeah and then when you get Precision is more on the python files is where you spend the time like adding checks creating this

python objects structures where you can use guard rails and where you can use like before and after kickoffs or where you can build flows so this kind of like where we get like where you optimize for precision but uh and funny enough is a piece that we spend less time on just because we let AI right out our agents and tasks for us but usually you do can optimize a lot of this just by kind of like revamping some of it and it gets something very good so uh another example of something that we have

done in the past and that was pretty interesting is building a crew that monitors AWS for us so AWS where people host a lot of like their software and we had it to actually produce um a like a markdown report at the end but in this one we spend more time describing like what exactly we need and this kind like the level that you can get where we have an executive summary table of contexts introduction a very detailed analysis and you could have actual link you can add guard rails to make sure that these links

are not like hallucination so if for example it's mentioning a link in here and you can click on that what you actually get is control traffic two resources Pages yes there you go so you can actually have exactly the page that is a valid one that's really cool yeah that's really if you spend the time you can get amazing thing like risk assessment Matrix road maps um this is actually a demo that we're going to be pushing with AWS in general so it's going to be like on their blog post and everything but uh this

kind of like the level that you can get if you're really double down kind of like u a lot of this all right hey we we covered a lot I mean time flew to be honest um I hope people stuck around like if you're listening to this and you're in in the deep end here with us like you know you realize you probably realize that I mean look at this look at the outcome that you can you could potentially get um but there's going to be zigs and zags along the way there's going to be

errors there's going to be you know it isn't press a button and it automatically you know you have this agent force multi-agent you know ecosystem but it is press a few buttons Zig and zag and and you and you get it out so um Joe thank you for spending time with us showing us around bringing you inside your brain and and really how you think about uh this multi world and showing us crew where could what do you suggest people do I mean I know use crew but like what what do you suggest people do

if they want to get started in this world uh I would say if people want to really dive into this uh I would recommend checking a couple courses that we have uh with Ender Yang on deep learning AI so I can send you the link so that you can put on the descriptions but uh there's a couple courses the first one is is pretty good it's like very introd like from introduction it's like very for beginners out the way to can like start building everything so if you do that we're going to step back out

the way to kind of like what is an agent like how agent Works what are tools what are llm so you can watch that and catch it up on that way and then there's a second course that we did that was very practical it's all about what we're seeing in companies out there um and there's an interview with the CTO of gen on PWC on how PWC is using crei in production and there's a few extra informations in there as well that you probably want to check it out so I would say those two like

we have over 100,000 people that enrolled on those courses so I definitely recommend you to check it out cool we'll include that uh get your hands dirty and and of course like and comment and subscribe uh if you enjoyed this video and uh I'll see you in the next one Joe I'll see you around see y thank you so much for having me thank you later [Music]