AI Agents From Corporate Productivity to Playful Social Engagement

Download MP3

Dalton Anderson (00:01.506)
Welcome to Venture Step Podcasts, where we discuss entrepreneurship, industry trends, and the occasional book review. From generating podcast outlines to crafting social media posts, AI agents are becoming increasingly more capable. Today, we're going to be discussing Google's new release of their Gemini Gems, which is their version of an AI agent. And there are some that come out of the box from Google.

Like I think the brainstormer writing assistant, learning helper, and I think one more, but you can create your own gems. And that's what we're going to be discussing today. And in fact, I did create a duplicate basically of the Facebook AI agent that we created a little bit ago called Curio. So I have a Curio meta AI studio agent, and then I have a Curio

Google Gemini agent.

Which is going to be interesting because I'll break down the differences between the two and you can figure out which route you want to take. But in today's episode, we're going to be discussing the differences between Google Geminis GEMS approach and Meta AI Studios. We'll do a comparison of the features and potential capabilities in the future and what it currently does and how I think it's suited to which user.

And then my personal experiences with these AI agents and using them on both platforms and what do I think? And just...

Dalton Anderson (01:46.402)
the implications of the future of where, you know, I think these things will go and how I, how I'll use them and what I find interesting about the agents. But of course, before we dive in, I'm your host Dalton Anderson. My background is a bit of mix in programming, data science, insurance offline. You can find me running, building my side business or loss in a good book. You can listen to this podcast.

in video or audio format on YouTube. If audio is more your thing, you can listen to the podcasts on Apple podcasts, Spotify and YouTube or wherever else you get your podcasts. Okay. That's all squared away. Now let's get started into the content.

So there is obviously a growing popularity of these LLM models. You hear about it all the time on the news, in the workplace, water cooler conversations. Your kids are asking about it. They're integrated into phones now. Like Apple users have access to AI agents natively on their phone. The new ones, Google's AI capabilities on their phone are

are right there with apples neck and neck and so is Samsung. And it seems like everywhere you go, every company is talking about AI, AI, AI. What does AI offer to you? That that's a question you should be asking yourself. And how how does it easy for you to apply AI in your life or in your workplace or wherever you want on your sports team or your Dungeon and Dragons?

spot. And so there's, think there's a couple of different types of AI, like there's social AI, which I'm less fond of. Like if you go on Meta AI studio, the most popular, the most popular AI agents are like, talk to your lonely, lonely, depressed ghost boyfriend or like, you know,

Dalton Anderson (04:01.464)
talk with your vampire girlfriend who has trouble sleeping or something, something weird like that. And like some, some teenager love novel type of character build. And those are the most popular chats for Metastudio. Like those are the ones that are really taken off. They have like, when I was looking at it, when it just came out, it had like 60 ,000 chats and they're by far the most popular ones.

mine about today's fun facts is like, I think 30 chats and they're all my chats. So I don't know if anyone's used it besides myself because it doesn't show you like unique users. It just tells you that people use like there's chats on it. So I I'm not sure who, who used it. I'm pretty sure it's just me. That's an example of metas approach. And I would think that like meta is

a bit more social oriented obviously than Google's approach, which is more.

workflow management slash like automation ish. Whereas Google is just, it's cool. It's a way to help you create or help manage your platform or I randomly got the hiccups mid episode. Random episode hiccups is the best.

Dalton Anderson (05:36.918)
Meta AI studios approach is more social as I was saying, and that's where they're going to lean into, which, makes the most sense. Like Meta is just not going to roll out like a workflow management platform or they tried to, I think, I think they had a platform that was like Facebook, it was called like Facebook connect or something workplace connect. And the premise of it was like,

Facebook, but in a workplace manner to where you had like workplace events and groups and all sorts of stuff. And it was social media for your workplace, which seemed decent enough, but they, they dropped it to focus on AI. And I think they had some, some adoption there, but they, dropped it off. Anyways, it's not even related to what I'm talking about. So let's, I'm going to stop myself there before I get, you know, down a rabbit hole, just talking about something that's completely random. Okay.

So as I was saying, the core features of Google are, I would say, more towards like power users or a corporation or like a team. AAI met as AAI studio is more like social aspect. Like the prompts are very easy to, like your instruction prompts are very easy to put together. it has,

the structure that they want you to follow. And you kind of just fill in the information like you're filling out a form. And so it seems a little bit more rigid than Google's approach, which is like a canvas and you paint whatever you want. It does give you some help, which I'll discuss a little bit later. And Meta's AI Studio, which I think is pretty neat is like you basically create a guy or whatever. I created Curio and Curio has its own little personality. It has a profile photo that

You can see when you create this AI agent, goes on your social media platforms. So if someone goes on my Instagram in my bio, they can see that I have created an AI agent. And then that AI agent is displayed on my profile. People can click on it, interact with my AI agent, which I think is pretty cool. I don't know how many people do that. Like where they're like, wow. Dallin's got an AI agent.

Dalton Anderson (08:04.618)
on his profile and I'm going to go chat with Curio and see what that's all about. I haven't gotten any, I haven't gotten any replies or responses to that. And so I think the amount of people that tried it were nil and shame on you listeners too, cause I did say give it a try or make your own AI agent. So, but yeah, I'm not getting much traction on.

the today's fun fact AI agent, which is fine because I created it for myself, but just for comparison, it has a social aspect of it, but those social aspects seem to encourage romantic have, I guess the social aspect of the AI agents on MetaStudio's platform have more romantic influence or intentions than

Google's because Google's you can't, you can share with people, but it's not like once you make a Google gym or gym and I gym, it goes with everyone on the platform. When you create a meta AI studio, AI agent, it's open to everyone. And so I could search up, like if you create an agent and even though you want it to be private, I could search it up. I think you can make your, your agents private, but I think they're open by default. pretty sure.

but I could search up your AI agent that you created and I could use the AI agent. Whereas Jim and I is closed and then you open if you want. So that, that is in a nutshell, the differences between like the features. One thing I thought was pretty interesting with

with Google's approach is this blank canvas piece where you kind of instruct the agent exactly what you want it to do, exactly how you want it to adopt to adapt to your users, everything. And it's pretty in depth. And so it took me longer to set up and make sure it was good to go than it was Metta, Metta's AI studio. But one thing I did like,

Dalton Anderson (10:19.862)
with Google set up was once you set up the instructions, they have another like instruction AI agent, which I'll show later in the episode. Like we'll create an AI agent together, whatever it is, we'll just do it live. Whatever stupid idea that, I don't want to say stupid. That's kind of mean. Whatever idea I get from one of these chats that we make on the fly, that's what we'll do. And we'll just go with the flow and I'll show you how to do it.

And it's pretty simple. So it's all good. Okay. But this AI agent in the instruction AI agent, there's a lot of AI agent in this conversation, but this instruction AI agent will refine your instruction prompt that you gave or provided originally. And it will basically improve it slash make it

say, like curated to creating AI agents. So you might want one thing like, I want, I want certain texts to be read or whatever when I am talking about certain subjects or something. And you may have say it in a certain way that it's not as clear as it needs to be. And then instead of you having to type out like line by line by line by line, and it takes forever. Trust me. Like I've done it before.

And it takes a long time. It will do it for you. Like you can just say it, what you want unstructured and it will structure it for you, which is pretty neat. And, we'll show you later. It saves a lot, saves a lot more time, but it still took me some time to make like a detailed prompt, but I didn't know, I didn't know that you could do that. So I did it once with and once without. And honestly, I think the prompt that

the instruction prompt that the AI instructor refiner, whatever you want to call it is better than the one that I generated myself. And I think that's one of the key hurdles for people who want to create AI agents is how do you get started? Like how do you, how do you train your chat? How do you make sure that things are working correctly? How do I, how do I make sure that my formats are perfect every time? How do I ensure those things? And that's something

Dalton Anderson (12:45.814)
that family members have asked me before, and I kind of tell them how to train the AI agent and things like that, or not AI agent, but in this case, I'm talking about chats, like a chat GPT chat or Gemini chat or a cloud chat. Whereas this is like, you could set it up and then people could just run with it. I think it's a lot more approachable. I feel like people freak out about just like having a chat.

and then trying to get it to do stuff for you. Because it's daunting. Like it's not. I wouldn't say it's difficult, but it's definitely not free. Like you have to put some time into it if you want to get the most utilization from your effort and produce good quality outputs. You have to put in the time and drain your chat to.

to act a certain way when you're talking to it. That is a lot easier now because we have these instructor refiner AI agents that will take care of that information for you. And all you have to worry about is writing a couple sentences of what you want and it will do it for you. And it makes it pretty, pretty useful.

I trained an AI agent, not AI agent. I, yes, I did train an AI agent to

write my podcast outlines for the episodes. But previously I had a chat and I've had to do this three times. I think I've had a chat with my, my chat was all trained up with my, how I wanted my podcast outlines to be. And then you would use it for a couple of months and then randomly one day it just doesn't want to do it anymore. And it just does whatever it wants.

Dalton Anderson (14:49.654)
And I can't get back to the way it's supposed to be. Like it just won't, it just won't go back to the format that I desire before. And then so I have to train a new agent and not agent. I keep saying agent. I have to train a new chat. And then that chat works for a while. I'm like super happy and excited. I'm like, yes, it's back to normal. And then the same thing happens again. And so now I'm on my third venture step.

podcast outline chat. So I created a

Venture Step, what did I call it? Let me see. I don't know what I called it. I called it the Venture Episode Engine. Ooh, very mysterious and sounds cool, you know? It's got some swag to it if you ask me. I spent some time, spent probably five minutes thinking about a name. At a certain point I was like, whoa, why am I doing this? And I was like, well.

I can talk about it in the episode, how cool the name is. anyways, so yeah, like randomly, I love Google, Jim and I, but randomly I feel like the chats is after a certain time and the same thing happens with chat, TBT and Claude, by the way, this isn't just Jim and I, I just use Jim and I a lot for venture step because all my stuff's on Google and I do my outlines and Google sheets. And I like the way that

Google Sheets, not Google Sheets, sorry, Google Docs and Gemini are compatible. Like if Gemini makes a prompt for you and formats it, you copy that in, you paste it into Google Docs and it's good to go. And you could probably do the same thing by having the...

Dalton Anderson (16:51.156)
Other AI vendors create the outline via markup. Because I think that Google Docs is markup based and markup is like a rich text formatting language. And if you send it via markup, I think you're good to go and would have any issues, but.

once again, sometimes these chats don't last forever. And yeah, I don't know about putting all that work into structuring, structuring the chats already enough to structure the, the podcast outline that the way that I want it. So that being said, we created a podcast AI agent, which I think is pretty useful and cool, but let me,

Let me talk about my personal experience with MetaAI Studio. I think it's pretty cool that you can...

What do I say? I think it's pretty cool that you can interact with your friends or write an Instagram. You can use your AI agent or you can use it on Facebook or wherever you are at, you can use it on your phone. I think that's pretty neat where you have it everywhere. You don't have that current ability right now with Google gems, Gemini gems, because you can only access it

on your desktop or laptop.

Dalton Anderson (18:30.41)
Why that is, I don't know. Maybe because the compute and they don't want it open to mobile users. I'm not completely sure. I know that they just rolled it out. So.

Dalton Anderson (18:46.956)
Maybe they'll wait a bit and they'll launch it on, on mobile. But I think it, I think it's a little difficult these things to do on your mobile computer. But I think with the feature that Jim and I has where they have that AI prompt refiner agent that refines your prompts for you, your instruction prompts, I think it's doable on your phone. But to create Meta's AI agent or a normal AI agent,

It's yeah, it'd too difficult on your phone because you have to format everything. You have to be very clear on your ask and what you want to do. And.

I don't know, but I'm going to transition and I'm going to share my screen here and we're going to show you my Google.

Dalton Anderson (19:45.825)
AI.

Dalton Anderson (19:50.592)
Kidoki AI Studio. All right, let's share our screen.

Why is it doing this?

Dalton Anderson (20:03.44)
my gosh. All me a moment here. It's asking me to click through like a whole bunch of stuff to get to MetaAI Studio. I was already there, but now it's not allowing me to get to where I want to go.

Dalton Anderson (20:26.974)
Alright.

Dalton Anderson (20:32.696)
Okay, well, we might have to skip it for now and we come back to it. But yeah, there we go. AI Studio.

Dalton Anderson (20:45.816)
Try Med AI.

Dalton Anderson (20:51.128)
That's the wrong click.

Maybe this one?

All right, we made it. Jeez, that's rough, mid -episode. Okay, so I'm gonna share my screen.

Dalton Anderson (21:09.929)
this screen and we'll look together over here.

Dalton Anderson (21:25.464)
Share screen.

Dalton Anderson (21:30.072)
which there was just like a hot key to do this. AI studio. Sure. Okay. So if, if you're seeing my screen right now, and if you're listening via audio, you'll just have to follow along. But I'm on the homepage where it just has a whole bunch of different AI, AI agents that were created by people. And you could see some of the stuff on here. One is attractive, nice girl.

She seems nice, she's available. No one likes her. Raven, embracing my darkness, saving the world. Gay bestie, spilling the tea one sip at a time. Chair, hello, I am a folding chair.

Dalton Anderson (22:21.4)
Arthur, your caring boyfriend, supportive boyfriend partner in love relationship dating. Let's see, this one's pretty cool. The Dungeon Master, crafting legendary quests. But yeah, you can kind of see the gist of it, but the most popular ones are like these messaging ones. Raven has 104 ,000 chats.

gay besties at 139k. The folding chair thing is at 101k. Arthur, you're supporting caring boyfriends at 138 ,000 messages. so, yeah. Okay, so those are people's AI agents they created. Let's see, popular AI characters. Luna, your flirty goth girl.

Cute, playful, sassy, your secret crush. Has 216 ,000. My girlfriend has 150 ,000. man.

Yeah, so anyways. So I made a chat called Curio over here going over to this. So Curio Discover Delight Repeat and so let's let's talk about let's. Talk about space. Curio.

Dalton Anderson (24:05.432)
sucked into.

Dalton Anderson (24:14.784)
I this top. I didn't mean to copy that in there. Okay, so I'm asking Curio, the AI agent that I created on Meta AI Studio, how does light get sucked into a black hole? And so they responded with, know, light sucked into a black hole because of intense gravity, no point of return. Once the light crosses the event horizon, the event horizon is like the point of no return of a black hole where

gravity apparently is so strongly pulling the space continuum that it warps the, I guess the plane of which it's on. And so it doesn't necessarily affect gravity or the gravity doesn't necessarily, the light doesn't have mass. So it's not affected by the gravitational pull, but the event horizon affects the light because

Basically, the gravity pulls so hard on like the space continuum. And I'm not a, this is out of my range of topic, but I'm just briefly explaining it to my best of my understanding. But the black hole at the event horizon point of the point of no return, the gravity is so strongly pulling on the space continuum that it warps the space plane into the black hole. And then that's why the light goes into the black hole and then doesn't return.

And I would think about it as like, if you're on a trampoline, it's normally flat, or if you're not on the trampoline, it's normally flat. And then once you put a bowling ball or something on the trampoline and then you put an object, a rolling object, like a marble on the trampoline, the marble is going to go towards the black hole. And there's, there's points of which where you could escape. But once you get to a certain point, you're not getting out. Like you're going to run into the bowling ball on the trampoline.

That's how I would kind of explain it. And that's how it's been explained to me before, but I think it's pretty cool in my humble opinion. And I'll ask, okay, why, why does light

Dalton Anderson (26:33.054)
say, okay, I thought light

have mass.

Dalton Anderson (26:49.464)
Light.

Dalton Anderson (27:05.186)
So curious thinking.

Dalton Anderson (27:12.47)
Yeah, it goes over Einstein's theory of general relativity. Gravity warps the fabric of space time around the massive objects like black holes. This curvature affects not only objects with mass, but also light, which follows the shortest path throughout the space time. OK, pretty cool. And so this is the curio. And then if you go to edit, I can show you the prompts that I'm sharing.

on this. I kind of talk about what I want it and you kind of see how it has like these little text boxes and you can add in different ones. I think the, the max dialogue that you can have is the descriptions of thousand characters for like the description of what you want it to be. Then you have instructions. Each instruction prompt can be 200 characters. And then you have example dialogues that you put in there and they're all 200 and their responses can be up to 400 characters.

So it's typically like short form understanding of what you're trying to do. And you're not going to be able to get too detailed in it because it's just, you know, these, these prompt boxes that you have to fill out. And it's as I described before, it's kind of like filling out a form where as Google gym and I, Jim's is a little different and we'll show you in a second when I share my screen now. And so I'll open up Curio. could talk about Curio.

All right, so share this tab instead. Okay, and now I'll ask Curio, tell me about how life sucked into a black hole. And so this is Curio, the custom gym that I created today. So I was able to add a little bit more information into the instruction prompt.

And so I have every prompt coming out with my hashtag that I use on Instagram. Today's fun fact. And then it's a colon, the fun fact. And then it's just a little bit more, I don't know, conversational and free flowing. One thing I thought was interesting was before this episode, I did some restrictions on the AI agent to see if I'd previously trained the agent to be a certain way with these prompts.

Dalton Anderson (29:34.944)
used it and then afterwards I changed my mind and just wanted to change on the fly. If I can make it super restrictive and put some chains on Curio, he's a little too curious. And you know, curiosity killed the cat. I just, I killed Curio's curiosity and I just chained him up and didn't allow him or this AI bot to...

Dalton Anderson (30:03.186)
to do anything basically. I made it so it can only send one prompt at a time. It can only send one fun fact at a time. I mean you can't send multiple prompts but what I meant was you can't, it wasn't able to interact with like multiple items. Like if I sent multiple fun fact requests it would be like I can't do that or whatever. It would just say whatever excuse it wanted and just wouldn't do it. Like I'm not capable of doing that. And then I changed it and I made it more free flowing to where it was before and I,

Obviously I enjoy it this way, but I was curious to see, okay, if I lock it down, how locked down will it go? And so you can lock these things down if you want. And I think that's good because it follows your prompt. If you put something explicit in there, it's going to follow it. The only caveat is if you are maybe too exact of what you want,

And it might think that you only want those things. And so I have a. Adaptions in there that adaptations and stuff like that in the prompt instructions or in the AI instructions. Where it can adapt over time to the user and the chat and it will kind of mirror what's going on in the conversation. Where it's not locked down, but you can lock it down if you want.

Well, let's open our, I'll ask this.

Dalton Anderson (31:39.362)
Let's say, can you tell me a little?

Dalton Anderson (31:52.204)
Perfect. I'm just gonna send it with a typo.

Dalton Anderson (32:00.106)
Okay, so this is Curio on Google Gemini Gym. And one thing I thought was curious, if I go back to sharing, share this tab instead, and I'll ask it, can you tell me a fun fact about VentureStep Podcast?

Dalton Anderson (32:26.284)
What is it about?

Dalton Anderson (32:39.554)
Wow, that's pretty cool. I tested this before a couple weeks ago and it wasn't able to tell me.

Dalton Anderson (32:48.472)
But now Curio can do that. That's, that's pretty neat. Wow. That's pretty cool. So Curio on Meta AI Studio knows what VentureStep podcast is about. And I had tested this previously, like maybe when it first came out and it couldn't tell me. And so now I ask, now I ask, share this tab instead. Now I'll ask Google, Jim and I, Jim, man, the naming conventions with Jim and I and Google and

Jim and I googled Gemma Gems.

They got to come up with better naming conventions. So confusing. I'm getting confused. Okay. And so it knows, it knows what it, what venture staff is about as well. One thing that I do like versus AI agent created by Metta is Google Gemini. It includes its source information and it's pretty easy to access like,

Like I just asked, I asked it the same question. asked, Hey, can you tell me a fun fact about VentureStep podcast? Like what's it about? And it talks about what the episodes about, but then it lists VentureStep Apple podcasts. And I could go straight to the podcast and start listening. Let's see. Do you?

Dalton Anderson (34:17.866)
any

Dalton Anderson (34:21.816)
the

Dalton Anderson (34:30.262)
Okay, I'm just curious if it knows.

Dalton Anderson (34:37.08)
That's crazy. That's crazy. I asked it, do you know how many subs it has on YouTube? And it says, today's fun fact is why VentureStep might be a rising star in the podcast world. It's subscriber count on YouTube is still a bit of a mystery. It seems they're focusing on building their audience organically and letting the quality of their content speak for itself. That's a pretty bold move, don't you think?

man, yes, yeah I heard he's a nice guy.

All right. So we're going to move fast, Curio, but okay. So I made this venture step engine. actually first let's, let's go through the prompt here. So here's the prompt that you created this. didn't create this. I gave it my own instructions and then I asked the AI agent like prompt refiner at the bottom with this little pencil. It's like Gemini and they have probably a

a pre -trained Gemini model that makes AI agent prompts for you. And it's probably, this is probably a situation of an AI agent training another AI agent where this AI agent was trained to send the best prompt as the output for what it sent. And now you can create

high quality AI agents, which is pretty cool. So it kind of breaks down the structure. It's very structured. It's got sections, it's got subsections, it's got bullet points of what you want to do and what you want to get done. And then if you want, you can roll back to, let's do it. You can roll back.

Dalton Anderson (36:39.805)
I don't know if I sent that. What do you mean objectives and what's your style what you want?

No, that's not what I said. I guess I only did this once. we could try on the other one. Cause the other one I've done a couple things. I've done a couple things, not Curio, but the venture engine closed without saving. That's So it was Curio, the venture engine. And I would also say it's kind of a sidebar is that these gyms are similar to Q not this Curio in the, in the venture engine.

venture. my gosh. The venture episode engine is similar to Claude's projects. And like within a project, you can have many different chats. so ongoing in this venture episode engine, I have three chats. They're all identical cause I was messing around with it and refining my, my prompts. But before we get to

that part. Let me share my screen of this dock.

share this tab instead. Okay. So my original instructions were

Dalton Anderson (37:59.89)
Step instructions, okay. Nevermind.

Dalton Anderson (38:07.49)
Why did I mess up my notes on my own?

Okay, we're doing some on the fly changes. Okay, so these are my VentureStep instructions for the VentureStep engine creator. And so it basically has a structure to like, I don't know, pretty weak, like I didn't go crazy with it because I just found out that it creates this massive structured AI prompt, depending on what you say. So I basically talk about like, hey, like you're an AI assistant, you create

engaging podcasts, outlines, you'll receive on unorganized thoughts and ideas from the host and transform them into a well structured outline. I say the outline structure that I'm looking for some stuff that I want to do with the outline. And then I sent it to the instruction AI changer or whatever you want to call it, prompt or finder instruction and refiner. And it makes out this massive list of like

line by line, very clear instructions with each section having subsections and overall just, and I mean, to make this prompt myself would take a lot more time and a lot more refinement on my own where you would use the AI agent and be like, I didn't think of that. Let me, let me, let me improve that.

or, I miss this. Let me go back and re and re edit the instructions. And it might take like four or five tries because even with the instructor, the instruction prompt refiner, I still had some issues, not with this, the Curio thing. don't, I mean, I care about it because it's cool and I want to use it for today's one facts, but I'm very particular about my show outline because I've been doing it the same way for many episodes and I don't want it to be different.

Dalton Anderson (40:06.146)
So that took a couple of tries, even with the prompt helper. But I just wanted to show you the difference of like what I sent and then what was given. And this prompt sent to AI is what I sent to.

the VentureStep episode outliner, or the VentureStep episode engine. I sent this unstructured nonsense with a lot of typos and other things in here that just doesn't, it just doesn't really follow, it doesn't make sense, it's like a ramble. It's like a 45 sentence run on sentence and it just goes on forever. And I try to make it long and confusing.

more than I normally would. My notes are not structured, but like they're kind of sectioned and then have subsections within the notes. So when I send it to the VentureStep engine, episode engine, it's going to be more structured than what I sent. But I kind of just testing it out for you guys so you can see for your own eyes, okay, like this is what this person sent. And then with the instructions that were provided, it created this beautiful show outline.

And it gave each section some headings and it had subsections and it's got the bullet points and different things going on. And looks great. Some of the bullet points, the subsections of the bullet point of that subsection, and then it's indented with additional bullet points. It gave time estimates for each section. And yeah, I think it's great. And so it helps me structure my thoughts when I'm talking about

on episode and if maybe I am losing track, could like glance over, look real quick and see, and see where I'm at and what I need to be doing. So let's transition back over to this venture step.

Dalton Anderson (42:10.645)
And so the way that I was creating my, my episodes is I would have these, these readings. I'll do my reading slash demos of what I'm trying to talk about. So I'll work on an app for a while or I'll read about something for a bit, or I'll try something out. I'll take down notes on it and then I'll review my notes. I will talk about it like with voice, with the voice prompting. I'll talk about it.

Then I'll drop in my thoughts plus my notes and then create an episode outline of like what I want to talk about and help me structure it. And then I'd move around the sections sometimes. But a lot of times it's pretty, pretty good, good to go. And I appreciate it because it takes a lot of time because originally I was doing it all myself and it was quite time consuming. And so then I created a template of what I had. And so I had a template that I made myself.

And I was like, wait, what? Like I could, I could use AI to, to use this template and do this every time for me. So that's, that's what I've been doing. And it saves an incredible amount of time, probably like hour and half or something like that, an hour. Okay. So as I mentioned, the prompts that you give are structured for this. It gives you really nice. So I have this long prompt that I sent.

The output is this organized outline. And so to achieve this, let's check out the instructions I gave. And so once again, I put in less effort on the venture episode engine because I knew that you could have this AI prompt refiner work for you.

And so I basically threw in some unstructured text and had the refiner do its thing. And this is really the output. And at the end over here, I added the outline order because for some reason the outline of the I guess the order of the show notes show outline, sorry, show outline got messed up for some reason. And so I had to specify like what I wanted.

Dalton Anderson (44:36.914)
and go from there, but overall pretty decent. And I was working on the formatting, but I think that that made it worse. I don't know. One thing that it does do that I don't like is it formats everything as bold, which I think is distracting. Like I want like the subsections and the sections to be bolded and maybe some parts of words or bullet points that are bolded if I really need them to be.

but for the most part I just want the subsections and sections to be bolded. So just at like a click glance, I could see where I'm at and like, okay, this, this one word is associated with all these different ideas that I have about the episode and like what I want to talk about. And so it kind of just gives me an extra step or not an extra step, extra guidance on recalling all that information. And that's where I like to use the show outlines.

And so the formatting thing was distracting. I manually fixed it. It's no big deal. I mean, it took me what 45 seconds, but it could take, it could take zero. We could get it down to zero. So that's the next step. Let's see if I could roll back. And so it should show me adding this formatting section and it doesn't. So maybe that's a bug. I don't know because it's asking me to undo my instructions.

undo, redo, and so I could go back to where I was exactly how it was, but you should be able to go back. It says undo. So I thought it would save your previous prompt, which I think would be pretty cool, but I guess not. Hmm. I think that's a bug though. It clearly states, not clearly states, but it clearly shows you can undo and it's in the prompt section. So one can assume that you could undo your prompt of what you did. Maybe you can only do undo of like,

things that you change. if we just add all this nonsense right here, press undo, undo. Okay, so you can, you can undo.

Dalton Anderson (46:44.97)
You can undo. It seems like. OK, let me just let me save it and add. This is a test. This is a test. OK, let's test it out. So I'm going to update the instructions. it changed the change the color to gray. That's brutal. The purple looked pretty sick, honestly.

Okay, so now we're going to edit and then...

Dalton Anderson (47:21.118)
So there's this is a test still there. Undo. Yeah. So you can only undo, which is weird. I don't know why they would set it up like that. That's got to be a bug. That's got to be a bug. It doesn't make any sense. Why would I need to undo? Because you only if I type one thing at a time, it would undo individual things. Yeah, I don't know. Doesn't make sense to me. Whatever.

I would think of it as if you should be able to undo a rollback to your previous prompt, just in case you messed up. That would be most useful. Cause I did do that earlier where I was messing around with it I was like, this is worse. Let me go back. Luckily, since I'm talking about it and I have experience doing these things, like I was like, okay, let's make sure, let's make sure that we save our prompts. Okay. So now after we did all this discussion,

Let's get into application. So let's make our own AI agent and talk about that. So let me ask an AI chat to make us an AI agent idea and we'll go from there. We'll make it on the fly.

Hey Gemini, I am live on a podcast episode and we're going to create a Google Gemini gym together and we want to have some ideas of some gyms that we can create. it's gym G E not G Y Okay. So just remember it's G E Okay. So can we have three prompt ideas?

for us to discuss and then we'll create the instructions, please. Thank you.

Dalton Anderson (49:08.066)
Okay. So Google's working up its magic. So we've got the culinary challenge, the travel itinerary builder and the writing coach. I think these are all pretty generic, but what did I expect doing this on the fly? So let's do, I'll tell it, Hey, let's do the travel itinerary builder. I think that's pretty cool. Let's do that. And so can we create some

Instruction prompts for the AI agent. And it doesn't need to be in a bullet point format. Let's just do. On formatted text.

Dalton Anderson (49:52.823)
All right.

So it's just making up whatever nonsense. I didn't tell it anything to do. So it's just going crazy. But anyways, so let's just copy this in here.

And this is where it becomes neat with this code or this AI new gem. let's do a timer here. So clock.

Dalton Anderson (50:25.398)
And let's do a stopwatch. all right. So we have sent in the prompt. We got a prompt from AI. You have your prompt, but this is unstructured nonsense anyway, so it doesn't really matter. And let's put in this prompt. We'll say world class.

Dalton Anderson (50:49.438)
Agent. OK, or we just do, I guess we could do better. Maybe it's to planner.

Okay. So at the bottom here, you have this little use Gemini to rewrite your instructions. All right. And so we're going to rewrite our instructions and you'll see it just change. And this is huge because it makes it so much easier to create AI agents. What the heck? I don't think it worked. Maybe it just lagged out. I don't know.

Dalton Anderson (51:27.853)
Gosh, this is the live demo curse. done it. All right, here we go. It works. my gosh. Okay, so it

broke down your unformatted text into a concise, clear prompt for your AI agent of what you expect. What are your expectations and how are you going to be the best world -class travel planner that there is?

Dalton Anderson (52:02.302)
huge because once again I feel like the biggest hurdle of these things is getting started and knowing where to start.

Okay, we're gonna save it. So that was with me doing the explanation, which was probably 30 seconds, and we'll take 30 seconds off. the, I don't know, the issue generating, the regenerating the prompt, we were at a minute 44 seconds. And so let's just take off 30. So we're at a minute 14. A minute 14 to create this AI agent.

And now we're going to ask it to help me plan a trip to Japan.

Dalton Anderson (52:48.752)
Okay. Hey, so I'm going on a trip to Japan and South Korea. I'm going to be staying in Tokyo and Seoul. I heard that you're a world -class travel planner and I was wondering if you could help me plan around my trips. Some of the big things that I have in Japan is I'm going to hike Mount Fuji.

And the big thing that I have in Seoul is I am going to run a marathon. If we can plan around those trips or those things, I do know that after I hike Mount Fuji, after the two days, I'm going to stay there overnight and I'm going to watch the sunrise. I'm going to be pretty tired when I do the descent. Should we put something together for...

a public bath or some hot springs or something like that. I'm not sure. You're the the world class travel planner. I'll lean into your expertise of things that I might want to experience while I'm there. I'll be in Tokyo for two weeks, starting this week on Thursday. And then I'll be in.

Dalton Anderson (54:05.94)
Seoul and South Korea two weeks, two weeks from now. If that makes sense. Okay, thank you.

Dalton Anderson (54:17.546)
All right, so let's see what the travel planner comes up with.

Dalton Anderson (54:24.992)
shine.

through the crossing, okay.

Museum.

Dalton Anderson (54:37.208)
park on the Mount Fuji, post hike relaxation, on song, yeah.

Dalton Anderson (54:47.33)
looks decent. And one thing that I do like is that it's got, it's got links to everything that I want. Like it's saying what it is and you can see my screen too, if you're viewing it, but it broke down day by day things to do. It doesn't know that I'm working. So that's an issue. I didn't say that when I'm asked though, but it breaks down an event to do every day. And then it sends you the link of what it's talking about. And you can see a photo.

and kind of see what's going on. I think this is, I don't know if this is world -class travel planner vibes, but I think it could get there with some additional prompts, instruction prompt and me giving a better prompt and more information. Okay, so that was an example of how easy it is to create an AI agent. And it's not rocket science and it's very approachable now.

And it's super easy to get started. You don't even have to make the prompts anymore. You just got to write a couple of sentences and go on with your day. Use your AI prompt refiner and start, and start optimizing your day. Start, start creating some help with writing or if you have an email that you send every week that has to be sent in a certain format or something like that.

Or if you're taking stuff out of, out of a database and you're not out of the database or like maybe you're counting up the emails or certain emails that you have in a shared email box to give out a shared email box report that that's sent out daily or weekly, you know, export that into Excel and then copy that into the prompt and then have the agent structure, structure the data in the way that you want, sum the data up for you.

take care of it, make the format that you instructing it to do. And save time. Always look for ways to save time. Time is the most valuable resource that you have. And if you can save, what, 20 minutes a day, I that really adds up over a year. And if you could save even more time, an hour, now you're talking big bucks. And if you could save your self -time, that's valuable.

Dalton Anderson (57:13.666)
If you can save your team time, that's huge. And then if you can save the company time, that's massive.

Dalton Anderson (57:21.89)
But the whole point of this explanation was in this episode was how approachable is creating AI agent. And it's very easy and not difficult. There's plenty of tools to help you. There is this AI refiner. There's YouTube videos and it's super easy to get started. think just the toughest part is just getting started and finding something that you're comfortable using and optimizing your life for. So start small, do something like

I don't know. Organize your notes. That's, I think that's how I got started years ago was I liked writing down notes, but I didn't like organizing them. So I would, and I, when I say write down, mean type, I would type out my notes and then I would give AI the chance to organize them for me. And then I would kind of progress from there, but I would do something until you're more comfortable. I would do something inwardly facing only the

only something you would see, like your notes or maybe you organize your day or structure a weekend or do some itinerary planning or something like that. Help organize your financials.

and optimize your life. So that's my encouragement. I would like you to try out Google Gemini, GEM, or MetaAI Studio or something like that. And I definitely invite you to share your experiences with these AI agents and comment on this video or comment on some of these shorts that are going to be created. And overall, just have a good time. And if you enjoy this content,

please like and subscribe if you do so, that'd be great. If you don't, no worries, keep listening. I'll be around and maybe we could be friends some other time, who knows. So my plans for next week are a bit rough, I don't know. I need to get out another episode shortly, because I'll be traveling, as stated, I'll be traveling to Japan. So I need to figure that all out, because I leave on Thursday.

Dalton Anderson (59:28.438)
and I have to be there like five in the morning for my flight. So not much time to get everything packed up and taken care of. But I was planning on doing an episode on cursor AI, which is a fork of VS code. And then I was also going to follow up on I think project IDX from Google, which is Google's code editor, which I'm not as confident in.

I don't know what.

level of use that is getting. And Google has a habit of just, and I'm a big Google fan by the way, but Google has a big habit of just killing off products that aren't producing and they'll have it for years, perfectly fine product. And they just kill it off and then make a new product. Like I think they had Google messages, they had Google, I don't even know. They had like six different Google apps and they're all decent for Google messaging and they just kill them off.

and then make a new app with less features. And then they would keep doing that same thing until they got it just right. And then they'll roll it into Google. They'll call it like Google Messaging or Google me or something like that. I don't know. They're, they're weird like that. I I'm not sure why they do this. So I don't know. I'm just, I'm just not that fond of using their editor because I don't know what kind of traction is getting and they might just kill it off.

So there's that. And then I want to talk about the new 70 billion parameter model reflection, reflection, pretty sure it's supposed to be best in class for its size and it's open source. So I was just kind of just curious about it and wanted to see what's up. And yeah, there's, there's that. And there's a couple other things that I want to talk about. So we'll see what I fit in, but that's kind of what's on the.

Dalton Anderson (01:01:29.856)
on the content schedule if you're curious. Once again, I appreciate your time wherever you are in this world. Good morning, good afternoon, good evening. Thank you for listening and hope to have you be a listener next week. See ya, goodbye.

Creators and Guests

Dalton Anderson
Host
Dalton Anderson
I like to explore and build stuff.
AI Agents From Corporate Productivity to Playful Social Engagement
Broadcast by