BONUS: A Total Beginner’s Guide to AI Agents & Automation

30:22No, they fixed that ASAP. They fixed that ASAP. Okay, I'm going to share my screen here real quick, just to show you what basic JSON looks like. I like Patricia's comment while you're doing this. JSON relieved us from the constraints of structured databases. At least I assume you're talking about JSON there. Yes. Yeah. All right. Oh, here we go. Here we go. Here's a good example. Here's a good example. I'm going to blow this up so you can see it. JSON, what it's essentially doing is carrying data from a database or to a database or from a database. And so you'll have different categories where you'll see like, what is this person's name, age, city?

31:05Anytime you go and fill out a form on a website, what you're actually doing is creating JSON behind the scenes. You just don't know it. It's what the form is doing for you because that's how it's sending it to wherever it's going most likely, whether that's a spreadsheet or a database, whatever the case is. But it's just a thing like this. You don't have to necessarily know how to write it, But what is good to know is that you can go through and do things like make changes pretty simply and recognize little problems. If you at least have a passing understanding of it, it will be very, very helpful for you because you'll be able to spot problems, fix problems, and better understand where things are happening in the agents you build.

31:50Yep. Another consequence of using JSON is, for example, somebody mentioned N8N earlier. N8N is like Make or Zapier if you've ever used either of those. It's like a workflow automation tool, a node-based tool. And you can actually export and import JSON directly. So all of the nodes that you put together, you're basically just putting together a big JSON file, and you can make workflows really easily from JSON. So you can ask AI to basically put the JSON together in a structured format for you so that you can create an automation with it. So it's nice to know that it exists and get comfortable with copying and pasting it and working with it.

32:36And you can always, if you don't know what's wrong with it, copy it, throw it into ChatGPT or whatever you're using, and it will absolutely find problems and help you line them out. But I dropped a five-minute JSON explainer in the chat, which is essentially what I watched to better understand it. And it was good enough for what I needed. And then the occasional tinkering. This is not a thing you'll use in every agent. It's not a thing you may ever come across, depending on what you're building. But the knowledge of it will make your life a lot easier because quite often when you go to download an agent, you'll wind up with a wad of JSON.

33:14We got a good question in the chat. Level yourself up. Corey will share one example for sure that he uses a lot on that. And I can comment more on that later. So basically the question was for anyone listening. So what is your best case use of your AI agents and automations that help you productive wise daily? We will get into that more in depth. Yeah. Yeah. And I'll walk through what I have, what I do as well. And, and, and it'll really hopefully help illustrate that because I have a number of things that get my morning in order and make sure I'm, I'm ready to kick all of the butts when I start my day every day.

33:51Yeah. And we have one more term, technically two that we're going to cover before we get into more. Yes, we do. Yes, we do. So you might have heard us mention connectors earlier, or maybe you've seen this term, and we said no jargon. This is the one jargony thing besides APIs and JSONs we're going to talk about, which is MCP, or model context protocol. And basically all you need to know about this is that it is a standard way for AI tools to connect to apps, data, and services. It's a metaphor that a lot of people stuck onto was USB-C for AI tools. But basically, it's just a structured way to make tools and data and services available to AI agents so that anyone can use them, whether you're in chat or any other situation where you're communicating with an agent and data.

34:42And they take the form of connectors inside the applications that you use. So Corey's showing some examples here. yes and uh and like all of these the click up doordash figma i think little caesars may even be connected in here it is yeah we tried and failed to order a pizza a couple weeks ago we're going to do that again soon um yeah these are all mcp connections behind the scenes and uh they they have a number of things that they can do that are defined by whoever owns the mcp uh you know whoever owns that service is able to say, I want them to be able to read, I want them to be able to write, but I don't want them to be able to, I don't know, add an image to a thing.

35:27So, you know, you wouldn't make that ability available. Yeah. So Connector is like the easier user-facing version of an MCP server. So you can connect Gmail, Drive, Slack, ClickUp, all the things that Corey just shared. And the agent can then use those tools with permissions that you set. So there's like a menu of tools, tools, functions that the agent can use when it's connected to stuff. And you can define which of those tools it's allowed to do. Yeah, MCVs are a little tricky, but basically they are, you know, simple enough that you can create them with the help of an AI. And you can certainly use them with AI.

36:14It's an easier route than dealing with APIs quite often. Yeah, it sort of was meant to replace having to teach the agent how to call each API and give it the docs for each service every time. It lets you authenticate your account as well, which is one of the difference. So like if you use the sign-in with my Google to use other sites, that kind of verification still works through an MCP quite well. We got a question about MCP versus plugins. You know what? Honestly, they're very similar. You can basically think of a plugin as a package of skills. And you can include an MCP server as part of it.

36:56So it's like a package of MCPs and skills and docs, basically just putting all of the stuff that you would need to do a series of tasks together in one place. We can go into that a little bit more in depth later when we get to Claude. but yeah the line between plugins and skills and mcp servers is kind of and so and apps it's all a little muddy there's a little muddy yeah that there's a lot of intertwining there a lot of apps are using apis and mcps and a lot of mcps are connected with apis and there's all of these different ways you can do things there so you know at its core though the thing you need to know is it's what lets you connect your app to your AI.

37:40Yep. And makes it easy. Somebody just asked if we'll be sending the recording. Absolutely. This will all be archived later. You'll be able to watch it for eternity here on YouTube. You can also take the, take the bar, the red bar and scrub it back all the way to the beginning. And you can watch from the beginning right now. Yeah. We just got a question here from level yourself up that I think this is probably a good spot to address anyways. When dealing with agents, model selection really, really matters. Where what they are talking about here is specifically like token usage within a subscription account.

38:20So whether that's your ChatGPT, your Google, your whoever, your Claude, whoever, there's a limit to how much usage you get in a month, depending on your plan. like the$20 plan, the free plan gets X amount. If there's a$20 plan, it gets this much. There's a$100 plan, it gets this much. And those go up and down, you know, depending on model usage. So it's not a, there's not like a firm hard line, but the best thing you can do if, if upgrading a notch is not a thing that's currently in the cards for you, which is totally cool. It took me a long time to get comfortable with having to upgrade mine.

39:05And what you need to do is better understand model selection. Like if you find that you just leave it on thinking and extended for every single thing you do in ChatGPT, you're going to run through your limit a whole lot faster than you will if you use it on instant and then go to thinking when you get an answer you think sucks. If that makes sense. Like, you know, it's... You get a finite amount of token value is what I would say. And different models eat at that value faster and in bigger lumps than others. So like instant is just taking a little nick off the corner where thinking is more like pulling a little hunk and a little hunk and a little hunk.

39:43And pro model is more like, roar, taking it all. So that was silly, sorry. Oh man, I'll let you wrap up, but I'm fielding some interesting news in the chat that Opus 4.8 just came out. Oh, God, here we are. Yeah, the last thing I want to say on this is, and this is more important with agents, just like it's important with your subscription, but I'd argue it's more important with agents because you're using token-based pricing, which is way more expensive. So you want to be sensible with your budget. Not every task needs Opus 4.8 or GPT 5.5. The fact is, there are a lot of tasks that like GPT 5.0 can still do just fantastically great.

40:32And it'll do it at pennies on the dollar. So understanding the difference as you build agents in which stages require certain things. Like if you just want to classify an article, for example, all you've got to do is like you could throw that to a nano model. I use Mercury 2 a lot in my agents because it's very affordable. There's like no latency because it's a diffusion model, but I don't want to get into all of that today. But it is an option, and there are ways to very cautiously choose the right model for your task. We've got an article on this I'll find and share in here too. I pound the crap out of chat.

41:10GBT at$20 a month. man, I beat that$20 chat GPT subscription. I have beat it to death over the years. Go ahead. I just want to address a comment. So Kitty Bradshaw, I really appreciate that you said this. So Kitty Bradshaw said, this is overwhelming. I think I need someone that can teach at a beginner's level. So I think hang in there. What we're going to do is we're going to actually visualize this now and try to put it in context. And we'll go slowly so that we can take all those terms that we just learned that we threw at you and actually like break it down and show you them in action show you what they mean what they look like in a system yeah yeah because this is intended to be beginner friendly so please let us know stop us at any point if you get confused if we're going too fast let us know um the goal is to try and help everyone understand and there are some of these that will absolutely do it for you is a thing to know like chadgbt workspace agents, for example, or ClickUp, which I'm going to show off here in a little bit, because I've just started using their AI stuff this week, and I'm unlocking a lot of things that have me excited, that you just tell it what you want, and it'll build the thing for you.

42:23Absolutely, that is a thing now that works quite well, and it's only going to get better. The next 90 days, that'll be just how agents are largely built, is with natural language. but understanding kind of the mechanics of it will help you better understand what you can do with them if that makes sense like if you know we can do this we can't do that that's not a thing that's going to work you'll be able to better look at your own tasks your own jobs the things you do in your day-to-day life that you want to try to automate and have a good idea of you know what it is and what you can make happen hopefully all right so should we show some of this in action here which one do you want to show grant uh what do you want to start with um i'm tempted so we have here chat to bt we could show the agent workspace there but um that that not everyone's going to have that so i don't know if that's yeah that's only available in the business plan right now is the only the only stinker part of that maybe uh make yeah we could do make makes a good simple one click up's a very simple one you're not building your own agents in click up either it's just you're just telling it what you want okay um so yeah maybe let's do those two first because they're a little more beginner do you have a do you have something you want to show and make if not i can show something i do not if you do that would be wonderful all right let me log in and then i'll pull up my screen so this is a good one to start with because this is more like an automation than an agent and I'll show you what I mean.

43:59Let's just take a sweet time. What are good resources for using AI with GIS?

44:12Geographic information systems, is that what you're talking about? Geospatial AI? Let us know and we will give you a good answer. Okay. Yeah. All right. So what I'm about to show you is an outdated workflow, photo workflow that I have in Make, and we can make some changes to it.

44:41Okay, so should see my screen here while it's loading. Okay, so this is an example of a node-based tool that is an automation workflow. And the way that it works is basically like this. So first it starts with a bunch of links that I put in a Google Sheet, right? And so that you can think of as the trigger, let's say, is that a new link goes in this Google Sheet, and then that starts this process. I can also click this run once button here, and that will also fire it off. Now, how have I connected this? I've connected it with an API. So I've connected it to my Google Drive here, and then I was able to fill in all of the JSON fields for the spreadsheet.

45:39So in this spreadsheet, I have all these fields. the spreadsheet's called news today we've got just basic sheet one that might be a good place to illustrate that grant because i think there's a code button there isn't there where you can see that you can look at it as code or just as the form like you can do it either way with these i want to say i think you're right that may not be a make thing i could be wrong yeah and then i think you can do that where basically it'll convert this to json but basically i've and then i've set a limit here which is like a guardrail so it can only run 35 rows at a time it's got my rows with headers here and then basically what happens is once I hit run or for example you know new a new row is added it goes through this iterator so if there's you know let's say there's 35 new rows that are added for each of the 35 which this is a this is a function that kind of runs it says for each row number.

46:36It's going to run these tools. And then what are those tools? I'm forgetting exactly what this does. But basically, it takes a variable name and it converts it into a value. So yeah, it's kind of complicated. What you need to know about this is that you just ask AI to help you set this up for you, and it should be able to do it. And I will show you an example of that. I'm trying to, you can't like click to drag, which is kind of annoying. Okay, there we go. Okay, so now what happens is for each of those rows that I add in here, it searches perplexity, and it looks up the links for all of those rows.

47:18And it looks up the story, it uses this model called sonar, it uses this prompt, which I put in here, and you can see it kind of shows you all of this, how it works. The different variables as they come in, yeah. Yeah. And so all of this, again, you can think of this as JSON. In fact, let's just show it here. It shows you the variables. And then what it does is it runs it through Claude. And this is Claude Sonnet, who reads basically everything that Perplexity finds. And then it sorts it. And I'm going to zoom out to zoom back in here. It's very tedious and annoying. that is can you is it a is it the one where you click with two fingers and it'll drag yeah there we go there you go i told you it's been a minute since i've used this okay so then what happens is uh on perplexity it then sends it to claude and claude does a sort a sort and it basically looks at what type of content is this and it classifies it and based on the classification it sends it to one of these different routes so it sends like okay it it says row number one is, let's see, around the horn.

48:30So it's going to send it to the around the horn path. And then it sends which one of these is around the horn. Quick call out here that LaShonda mentioned is that there are good learning modules in Make. Most of these tools have them as well that will really kind of walk you through the nuts and bolts of building one from scratch. Yeah. Or from a template. Yeah, exactly. Yeah, but basically it sorts it into one of the four categories that goes in our newsletter. And then based on that, it runs a prompt to write it in our style. And then it adds the updated written content to the spreadsheet. So something I think would be helpful for people to understand is while you show us that, Grant.

49:10Yeah. When this happened, what was the task you were trying to automate? What was when you decided I need this? What happened? What did you first decide? Well, the first thing I decided is I said, hey, we get a lot of news that we read. every day and it's overwhelming. So I need some help sorting and categorizing it because there's some very obvious things that we do every day. You know, if it's a, if it's a tool, new tool that just came out or a new feature on an application that we use. One summary of things. And yeah, that's treats to try. So I just need to categorize that, put it in, put it in this bucket over here.

49:46If it's a news item, like, you know, OpenAI announced they just raised$6 billion. I need to categorize that and put that in around the horn because that's probably where that's going to go um and then you know there's some more niche ones like hey is this a funny thing that could be good for the blurb I'm going to categorize it for the blurb and put it there and if this is uh more of a thought piece or a think piece or something well that's going to go in intelligent insights and so that's going to put it there and so I said I need to come up with a better way to sort and categorize this flood of news that I read every day how did you decide on your trigger uh your trigger is the line creation in the spreadsheet correct yes so basically once the once the row is created with the content in uh column a let's say um you know the link to the news item is in column a then it runs for that and what made you think you know what a spreadsheet's the best way to do this i should start with that forgive interviewing you here on the spot with no I think the reason, well, this was like, I made this over a year ago.

50:48I think the reason I did this was because I'm already looking at content and pulling it and putting it in some sort of database already. You know, I'm already putting it in a spreadsheet. So it makes sense to have the automation live in my spreadsheet. I think is why I did it. Yeah. Okay. Yeah. And then how did you, so, you know, I'm going to start with a spreadsheet. And I want it to go do all this stuff in the middle and come back to a spreadsheet. Right. How do you decide what each step in between should be? It's a good question. So first of all, I think when I put this together, I was working with Claude to try and come up with the exact workflow.

51:32And things like iterator and tools and this kind of more data formatting stuff, I wouldn't have thought of myself. So I think it's suggested to do that. Yeah, using an AI model, it knows, I mean, you can take it the docs page and be like, here, look at all these features, tell me the best way to make this. And it has more knowledge of that than you ever will. yeah then I knew that I needed a an AI that can actually read as many websites as possible so at the time that I set this up perplexity was the best at that category it could read almost every website that I would put in there and there's a part of the workflow it says if you can't read it you know tell me you can't read it and then I'll go check it manually and perplexity was the best tool for that job.

52:22So that's where I came up with using the tool iterator here. Sorry, the tool router here. And then I knew, okay, well, I have these four main categories that I cover every day. So I need to come up with a system to basically sort them. And that's where I came up with the sorter step here, and then all of these different iterators. Okay, quick question. What's populating the Google Sheet? Are you doing that? Or is it another agent? so you could absolutely make it another agent for me i love to make my life miserable and i manually search for news every day i have not yet come up with an agent that finds everything that i want in the day so therefore i still have a job um but you know in theory you could come up with three or four different agents and the way that i would do it is i would you know come up with a grok agent to search x.com links.

53:15I would come up with, you know, a Gemini agent to search YouTube. And then I would have, you know, whichever AI can read the most amount of websites. Right now it's probably ChatGPT. So I would have ChatGPT go and read all the other sources that I usually check for AI news and do it that way. And then what I would do is I would have that on a schedule. So I would say, okay every day at 8 a.m run um you know check the all of our usual sources for anything that's published you know in the last 12 hours or eight hours or whatever it is uh and then i would have it then go out find them list them check to make sure that there's no duplicates with anything that's already in the spreadsheet and then fill in the spreadsheet based on all the new stuff And then it could iterate through this 35 at a time or change whatever limit I wanted it to do here to then go through and, you know, based on all of the new items, run it.

54:13Here's a good comment to address, too, from Luke Schwetman. What deters me is having to have paid accounts on multiple platforms to make these systems work. and on that is that is that is definitely a real concern but there are definitely there are also ways around it i saw someone here in the chat i missed who it was uh mentioned open router uh where you can access api connections for many models pay one bill in one place like for using the apis there's not a monthly fee for using the apis it's usage based so if you use it and you're able to go in and like set a budget. You're able to say, listen, not more than$25 a month or whatever the case is.

54:59And it can spend that however you wish. So, you know, there are ways around it. There's absolutely some amount of money that kind of needs to be spent, but there are ways to minimize it. And and what we were talking about with model selection is also very helpful. Yeah. Yeah, the nice thing about OpenRouter 2 is that you can actually set guardrails as well, which is nice. So you can come in here, you can set a guardrail for SOC 2 compliance, like it said. Yeah. Description SOC 2 compliance agent. And then you can save that. And then you can go in and you can define, you know, what API keys can you use, budget policy model and provider access, you know, how to defend against prompt injection and sensitive info detections.

55:54This would be like PII credentials, et cetera. So that's putting guardrails, which we talked about earlier to use inside of your agent workflow. And they're easier, those tools are easier to use than it seems. It's another one of these things in AI that looks intimidating from the outside, but quite often once you get in there, you realize, oh, I can do this. There are times you will be mind-numbingly frustrated, but please know that happens to everyone. Yeah. Yeah. But yeah, so this is an old, this is an automation workflow, right? So this is a recipe. The process works exactly the same every time new data goes into the sheet.

56:29It runs for each of the rows that are new and then it sorts it. It's the same process. When you're working with true agents, there's going to be a step in here where you add what's called an agent node or an agent layer where the agent is going to make decisions and it's going to have a list of tools that it can use and it's going to decide what to call when right now you can give it pretty strict guardrails you can give it pretty strict instructions like hey always use perplexity hey always sort you know according to this prompt hey always write you know this category the same this category the same this category the same etc but um the agent you put the reason you would put that in here if there's a lot of variability in what you're trying to do so if you're getting customer uh like messages right you're you can't build a one-size-fits-all workflow automation for you know anything that your customers might send your way it's just not going to be efficient um so you have to have something in here where you're basically adding a brain to it and really this is kind of like an intelligence layer but this only runs one prompt It can't make decisions and call its own tools.

57:40Now, there is a way that you can do that inside Make, and that's where you create an agent. And that's the new version of this. Yeah. Something that was just said here, and another thing we'll address in a minute, was that the real key to building an agent is less understanding the tool and more figuring out what you want. So you need to sit down. And and I honestly I use graph paper, I mean, like real paper and a real ink pen, which I know makes me sound like I'm one hundred and eight. But I like to think of like, here's what I have. Here's what I want. And then I think about how can I do that?

58:21And more often than not, what I will do is I will take that to, in my case, likely GPT-55 or something and be like, what's the most efficient way for me to build this as an agent in, I don't know, in ChatGPT workspace agents, for example, is where I've been leaning a little bit lately. I hope that gets a broader rollout soon because that's a real difference maker in agent construction, in my opinion. Are you talking about this one that I'm looking at right now? Yeah. Yeah. Yeah, so this is basically an agent node that you would add. And the way that this works is exactly how we defined earlier.

58:59So there's a connection. So in this case, we've already connected it via the API. So we've already set a key and credential. Then we pick the model, like Corey was talking about earlier. You don't want to run the highest one possible for everything. You want to probably run the least expensive, smartest one you can do. So then you would give it instructions to explain. This is like classic prompt stuff. So, you know, you give it a role, a goal, and guardrails there. Then you would take the input. So what is it iterating on? So in this case, you would maybe have a step before the agent. Ooh, I like this.

59:38Let's look at this. You would have a step before the agent where you would maybe have, like, a trigger come in. So if we look at... Discard changes. Thank you. Oh, where did it go? Sorry. So if we look at this entry point when I clicked Create Agents.

1:00:02basically it has all of these nodes up here right so you can have a web hook that starts a node a node is just one of these um basically like one of these little circles in this step it's a step in the workflow basically so let's say here custom mail hook right so this is one node right so this is like one step in the workflow and then you would click create web and then you would connect it to your your gateway let's just save it then you'd have like a little thing here and that it's like an address that you can work with you'd save that then you would connect it to what was that thing i was just looking at that's a good word it's a step it's a step I was like, do you want me to find the shape of the node?

1:00:53Did you want me to? But basically, okay, just to visualize this. So you would have the webhook that is the trigger, right? So you get this from somewhere, something else that you would define. Like maybe you email, in this case, you would email this little address here, right? So this gets an email, then it goes to this agent node. So for the input, you would define whatever you were emailing from this webhook, whatever data was coming here the agent gets it and now that's what goes in input right and then this could be files so you can add files that's part of the context what we talked about earlier and then um there's a conversation id here just to save it i guess so you can look back at it later um and then from there you can add all of these elements to the agents these are things that we talked about earlier so how do i get that to come up yeah okay so you can add context you can add knowledge these are documents files um if you wanted to always when it's working on this task read like your employee handbook or your you know whatever the case may be whatever documents that that are really important like for us in like the news game it's always like style sheets and things like that that are right right for us yeah um then you can add tools so this would be you know functions you know can it can it work with google sheet um you know can it call http and search the web can it work with your gmail etc etc um then you can add an actual server an mcp server which again like we said it's kind of like a connect it's kind of like a connector so in this case there's built-in mcp server so you can connect it to hubspot you could connect it to paypal you You could connect it to any of these, linear, square, whatever you wanted to do.

1:02:43And then you can connect it that way. And then you can also chat with it. But it has errors, so we can't chat with it in this case. And then from there, you can add another step in the workflow, another node. And then that will determine what happens from here. So in this case, you get a webhook, you get mailed this data, right? Now the agent is given instructions for what it's supposed to do with that data. And then it can then take that and it can do another step with it. Maybe it's just. That may just be depositing it somewhere for you, like sending it in an email. Put it in a Google Sheet, put it in a mail, you know, send you a message on Slack, put it in a data store.

1:03:23There's all these built-in features here for what you can do, which is really exciting. And the way that Corey was explaining this earlier, I think is the best way to think about this. don't come in here and try to figure out which of these tools you should connect and and how to put this together on your own what you should really do is you should define the exact thing that you want your agent to help you with or multiple things that you want your agent to help you with then talk with a chatbot or another agent tool like codex or claude co-work or something like that and have it help you define what you should actually connect and have it run through a series of steps, like say, hey, I use, you know, I use Gmail, I use Slack, I have, you know, customers are messaging us on our website chat tool, you know, give it all of the information possible and have it help you basically build the perfect agent with all of the perfect tools to, and, you know, all the context it would need.

1:04:22And, you know, if there's any MCP servers that you should connect to, have it help you think through that. Because sometimes like, like we're, again this is supposed to be for beginners you don't want to get overwhelmed with all this stuff right you just want to know kind of what they are and then when the agent tells you like okay this is what you're going to do you're going to set you know this web hook here you're going to connect it to these tools in this agent you're not like running in blind you kind of know generally what those things are right i'd like to real quick here answer patricia's question about what what do mcp connections give that a solution like zapier can't i what i would say is that zapier solution is just more more rigid for lack of a better way of explaining it you you get more flexibility with an mcp with regard to the ability to have it reason and have the ai decide what tools it wants to use versus you having to pre-define every single element of everything you can have some of that be more in flex with mcp is my understanding you don't have to assign every single tool you just have them all exposed yeah i'm actually going to look at zapier and see how it compares because it's been a minute since i've used it um i'm just setting up an account real fast steven i'm just seeing your uh note about a hypothetical example and i'm not sure where you said that uh if you If you could clarify what you'd like the example on, we'll be back.

1:05:50We'll sort that out. You can do this on Cloud Code and Codex now. Yeah, both of them will do it, and they do it great. Yeah. Top apps for project management. I don't know. Let's just say ClickUp. Okay, continue. I should add both of those apps, Codex and Cloud Code, both, make agents very, very easy. You know, you have an automation section. And so like even if you use ChatGPT, you have access to Codex even if you're not using it. If you use Cloud, you have access to Cloud Code even if you're not using it. It's a different app you would download to likely your desktop. And there are a number of tools it makes easier that you don't get in normal ChatGPT plus in the web interface like Skillet, which you do in Cloud, but you don't in ChatGPT.

1:06:43automation through their plugins the ability to add and create apps real easily don't let those don't let the the names Claude Code and Codex intimidate you away from going and trying the tools the truth is both of them are great for things far beyond coding like I write in Codex most of the time I research in Codex I most of I'd say half of what I do is in there now Yeah, so just to close the loop on the Zapier side of things, Zapier has everything that Maycast, right? It has Zaps, which is basically automated workflows. It has agents, so you can do an agent node as well. Looks like it has a chatbot, so you can do an AI-powered chatbot.

1:07:27MCP tool integrations and even forms. So you could just go to the co-pilot here and say, you can use the dictation tool. Let's do an example together. What would we want to build an automation for? You know, we had someone ask just a few minutes ago if we would show a basic agent for like searching the web, something for folks with really non-technical backgrounds. And I think speaking it into here is probably the best way we can come up with. Like maybe what we want is it. Let's create a basic agent for searching the web. And let me just make sure that's working. Okay, great. What were you going to say, Corey?

1:08:08Maybe it can analyze and, or maybe it can, let's say, analyze and recommend or not recommend certain sources. That can analyze and recommend or not recommend certain sources for the purpose of writing a daily newsletter about AI.

1:08:31All right. So now what's going to happen? This would be like if you were working with Claude or Codex to put the agent together. You would basically do the exact same thing where you define what you want to automate and you would have it. So I'll just say you decide because this is just a demo. So what it does is now it's thinking about how to actually put together the automation based on what I requested. And now what it's doing is this is actually an agent we're working with here because it's using tools. Yeah. Yeah. Basically all of these tools, like whether it's make, whether it's N8N, whether it's Zapier, they are all copying each other.

1:09:12They're all trying to be the same tool because they just want to be the one that you use. Right. So it's important to remember that they might have little nuances that make them easier or harder to use. They might be more your style or less your style, but they are all the same. Like Xavi or more expensive, but also has a gajillion. That's true. They're not the same on top. Yeah.

1:09:42So there are differences. And sometimes it depends on, you know, where you can integrate the app you want to integrate. All right. So now what happens is it sent me to the Zap editor. And what's happening here is it's still working on the left-hand side, but it's now putting the editor together. So instead of doing a tedious manual process of you deciding which nodes you need to string together, you basically give it a goal, and it's going to come up with, as well as constraints. So what applications do you actually use? What sources do you want to limit it to? What sources do you don't want to listen to?

1:10:21etc etc and it can now start putting this together for you yeah yeah ryan patrick says zapier has been around since the days when real developers had to code each integration by literally reading their api docs gosh could you imagine how horrible now you just point the the the ai at the api docs or even the mcp server and say okay told me how to do this um go dude yeah so as you can see here it's now putting this together. It's got a schedule, so it's running every day. It's got an AI by Zapier, and you can see it's using, it's functionally the same. It has the provider. It has the authentication.

1:11:03I guess this is the, looks like we have to put a, that might be like the credentials there. It has the model it's using, has the prompt, and it wrote all this for me. It defined all of these terms. I could come in here and edit this, but it's a good start, right? Yeah. And then it hasn't connected any tools yet per se. It says, oh, it needs to authenticate the AI step first. Okay, so here's where I would come in and if I was actually going to use this, I would sign up with Zapier and put in my credit card information, get my API key, et cetera, et cetera. I'm not going to do that, but this is just an example of how you can go from zero to here.

1:11:46Pretty sure you can do the exact same thing in Make. So if I was going to start from scratch on Make,

1:11:57you could probably start from a chat.

1:12:05Maybe not. They have a lot of pre-built libraries you can start with, I guess. That's one way you can do it. So, for example, there's all of these custom workflows in here. Where the workflow's already built. You just go in and connect the dots. Yeah. So you would find the thing that's the closest to what you're trying to do and then change it. Yeah. Like content draft creator, maybe that's close, things like that. And the other one that people use that's very similar to this is N8N. Yeah. So that's this tool. And it's basically the same thing. Where do they have the workflows on here? I feel like they kind of hide them.

1:12:45Oh, templates. Here we go. Ooh, here's a question, Grant. Yeah. How many things do I need to pay for to get this to work? Great question. Great question. So let's say you are wanting to spend as little as possible. And it very much depends on what you want to do as well, I would add, real quick. Yeah, yeah, that's true. if you want to spend as little as possible what would you recommend Corey? I would think ChatGPT Codex would do most of this. I would do Codex absolutely because you can do it with your$20 subscription frankly. You can do it better with a$100 subscription but you could do it with that and honestly it would do just about anything you need as far as I know.

1:13:34I mean I can't think off the top of my head of like oh you're not going to be able to connect to this or that. I mean, if you want to run multiple companies' models, if you're like, I want ChatGPT to do this stage, I want the other to do this stage, I feel like there's a – that's the only place you're really going to come up short with built-in codecs or built-in to quad code. Yeah, because basically what's happening is all of those other connectors that we were kind of looking at of the other applications, whether it's linear, whether it's Canva, what have you, they're all building so you can basically run them inside Codex or Cloud Code.

1:14:13They're trying to make themselves as easy for people to use where people actually want to use them. And so in this case, you wouldn't have to necessarily string together the entire automation in a separate platform. You could just go to this thing here, automations on codex, and you could then, in theory, have it do everything that Zapier was doing for you, but inside codex, and then you can run it. and if you are curious about how you might do that there's you know we did it we did a full automation where we talk like okay setup sorry we did a full stream where we did everything from what's a workflow you do every day all the way to scheduling it and running it as an automation so you definitely check out the ai for total beginner stream that we did where it shows you exactly how to basically do that in codex if that makes sense yeah yeah it's been a minute since we've done one of those we'll run through those maybe we'll do that again one day soon because codex and cloud code update about every third day now so uh or at least every thursday i just heard a new codex drop this afternoon too right as uh oh my gosh really yeah yeah the number and supposedly new features or something i'm not sure how big of an update it is it may have just been a normal update but the one tweet that popped up on my phone seemed very excited about both of them having these things today so i don't see anything yet but maybe it's on x maybe yeah it may just be an x thing uh okay well now that we've gone through make oh sorry no i was gonna say i don't know you know it seems like we lost a good amount of people in that in that example so um i would like to show people some more practical examples of how we're using these tools today.

1:16:12Yeah. Well, one I'd like to show real quick, because it's also very beginner friendly, and I'm not going to spend a lot of time in here, is that I don't know if you use project management tools where you work, but we use ClickUp here at work. And we've recently gotten access to all of their AI tools. And I've had a lot of fun, like, tinkering with these and seeing what there is and uh it's really simple you'll just click the ai bar over on the left and you can just come in here and tell it like hey i want an agent to check my list swat runway for any incomplete tasks once a day and people don't know is a workplace management tool so it's like where you have all of your basically the way we use it at our company is we have all the work that needs to get done we track it we monitor it we you know track its progress and um yeah yeah market and and the nice thing is like every i realize every tool on earth right now has integrated ai into their app uh but this is one of frankly quite few where i feel like they really did it very well into a traditional app like it solves a number of actual problems i had in there uh so it's going to come back with a quick question or two a simple note like you have x incomplete tasks would you like a full list of the tasks with links i'm going to tell uh uh yeah and it gives you you know pre-recommended things to hit here so i'm going to hit that one which is give me a full list with task names and stuff and it's gonna it goes through here and it builds it while you wait or while you go i also like this like a coffee dog out this digital biometric scan happening on the right hand side yeah yeah yeah plus one for the graphics for sure because that's that's a really cool thing that suckered me at what time do i want it uh 8 a.m it doesn't matter i think the very cool thing that all of these platforms are kind of demonstrating is that um gone are the days of you having to manually go in click and sort and do all this stuff um pretty much every platform now lets you just via a conversation set up these types of agents and these automations like yes i want to comment here on is it w double uh it made a comment i to be fair this is not a thing i take lightly complimenting the inclusion of ai into a normal SaaS tool.

1:18:53You should know I spoke about it. You were very anti-Abit for a while. Oh, I was on a tirade for a year and a half because the instant the word agent ran past its first marketing team, every tool threw it into their apps and were doing these things that were absolutely not agents over and over and over and calling them agents and really watered it down. But I'd like to say that I think three things changed that. I think Claude Code, Codex, and open claw i as a as a group really really changed what they are yeah only half joking i get it uh but yeah this one i really do like like check here you know it's done uh it's built it it's given it a bizarre picture i'm not maybe a fan of some 3d glasses over here or a mardi gras mask i'm not sure uh but it goes through here it gives it its rolling objectives capabilities and scope gives it instructions of what it's supposed to do uh calls out edge cases that it should be on the lookout for tone and personality context mentions a variety of different triggers but you can then come in here and just instantly hit uh run and turn it on here we go yeah and it'll it'll start up a private chat with the agent and be like all right hey let's get started so now it's just going to run and that's that uh it may take it a few but ideally the thing to get used to is i'd say that you can um you don't have to sit and stare at these things you're not gaining time if you sit here and watch water boil absolutely when your agents are doing things go work on something else or take the dog out see some sunlight do something nice and like i regularly I think yesterday afternoon at one point I had four chat GPT pro windows running with various tasks and pulling reports and some other things that were really interesting.

1:20:50So, you know, there's a lot you can do and a lot of different ways you can do it. Grant, are you still here? You look frozen. Oh, no, you're not. I was catching up on the chat. OK, cool, cool, cool.

1:21:07One basic tool. Yeah, I was just reading up everything up until now because I was in the zone with that make example. Good feedback here, too. Suggest you ask for real-world examples that beginners could easily accomplish with one basic tool. Give us an example. For the beginner path of your sessions. I'm basic beginner stage, and this is leaving me over a while. Okay, that's fair because we've been jumping around a little bit. I think the hard thing about this, and we're actually working on something where we're going to break down how to use each individual tool from a beginner point of view for that tool.

1:21:46So this is kind of like a broad overview you can look at it as, which is why we're not trying to go too deep into any one tool or any one use case. We're just trying to get people familiar with it. but the best way you can look at it is start with what tool you actually already use. So Beth, for you, for example, do you use ChatGPT or do you use Cloud? Or do you use any of the other tools we've been talking about today already? Or all of them. Yeah. Grant and I tend to use several. Well, the idea is I'm trying to narrow down the focus to one tool. Yeah. yeah yeah i guess the question is if we were going to do an example that was really beginner focused honestly i think it would probably be in make i i i would rather it be in chat gpt's workspace agents but it is only available to the business crowd right now yeah uh unless we did one in codex codex is probably as easy a place as any walk through the how to download and show them how to build an automation yeah would you would you be comfortable doing that on your side

1:22:56what i need to know what i need to make sure of is that it's updated yeah let me run my update why don't you go check that in the meantime so beth said i'm already using so first beth said pick the free level of a tool so here's the the tricky part about agents right there is really no free agent unless you're going local so um you can use what's called a local AI model for free on your computer with something like a tool called LM Studio, which I'm going to share in the chat. Another one, I believe Lumo lets you use local AI tools for free. You might have to pay for an account for certain access, but I'm going to share both of these in the chat one sec so lumo's from proton it's privacy focus okay free ish the closest free ish you're gonna get is probably um you're using clawed code and have a working app um but i was once a beginner and overwhelmed that's fair that's fair um the the closest to free you're gonna get is codex so i think it would be helpful to show this in codex um it's 20 it's not free free but it's the most uh bang for your buck so to speak supposedly it's the best best deal in ai what i've heard for ages is that the 20 chat gpt plan will get you more mileage than just about any other one yeah i mean if you're willing to go up to the hundred dollar tier for claude that's probably or or codex you get more for that yeah but um for just what you get for 20 a month on codex is pretty pretty good yeah it is it is let me uh okay yeah there was also a question in the chat about hermes which is really exciting and i was actually playing around with hermes um recently so i could talk to that but someone mentioned mercury so i sure did mercury is a model that um you you know without getting two in the weeds because this is for beginners um that's very cheap um because it It uses a different process than the regular chat models do.

1:25:09And has like no latency. Yeah, it's very cheap and very fast. So Mercury is a good model to use with a more open tool. Like for instance, if you're on, is Mercury on open router? Cory? Mercury's on open router, I want to say. Mercury's native in OpenClaw. It's in, you can just come back and forth. Yeah. I use it. it runs most of my open claw, honestly. Yeah. Let me show you this. What I would say is they compared it to like a Sonnet. I want to say a Sonnet four or five or something or four. God, I can't remember the four dot what, but to that and I think chat CPT mini, it's kind of a midsize model, upper midsize, but it's good.

1:25:59It's fast. 90 % of what I need it is plenty model for. There are definitely things that I'll save like a 5.5 Pro or something for, but the idea is that they're very expensive and I want to use it as little as possible, but make sure that where I am using it really counts. Yeah. So Mercury, for example, as you can see here on OpenRouter, it's very cheap. It's$0.25 per 1 million tokens. So that's roughly 750 ,000 words, maybe a little bit less, a little bit more. But then 75 per 1 million output. So input and output, you can think of input as what you put into the model. Output is what the model spits out back to you.

1:26:46Context is how much it can look at and think about at one time. So this is like 128K tokens. So that's maybe like 100K words, right? I feel like we might have gone down another more complex rabbit hole. Okay. Well, when you're ready with codex, let me know. I was just in thinking about it. I was like, it is. And then I was, we could do a thing on that once. I hope there's interesting news out of them next week. We'll see. But for example, you can see here, like you can compare Mercury to GPT-5. This is on OpenRouter.com. You can see, you know, how much context you get, how much it costs, right?

1:27:26So if you were going to run it in an agent, in an open agent, let's say, you would probably want something more like Mercury than this. Yeah, it's the difference in I have, thanks to caching, my open claw costs me less than$40 a month using chat GPT. It was in the couple hundred or so range. I do not know how people are doing it with clawed API tokens. and still paying thousands of dollars a month. Thousands of dollars a month. Yeah. So, you know, the economics of this is really important if you are like us and don't just, you know, roll in money. Yeah. Now, I would say, well, while you're getting, or if you're ready to go on Codex, let me know, but I'll just share one more thing.

1:28:13I do. I had a fun question back a minute. I'd love to just address for a second. Go for it. Do you find your intellect and cognitive skills being enhanced since working with AI tools or are you finding your building using new intelligences? It's funny to me that this framing doesn't include the opposite. Like if you feel like you're getting dumber using AI. You know, and you see the headlines about people becoming dumber. But like I knew zero about software engineering. I was not, you know, I've been learning all kinds of things. And I would say that if you're using AI as a tool to broaden your horizons, there's not a limit to how broad you can go.

1:28:55But if you use it lazily, you will probably experience the opposite. I think it magnifies both of those. It can magnify the desire to learn into much greater knowledge, but it can also squash the desire to do nothing into doing way, way less. I think it depends on who you are. yeah edsy trone famous ai hater had like a a good line about this the other day but i'll spare i'll spare you corey does not like him um he's all right not for me so um look at it like collaborating with a smart friend we both gained from the interaction i think that's right i think you have to think of it as don't outsource your thinking entirely to the agent use the agent as a thinking partner to enhance your thinking like someone.

1:29:50Yes, 100%. Say like, here's what I'm thinking, react to this, and then get its reaction, process what it says, and then go back. Exactly. It's a back and forth. Have a conversation like you would if it was someone else. I realize that may seem a little weird, but the truth is I get a lot out of it that way from an intellectual standpoint especially. So here is Codex. when you come in there's a couple settings it's going to make you do that are going to seem a little creepy at first like it's going to ask you about you know like a source folder what you want just tell it to create you a folder called whatever you want in this one because this is my work computer my personal computer I have just taken the the all the guardrails off and told it just go do things but on this one I am much more cautious so it's got a folder that it can access And in there is where you do everything.

1:30:43And that's where just your normal chats are. You go in here and it'll go back to one you did before. It's going to take a minute to load because this little computer I'm on right now is, aside from running a live stream, doing an awful lot of other things. All good. All good. But here's the tool you're looking for. When we're talking about agents and what you could do with your$20 chat GPT subscription right now, this is where it's at. You have automations right here already inside codecs that you can use. It's loading up. So it's got a couple of examples here. You can go up here into view templates, but the truth is most of these are a little more developer focused and may not be.

1:31:28Thank you. I'm asking about cognitive skills because I'm 77, a beginner that really likes using it. I like that framing as a thought partner. Yeah. Heck yeah. That is killer. so what I would say here is like most of these aren't going to be that useful if you're not a developer that these that they have but it has right here the ability to create via chat so you can say I want to set up an automation I want Codex to search the web once per day. Find three relevant AI news articles. Hang on. We're going to word that differently. News articles about the AI industry.

1:32:24Write a brief analysis

1:32:32and send them to me in an email

1:32:39at cnols1980gmail.com default permissions you don't want to hit full access unless you're willing to accept the risk that comes with that you choose your way could you explain that in the context we talked about earlier so guardrails guardrails good call good call and this is really important full access here in codex means it can use your whole computer it can create things it can delete things it can go into any folder you let it there's just there's a ton of things it can do now and it won't ask you for permission first and it won't ask you for permission first that unlocks a lot of amazing things but it also unlocks a tremendous amount of risk that i would never, ever, ever do on a work computer on like your only small business machine or on something like that.

1:33:28But, you know, if you have a have a burner, it's fun. But for the most part, default permissions is probably where you want to be. One thing that's different in codex is over here is where you select your model. And, you know, you've got options of all of these recent ones that are great. But up here, you can select how much reasoning you want it to spend. Do you want low reasoning, medium, high, extra high. Each of these uses a little more of your token quota over the course of the month. And you can also go fast. Whoa, you don't want fast because it's incredibly expensive. Like I want to say fast is it's the one that uses the Cerebris chips.

1:34:08So we're going to go high on 5.5. We don't need anything over here, but you can put it in plan mode, which is like plan the agent. And that's what we're going to do with this is let it plan our agent and then we'll have it after we know where we want it uh where we where we are now you've got plugins you can pursue a goal but we're not going to get into that today oh we will we will i want to touch on that i'll handle that part though okay i didn't know how relevant it was to this so i wasn't sure very relevant very relevant it's a pretty cool feature though yeah i'll i'll touch on it to the extent people need to know about it.

1:34:46Okay. Okay. So, well, this is going to work here for a few. The nice thing about Codex is, you know, I can just go do something else. I can go back over here to plugins. I can come, oh, wait, where am I? Yeah. You go to plugins. You also have access to skill files of which I don't have many loaded in this account, but it's all here. You can go around. You can work on other tasks as you go. And when it's done, it'll pop up a notification and be like, hey, this thing is done. Oh, hang on. It's done. It's awaiting a response. How should it be delivered? Gmail. 8 a.m. Okay. What should it count as most for a business impact, technical breakthroughs, broad mix?

1:35:37Okay. So now it's going to work on it some more. and it'll be back. Grant, do you want to go ahead and talk about that while we're in here? I can show you. Yeah. Yeah. So basically what Corey's doing now is what's called plan mode. And in this case, to level up what we're talking about here, so Corey is using Codex as an agent. And what he's doing is he's defining essentially the subagent that he wants Codex to build for him and how that's going to work. And in order to make sure that the agent thinks through, you know, everything that that means before it just goes off and tries to write it itself or implement it, it makes a plan to make sure that it's thought through all of the edge cases and all of the components of how it's going to work.

1:36:25You can kind of think of it as like using thinking mode or reasoning to go through the plan. Then once that plan for how the agent is going to work is set up, you could use what's called goal mode to actually set the agent off towards accomplishing a goal. But in order for that to work, you need to define what the goal accomplished looks like. And then basically the agent will run until it accomplishes that goal. So a practical example of this would be, I want you to go out and search the web for everything that has been published in the past 24 hours. And the goal is not everything. Everything in XYZ industry has been published in the last 24 hours.

1:37:08And don't come back until you've captured every, you know, to me, done is every news article related to AI that has been published in the last 24 hours is in a spreadsheet. And that's what done looks like. you can go much more in depth than that, but that would be essentially the goal. And then it will just run until it accomplishes that, or it believes it accomplished that, for example. So that's important for, you know, setting an agent once you've created the agent to then go off and do things for you and actually get them done. Yes, it is. And I honestly, I haven't played with it much yet, but I, I want to.

1:37:46The truth is I've been a little fearful of its impact on my limits. That's very real. As often as they reset them over there, like I think Codex crossed 5 million already. Every million, they just reset the rate limits and it's happening every, I think, couple weeks right now still. So every million new users. Oh, automation created successfully, pulling it back up once to verify the saved schedule. Oh, okay. Here we go. Patricia has a really good point. So when we're talking about free, free-ish AI or the$20 tier, and we're using Codex with the ChatTBT subscription, what that means is that we are using the tier of usage that ChatTBT gives us access to for paying$20 a month, which means we're not actively managing tokens.

1:38:40So remember earlier when I was showing the comparison and price between Mercury and OpenAI? Well, basically, that's if you're using the API directly. So if you were using the API directly in Make or in Zapier or in any other agent tool, you would be paying OpenAI on a per token basis or any other tool, Mercury, whoever, on a per token basis. When you use a subscription inside of Codex like we're doing right now, you're not actually paying for, you know, you're just paying the$20 tier and you have limits to how much you can use. Anyone who's used Claude knows about limits. Oh, yeah. But you have a certain threshold with which, you know, they say, okay, it's not worth it for us to let you use it anymore because you've only paid$20.

1:39:25But it's a much more cost efficient way to use the AI and experiment with this stuff. but you might get what's called rate limited where you've been run out and you have to wait 24 hours before you can use it again or something like that. So, yeah. Yeah. And the other thing to know about that is, you know, we, we, we shared some terms today that, that I realize are a little, a little deeper than maybe most beginners are comfortable with knowing. But part of the reason that matters is because like the thing to know about APIs is even when you're in Claude or chat GPT, you're using their APIs to communicate.

1:40:04You just don't see it. It's all happening behind the scenes instead of you wiring it up. So understanding the limitations and capabilities of those things, I think, is really important to understanding agents. So that's why we didn't go super in depth with them, but we wanted you to just get a definition and kind of a loose understanding. But how does this differ from the Cloudflare worker process? Honestly, I have never messed with the Cloudflare tools. So Cloudflare worker is like a server, run this code when something happens. Whereas Codex is AI, go to this recurring thinking coding task for me.

1:40:49So I think the difference is like there's not a server involved is the way I would think of it um well I guess they are technically serverless um you know they run deployed javascript typescript whatever um yeah they're very similar I would say I'm not educated enough on on cloudflare workers to answer that question but you have aroused my curiosity well we'll get an answer we'll get a strong answer for you with the write-up that goes with this but yeah i would just say that basically it's like uh the agent running you know an automation is open ai you know putting compute to the process and then the agent goes off and uses tools and does what it wants to do and a cloudflare worker is a bit more like deterministic i would think that makes sense runs reliable like reliable execution so that would be similar to you know a more deterministic workflow automation that we had before.

1:41:52Very cool. Grant, do we have, by the way, the automation appears to have created. It has, and it's been down here running for about a minute. The truth is I don't know that the email account it needs access to to send the email. I don't know that it has access to that email. This may stall out. However, I just wanted to show that that's all there is to it. It's nothing fancy. Now, you can also go in here and, like, I would cut this down to 5.2 because it's just reading the darn news. It can do that. I would keep reasoning at medium. Codex medium is pretty good. I use high sometimes, but I rarely touch extra high.

1:42:36I only save extra high for, like, really deep thought. Like, the thing I was doing yesterday with, like, assessing videos, Grant. That was, like, I want extra high. using Cloudflare workers to replace Zapier and their main integrations to eliminate tasks. God, I'll bet it eliminates a lot of it. Yeah, I would definitely use it in that context. N8N, you know, you could also use N8N to replace Zapier costs because they charge you per automation. I think MAKE does that as well. That's not a practical... I mean, I guess it's not that different from charging you per token, I guess, usage-based, but I don't know.

1:43:14Not my ideal form of billing. Now that we've been through a number of these and we're down to about 15 minutes, are there any questions we've missed? Do you have any outstanding questions? This would probably be a good time to run through those. It would also be a good time if you haven't yet to go reach up above and click the subscribe button and subscribe to our channel. We really appreciate it and appreciate you being here and love having all of you as part of the community. Also, pop by the Neuron.ai and sign up for our daily newsletter and join a whole bunch of people who read it every morning.

1:43:47We'd love for you to be one of them. What have I missed? We've missed questions. I know we have. Okay, so one that just came up, Love and Lie 42. This doesn't allow protection of your intellectual property, though, does it? So you need to have a business agreement, a BAA with OpenAI. So you need to at least be on their business to your account. We talked about this a bit more in the workspace agent stream that we did. But basically, you can form a business account with just one other person. so if you have someone who you work with at your business or if you just want to go in with a buddy and form a business in order to get the business account you can do it that way and then your intellectual property is supposedly as much as OpenAI can be trusted on this yours and they don't train on it etc etc this is a thing that we have Ryan, I appreciate you calling this out Twitter spaces are a thing I have thought about doing a ton of times We have a Neuron account.

1:44:46I have my own account. Grant has his own account as well. And I have thought before about going on. But the truth is my following on Twitter is quite small.

1:44:59So I've never felt like there would be anyone that would care. If that is a thing people would care about, we would absolutely love to do that. Make sure you can hit us up at I'm at Corey Knowles. C-O-R-E-Y-N-O-L-E-S. Yeah, you should be good. You should be good, loving life. Just responding to his comment, he said, got the business, or they said, got the business, don't have a buddy.

1:45:26Okay. Someone asked about Hermes. I would like to talk a little bit, but this is going to be a bit more in the weeds. So I'll just do a very broad overview about managed agents and Hermes. But let's answer any other questions people have in case that's too complicated and people want to jump off yeah i'm scrolling back to see just if there's anything close we missed yeah yeah oh yes chris russo has a good comment for you love a knife turn off the improved model for everyone um which does allow your content to be used to train our models yeah turn that off we show how to do that in our um ai for total beginner stream so if you have not watch that why don't I link that in the chat?

1:46:09Cause I think that would be helpful. You can always just show how to do it even too. You know what? I can just do that. 77 year old has already subscribed. In fact, introduced my millennial son. I'm a long time nerd. I'm going to spend more time on this to become good. I am so glad to hear that. That is absolutely awesome. You know, we're, we're not kids either, but we're, we're, we're not 77, but we're not kids. And it's been a, fun journey that I never knew I would take in life. So it's been exciting. It's data controls. You know what? It's right here. See this? Improve the model for everyone.

1:46:52Under data controls, so you click personalization. You click your little profile in the bottom left corner. You go to personalization, data controls, and improve the model for everyone. You make sure that this is turned it off but if you have a chat gpt business it automatically disables training yeah yeah but the business account it's automatically shut off because they won't train your data the truth is on my personal one i think i've got it on still i don't really care i uh i feel like like like like i'm doing my part maybe so you are the reason that um chat gpt 5.5 really loved goblins i am i am the reason it's great

1:47:34i like to think that it was following the things i do that's hilarious um for folks who haven't watched it yet and i for total beginner stream check that out right 15 000 views and counting um grant somebody asked our thoughts on manis and i was curious i think you have used it more than i have am i right or am i crazy so we just talked to jen spark who is sort of like a manis competitor and uh you you could use either of those interchangeably they're probably easier to use agents than a lot of tools right now um why don't i go ahead and show you bring one us manis i like that you two always scroll past the the sponsored one and clear yeah don't click that even if it's a segment like i'm not gonna get make them charge money for a search i was already doing so manis for example this is you know obviously most most ai applications start with this chat window right now um you know you can do all sorts of things here you can do schedule tasks you can develop apps you can create spreadsheets visualization my only my only reason i don't recommend these um as much as we could is because you're basically paying on top of cloud or open ai so you might as well just go to cloud or open ai i feel the same way about copilot to be honest like i just think like if you're already paying for these models so like just use the the codex or cloud code and yes it can do all of those things and it and if it's useful for you um you know to not have to think very hard about it and just like come in here and They're like, yeah, create me some slides for, you know, whatever photorealistic, you know, you have to sign in.

1:49:23The thing I do like about GoPilot is that you get all of those for the one subscription. Yeah, which is nice. It's not a separate subscription. Yeah, yeah, that's true. That's true. It's more cost efficient if you're looking at it that way. My point with this is these are very powerful tools and they can do a lot of this stuff, but it's the same stuff that Cloud Code or ChatGBT Codex can do because they're built on top of it. So it's up to you. If you want something that's maybe a little bit more intuitive and easier to use than Maker Zapier, you could go with Manus or the one that we just we interviewed their COO recently.

1:49:58And it's a very cool tool. GenSpark is an alternative. You can do that.

1:50:06GenSpark is very similar here. It shows you all the things you can do. Got the office suite. You can create AI slides, sheets, docs, design code. it's a pretty good setup too you can even create your own uh open claw agent which you know it's maybe one of the easier ways to do that to be honest um it is it is speaking of uh we were we were asked at some point i don't know if the person is still here or not um oh hang on we do have a question here we ought to address first i've informed i've informed i do not use any personal information of anyone when i used ai is it still true since we are allowing agents to access our data i would say if by informed you mean you clicked the box and said no it's probably good but i would say that if you mean you like told it to in a chat it is probably not yeah here's the thing like i've heard a number of people who said that that's how they did it.

1:51:13Yeah, I would not trust the AI to do things just from asking it. I would not trust the AI to manipulate its own user interface, if that makes sense. Yeah. I think you have to physically go in and do that. Now, it can add memories. It can do things like that. But just because you add a memory, like, don't remember this. Yeah. It's not necessarily going to work. You need the hardcore, like, you know, yes, I'm 100 % confirmed that you are not training on my data. Yeah. Patricia Wilson said GenSpark is incredible. I love GenSpark. A lot of people like them. You know, I just go straight to the source with Cloud Code and Codex.

1:51:51But, you know, those are great tools as well. You know, very, very, like, they're basically building a harness, which we talked about this very, very briefly at the beginning. they're building a harness around the ai models and it's an alternative harness to like codex or cloud co-work um that helps it make um control the agent for you and gen spark i know for a fact because we just talked to their coo uses multiple models under the hood right so you're not just using cloud code sort of like the co-pilot version where you're using as many models as you know they have access to local models etc etc so um by the way one day next week uh i'm gonna be at microsoft build in San Francisco from Monday to Wednesday and one day next week someone from the I think the VS Code team is going to do a live with me it looks like from there you know kind of going through and showing how to start vibe coding using if you want to use tools like VS Code and Node.js I know there's your maybe sound a little more complex but it's it's it's a very doable approach and it's honestly it's the way that i would say normal engineers coded for many years but this is doing it with ai which makes it very simple and easy and he's going to come on and answer a lot of questions help us through anything we don't understand um some other stuff but i don't know who or what else is going on yet but i will i will be there if any of you are in san francisco want to grab a cup of coffee by the way hit me up i'm reading i'm reading ryan patrick's comments In the end of 2024, I had my Amazon SES key stolen from me like a week after giving it to a modeling cursor.

1:53:32Woke up 100K emails sent for my domain. Yeah, good rule of thumb. Don't paste your API keys directly in the chat. Ask the AI to direct you to the file or the credential manager where you're going to place them and place them there.

1:53:52Okay, I know we have six minutes left. I was going to talk about Hermes. I was going to talk about managed agents. I think maybe we got to do AI agents for total beginners part two to cover all that. Like maybe we do just a, yeah, maybe we do an all the claw kind of like all the claws. Yeah. Like just to do like a full claw. Yeah. That'll probably be the biggest. Including, including Hermes agent. But yeah, all of, all of the, the mountain of things that open claw has spawned into existence since January. I'm going to share two things real fast. So if you have not heard of Hermes, so I actually think Hermes is the new open claw.

1:54:37Okay, I think it's easier to use than open claw. I think it's, you can kind of think of open claw as the open AI of agents. And Hermes is like the clawed of agents where it's made by a smaller team. They have more taste that they put into the model where it's open clause, quite literally very open, right? You can do anything you want to it. Hermes, I think they have a lot more of their own personal sensibilities that they put into it. And this is an open agent model. So, you know, this is definitely not for beginners, but if you wanted to create your own agent that you could customize and work with any model you want, this is a really cool tool for it.

1:55:16And it's probably the cutting edge of these tools, I would say at the moment. Why is it at the cutting edge? Because basically you can I don't know if it's going to work. Let me see if I can. OK, I'm just I will pull up my screen on that in a minute. But because it it basically automatically continuously updates itself based on what it learns when it has you do a task. So let's say you have it go off, build your task, comes back with the result of that. You said, hey, I don't like it or, hey, you know, this or this didn't work for me. Then it will automatically do that. So these are things that we taught you how to do in our AI for Total Beginner stream with skills, but it kind of automatically manages the skills in the background for you.

1:56:00Yeah, and I want to say OpenClaw already had the ability to do that, but it was more of an add-on sort of thing. My understanding is Erme's made it simpler. Yes, they made it simpler. It's a little bit more abstracted away. And I think because it's not, I don't think anyone can push updates to it. I'm not sure. I think they're a little more strict. It just doesn't have as much bloat as Open Claw has now. And Open Claw seemingly breaks every time there's an update. So they're working on it. Like, I don't rule out Open Claw. But yeah, I would say Hermes is sort of like the claw to Open Claw's open AI at the moment.

1:56:41Yeah. All right. What did I, okay. The second thing I wanted to show real quick before we go. So let's say you, it's amazing. So let's say you wanted to not just have an agent that you run for yourself, but you actually wanted to have an agent that you run for customers, right? So I would say manage agents is one of the most advanced ways to do that. It's also one of the more expensive. So I guess Hermes? Yes, I'm about to switch. Oh, okay. Yeah. Managed agents are one of the more advanced ways to do that. And basically the way that this works is, let me mute this. The idea being you could produce agents and sell other people access to them.

1:57:26Yes, exactly. And I think that you would want to use this most importantly. This is kind of messy so that they're showing, this is like the dashboard for managed agents. But basically the way that this works is it creates a sandbox that the agent can run in, and then you can create a front-end application. Or you have your own website, and your customers can interact with the managed agents in the cloud. And you can give it credentials. So to Ryan's point, where Ryan accidentally gave his keys to the agent, they have this thing called, if I can remember it. Well, basically, they have a credential manager and a memory manager, and the credential manager holds the credentials on cloud.

1:58:12It holds the API server, so they can access those depending on the permissions that you give. And this would be more equivalent to, at least from my understanding, what a cloud worker would be like. But it's very easy to set up, even though this looks very complicated. oh man there's all this like code and stuff well really all you do is you go to console um you know go to the anthropic console and they have a quick start guide there and i will link that in the chat AWS secret manager, nice. but yeah i think we're going to do a whole stream we'll do a whole stream about clause and we'll do a whole stream about manage agents because we don't have time to do today is not necessarily the most beginner friendly but i'll just go ahead and point that out as another resource that you can use now i don't know what open ai is equivalent to this is yet i don't know if they have one yet but I imagine something like this is coming.

1:59:15So, oh yeah. Yeah. And I imagine you can expect chat GPT's workspace agents to roll out across chat GPT at some point. I mean, a lot of times the reason they'll drop these out to smaller groups in like the pro tiers, partially because people are playing enough that they want, they want a little something for their money, which is to touch the new toys first sometimes. But the truth is it also puts them in the hand of a smaller group of people so they can assess how this thing is used, which is what they've done with chat CPT finance is put it out there so they can see how do people use this? What can we expect the average user to, to cost us with this if they have it and, and figure out where it can fit into other plans as well.

1:59:59But that's it for today. We're at two o 'clock folks, everyone. Thank you so much. This has been a lot of fun, a lot of great conversation today. I can't believe as many people stayed on to the end. That is amazing. again be back next week with something from Microsoft and probably another one later in the week we'll also have we had a great video drop yesterday I don't know if you've seen it yet maybe Grant could drop a link in the chat real quick with Jeff Shainline of GreatSkyAI what they are doing is building a brain inspired computer interface so instead of just a neural network AI they're building a computer inspired by the brain and it's really really interesting and worth a look so uh we hope you watch that take the time subscribe do all the fun things uh and see you back next time quote me on this in five years this or something like this will be the new thing that everyone is talking about like yeah i think so i i think like what gpus were in like the 90s to today like this type of architecture or something like this i'm not saying this is going to be the one, but something in this field like analog computing, neuromorphic computing, like it's going to be a very big deal.

2:01:16AI has accelerated the development of quantum quite a bit too, even. Quantum has some issues, but it's really sped that up quite a bit. Yeah. Like we said, we said in the, in the video kind of jokingly, we were like, yeah, this kind of takes all of the good things of quantum and sort of all the negatives. So I I think probably something more like this will be used more broadly than a quantum computer even five years from now. But we'll see. Yep. Good to see everyone. Yeah. We sure appreciate you. And we will see you back here next time. Farewell for now, humans. Bye-bye.