EP 395: AI Agents - Everything you need to know
53:24 Share

This is the everyday A I show, the everyday podcast where we simplify AI and bring its power to your fingertips. Listen daily for practical advice to boost your career, business and everyday life. Picture this billions of A I powered agents doing work autonomy, ously without human intervention, powered by large language models, doing many of the manual, tedious tasks that many of us hate.

This isn't some future. This is possible now. And I think, especially over last week, A I agents have been all of the buzz. So today we're going to be talking about AI agents, what they are and why everyone is suddenly talking and scrambling to figure out in implement A I agents. All right, before we get started, a quick word from our partners at microsoft.

So why should you listen to the workplace b podcast from microsoft? Because it's made for leaders who know they must adapt to stay ahead. Work lab is the place to find real world lessons and actionable insights to guide you and your organization through your A, I transformation.

That's W O R K L A, B. No spaces available wherever you get your podcast are thanks to microsoft. Let's die back in. I'm excited for this one. You what's gone on money is Jordan and this is everyday ai ah if you're new here, thank you for joining us. If you're on the podcast, as many of you are, make sure to check out your show notes.

There's going to be a link yet got a click IT is our website your everyday A I dot com on that website, hundreds of episodes, hundreds of hours of whatever you care about in generate A I. It's all on our website for free, for you to learn like a free generate A I university. And this show is for you doing IT every single day the live stream, the podcast and the free daily news letters.

I just jump into A I agents and you know what, i'm going to start here because when was this a year ago? Yeah ah at the end of last year, the end of twenty twenty three, I came out with twenty four bold AI predictions for twenty twenty four, right? So I think this was probably in november.

So about ten months ago, one of the things that I claimed that would happen in twenty four, twenty, twenty four, and these were all, let me be honest, very bold and wild predictions. And the weird thing is many of them came true. And when I put out these predictions, not a lot of people agreed with me, but I did kind of a mid term show in june.

And strAngely enough, almost all of them had either come true or we're on a on pace to come true and one of them that had yet to kind of come to fruition. As I said in twenty twenty four, we may see more A I agents than humans. So here we are in september twenty twenty four in the last week has been absolutely bunkers for A I agents.

It's like every single big trillion dollar company in the world got together and they said, alright, let's go ahead asians on three, one, two, three brick and then everyone went out and started either announce or pushing or highlighting or updating and improving their agents capabilities. So on today's, we're going to go over some of the basics um and I also say I want to hear from you so I want to hear from our lifestream audience what questions do you have about agents? So everyone joining us, you know kobe and Michael and antanas and mario, everyone's sliter a brian.

Thanks for joining us. What questions do you all have? But also podcast audience, I always leave my a my linked N U R L in my email in the show notes.

So make sure you go check that out. Reach out. Let me know what questions you have about A I agents. But a one thing that I want to first talk about um because we're going to be going over this over the course of today show is there's different kinds of A I ancient as well. So there are A I agents that will either be triggered manually, right?

So I can sit here, essentially click a button and have that A I power agent essentially go execute a series of task, right? So there are a manual A I agents h, there are autonomous A I agents. So those are, uh, A I agents that are essentially running around the clock OK.

And then there's something in between. These are called some I autonomous agents. And am I making that term up? Yes, I am making up these three classification.

So if you go google, then you're probably not going to find very much, uh, but I think we have to separate them in those three different categories. So keep that mind as we talk in kind of the middle, this my autonomists agents, uh that something in between. So that could be triggered by a workflow that could be uh triggered uh by something um a customer inquiry.

Uh IT could be something that is scheduled to happen, right? So it's those three different types. So something that is manually trigger by a human, something that is fully autonomists running kind of around the clock, and then something that is may be triggered by in events, a web hook, a customer inquiry, a zai er eta are.

So keep that in mind as we talk about everything that's going on in the world of A I agents. So let's start with an overview because AI agents are not science fiction anymore and they're not new. We're going to go over time, mine later.

But you know agents have kind of been in the conversation for actually more than a decade, right? But IT is just with the recent advancement, obviously in generate A I in large language models that have thrust uh, A I agents not just into the conversation but also they've crept their way into every major company. The biggest companies in the united states and therefore the world have all highlighted in emphasize their investment in I agents.

So this is no longer one of those uh fringe discussion topics like I think two years ago, IT kind of was right. You had to kind to be a large language model or genitive AI or an OpenAI dork a like myself you know two or three years ago to really um be talking actively about AI agents. But it's not like that anymore.

Uh the everyday person, especially here in the U. S. You are going to be hearing about you probably already hearing about them this week, and we're going to go over some of the pieces of news there.

Uh, but if not, you are going to be hearing about A I agent attune. So I think it's important to set that ground work. And here's the other thing, like I said, there are not science fiction anymore.

Uh, they're not even a friend idea anymore. They are here, they are life, they are available and they are the other thing. That's why d is they are both no code and low code. So even if you want to deploy these uh in your organization is not like you need to have an army of software developers.

If you have someone that can understand uh, human language and type on a keyboard and click a mouse, that is all you need uh, to connect A I agents, whether those are my uh semi autonomists uh tony ious or uh autonomists or manual, if you have a human that can speak to essentially an A I system and click a button, you can start deploying these for your organization today. Um and like I said, I think this has the potential to be both a very powerful and exciting right as this can lead to billions of dollars literally in saved revenue ah for companies right of huge companies that are already are using A I agents as an example. Uh I believe amazon uh is spending nearly a hundred to billion dollars a year in research and development, right? So when you talk about the potential to save billions of dollars for companies, that's not high probability.

That is actually what's happening um but you know productivity and obviously can skyrocket right when you think of the manual test that you do over and over and then essentially saying, hey, why don't I train an agent to do eighty percent of this, although it's possible today, right? So there's there's great promise in terms of productivity, business growth, new opportunities in etra. But then there is obviously the downside, right? Sobering reality.

You all and I I never liked you here on the everyday AI show because what you hear I think is a false narrative uh, because it's it's a complete security blanket when we say things like, oh, A I won't take your jobs. Someone using A, I will, that's B, S, A because that person using A, I, uh, especially if they're using A I agents right in theory, uh, the work of that one human can do the work of two, three, ten, twenty, thirty humans, right? Ah so there's obviously a great chAllenges and pitfalls when we talk about the the safety and the ethical aspects of A I agents.

So are they are both um terrifying and exciting at the same time. absolutely. But that's why I think now is the right time to have the conversation about this.

So let let just jump straight into IT. And I want to talk about some of the latest breakthrough and why this is especially timely. Now, right? I've been thinking about having this conversation about A I agents now since day one, since I started every day.

A I you know, more than a year a half ago. Uh, but I think now is the right time because literally over the last ten days, three of the most uh either the largest or the uh some of the most consequential companies that control how we work have gone all in on A I agents. Let's talk about a little bit.

First, microsoft, right? So we talked about this a dedicated show literally yesterday. So if you did not check out that show, make sure to go do so.

Uh, so we talked about kind of microsoft copilot in their a copilot wave to so part of this wave, two of microsoft co pilot was talking about the microsoft copilot studio. Essentially, IT is a drag and drop AI agent builder are. So these are customizable agents a for microsoft copilot three sixty five.

And they have the ability to automate complex workflow across apps. Also, here is the important thing, right, because it's all about your company's data. Can you work with dynamic, right? Because what good isn't agent if it's dumb or if it's slow, right? Or if IT doesn't have access to your company's data? That's why I wanted to start off by talking about microsoft because of this is rolling out. I believe IT should be a available by the end of september here twenty twenty four.

Um so having that contextual memory uh for personalized business interactions is huge, right? And and I think that's one of the things that's been you know that has created this kind of gap, uh you know over the the the promise and promise of agents over the last two years when he combined with large language models to actually seeing them in production IT has been this ability to number one is IT technically feasible, right? Uh, because two years ago is pretty difficult, right? You had to have a bunch of dorks, right? People do here than me are.

You had to do a lot of dcc taping in the giver to make this work, not anymore with tools, literally like microsoft copilot studio, the ability with no code in low code. So what that means is, is typing something 啊 to an A I and that A I helps you build in A I agent。 And you might need to click a couple of buttons here and there to connect your your database, right?

So maybe from share point one drive eeta a microsoft, you know sweet of a microsoft 3 sixty five products。 But at that point you can integrate with both external tools, internal tools and your live data sources. This one cannot be overlooked, right? Uh I think the wave to uh announcement from microsoft, I got some good play.

Uh, but I also think that wave two is kind of what closed the gap, I think from the uh, original hype around copilot, right when I was first announced in eighteen months ago til where we are today. I think this wave two announcement really close. That gap um in those things is bringing this combination of large language models that can do autonomists work via A I agents and tap into real time data right? That's not the only one, right? This one is extremely important.

Salesforce, yeah literally this week uh is the sales force dream force conference um and we've seen A A sales the sales four CEO essentially say, hey, we've been A C R M company for decades. We are doing a hard vivid. We are in A I powered agent company now.

Literally, that's what he said, not me. Um so you have to look at their new offering called agent force. So we're going to break this down a little bit here in a while. But you I mean, salesforce is one of the largest companies in the world, right? One of the leading tech companies.

And if you are a large enterprise organization that sells something ultimately to to customers, clients, other businesses, right? So whether you're in b to B, B to c, there is a high likelihood you're using sales force, right? And there is a high likelihood whether you are on a sales team or not, there's a high likelihood you spend a decent amount of time in sales force going through all of this data, uh, you know, to help you Better manage your customer relationships.

So with agent force, these are autonomists agents force sales, uh customer service and marketing uh with a deep crm and data cloud integration with sales for a sales force as well as a low code in almost no code environment for quick deployment. And also here's the other thing, the invidia collaboration, right um that that piece is huge there for sales force, right? And kind of the third of peace that I wanted to talk about is open eye.

Now obviously OpenAI is not a uh company like microsoft and sales force has been dominating uh kind of the tech landscape for multiple decades. But I think they actually are one of the most important companies in the world. Here's why a right. I just mentioned as an example, microsoft powered by OpenAI GPT four O A, another big company that we probably use every day, right? Apple, if you are human being, living and working in the united states, you either probably use microsoft windows every single day, or you use apple or mac every single day.

And guess what? Apple and mac are both going to be powered by OpenAI GPT four o right? The apple intelligence, yes, they have their own kind of H A I uh small language models handling certain queries uh locally but then for other queries es they are going to be sending you uh to OpenAIG P T f our o a nd w e t alked a nd w e've h eard e ven f rom m icrosoft t here, uh willingness to integrate OpenAI newest model OpenAI o one okay, and this is a IT was formally code named q star. Then IT was code named strawberries, right? So if you've heard about q star or strawberry, this is a model that is completely different I um and also hate can we stop calling this G P T O one that's not his name, right?

So there is the GPT class of models, uh right? And then there is the reasoning class of models, which is the a one, right? But so many companies, yeah I mentioned, uh, microsoft and apple are actually using OpenAI technology, but there are A I wish I had an efficient count I would venture to say tens of thousands or hundreds of thousands of companies that you probably use fairly often, right? You're not using tens of thousands of them, but there are uh thousands and thousands of companies that we all use all the time that are actually powered by open the ice technology.

So when you're using, no, you think you might be using, oh, this A I powered real estate APP guess what? They're using a GPT four. Oh, right. So we also have to pay attention to what OpenAI is doing in this agents or A I powered agent space.

Um and for that I think we have to look at their recent model, this uh strawberry uh q star a one right h so last week a open a eye released o one preview and o one minute. These are again a new class of models um and we don't even have access to their most powerful model, which their most powerful reasoning model, which is a one OK. So we essentially have a one preview and a one many, but this is an agent's model that is capable of reasoning IT is capable of thinking under the hood.

And that is one key aspect that can bring this agented workflow to thousands of the software and services that we all rely on every day. So think of how I said, right, as an example, open uh, microsoft has their new copilot dio agents. Sales force has gone all in with agent force, right? This is not some one off trend.

We are going to be seeing this probably on every single big piece of software. You're going to see agented capabilities and there's a good chance they might be powered by open eye. So hey, this is fresh off the press right um couple hours ago, sam ultima tweed, incredible outperformance on goal three even though IT took a while and then he left a link uh to an OpenAI technical goals uh blog post and here's what uh go three is build an agent with useful natural language understanding.

I am going to read this quickly. We plan to bill an agent that can perform a complex task specified by the language and ask for clarification about the task if it's ambiguous. Today, there are promising algorithms for supervised language task such as question answering, uh, syntactic parsing and machine translation. But there aren't any more, but there aren't any for more advances linguistic tic goals such as the ability to Carry a conversation, the ability to fully understand a document, and the ability to follow complex instructions in natural language. We expect to develop new learning algorithms and paradigms to tackle these problems.

So this blog post here that sam altman just shared about is an older blog post but uh he is clearly hinting that with open a eh one model, they have essentially outperformed this goal, right where they said we expect to develop new algorithms and paradigm to tackle these problems, which is the ability for an A I agent to understand a document and follow complex instructions in natural language. So we're going to be talking a more about OpenAI e here in a bit and hey, uh, lifestream audience, if you didn't not get your question in, try to get IT. Now I am going to tackle them, uh, try to tackle them at the end, right?

So let's talk about what actually makes in A I agent, right? Like what the heck is an AI agent? So I kind of wanted to start off with some of these recent examples because it's actually uh, not here than a squirl on kito that all of these things from you know microsoft sales for open a eye have happened over the course of like six days, right?

That can be a coincidence on where the industry is adding. But I think we also have to just talk about what makes an AI agent, right? This is midlife yall. This is six core functions of an AI agent.

There's other list out there essentially what defines an AI agent, right? Because IT just sounds like a buzzword, right? And in the same way that companies wanted to throw out just A I or generate A I or large each language models, right, they try to spit out those lood blood words as quickly as possible on their earnings calls.

Incorrect forecasts right now you're going to hear the same thing with A I agent. So what the heck is an AI agent ah in what's the difference between an AI agent in a large language model? Well, first of that line is probably going to blur as large language models add to their capabilities.

But i'd say here are the six core functions that kind of constitute in A I agents. So you got ta check all of these box boxes, right? Uh, number one is that needs to be powered by a large language model, which enables IT to have natural language processing.

What that means is the average human needs to be able to talk or type to an A I agent in plain english or playing whatever language you speak. And IT needs to be able to understand human language. That's the number one.

If you have to write python on the front end for something to happen, in my opinion, that is, not in A I agent. Can IT be an agent, sure, right? But not by my definition.

And needs to be powered red by a large language model. And IT needs to have N, L, P. Or natural language processing, need to understand humans, uh, number two, he needs to have tool interaction. IT needs to be able to use outside tools, right? Number three, IT needs to have the ability to plan on its own right to handle complex test uh in A I agent needs to be able to plan how they uh plan to do that right um and sometimes you might see this uh chain of thought reasoning, which is kind of what we see h with you know strawberry or open A S uh one model.

Uh, number four, you you need to be able to have memory and or access to company data, right? An agent is not useful or even I I I wouldn't consider an agent if he doesn't have a memory or the ability to store or access your company's data. And that might come through number two, right, tool interaction, but he needs that.

Number five, this is a big one. And you should be able to actually execute tasks on europe half again, whether that is a manual trigger, autonomists trigger, semi autonomists trigger, IT needs to be able to actually execute something, not just go. Here's how you would execute this in theory, right? That's one of the things that, uh, you know one of the lines in the scene, so to speak, that uh, differentiates a large language model within the AI agent.

Can IT actually execute a task, right? And I we're seeing that with a especially right now with agent force and with microsoft copilot studio. There's other offerings that we will be talked about here a bit a from google meta eta.

But IT needs to actually be able to execute task, which is one of the new things that we saw from a microsoft three sixty five copilot in their studio. Uh, essentially their A I agent builder is they can not execute tasks on your have you have to give IT access or you have to kind of no click. Yes, you can execute this task, but I can.

And that number six, learning and adaptation, right? That's the other big one. If if an agent cannot learn and improve, right? And Normally if this is power by a large language model, IT can learn and improve.

But I need to do that are, let me recap those things quickly. One largest lengths model in natural language, natural language processing. Two, and he have tool interaction or outside tools. Three, the ability to plan or chain of thought reasoning for memory and or access to company that to. Five, the ability to actually execute test, and six, the ability to learn and adapt to become Better are we have to take a real quick break to tell you about work lab from microsoft.

So why should you listen to the work lab podcast from my ico? Soft IT explores the questions business leaders are asking, how can they guide their organizations on their AI adoption journeys? How can the technology help them create new products and business models and maximize value? How should they help their teams re scale for this new era of work? And why is that important to be completely transparent about when and how you utilize A, I find the answers on work lab. That's W O R K L A B. No spaces available whereever you get your podcast.

So let's get back to the show, right? Those are the six core functions, and that is what differentiates AI agents from large language models because right now, large language models, uh, with the exception, I think, of what OpenAI o one model will do in the future because right now a one does not have tool access, right? If i'm being honest, if I had access to all the tools that GPT four o has access to, for example, code interpreter slash a advances data.

Now, if IT had access to, you know, daily, even though I don't think dolly that great, if I had access to the ability to upload files, if I had access the ability to um browse the web here, the browse with being integration, if if o one, if the one model had access to that right now, IT would be an agent, right? IT would be an A I power dating right now. IT doesn't have access to those things, although OpenAI said, uh, that should be around the corner so essentially you have the two different pieces with ChatGPT and open eye right now.

You have the GPT four model, which doesn't have, uh, you number three on this kind of the ability to plan chain of A H chain of thought reasoning and tax tax execution number five but uh you know GPT four models um don't have that but the new a kind of reasoning model does so OpenAI has all the pieces which is why you know I think that not so cyp tic tweet from same ultimate means a lot more than we think. So, like I said, the difference, large language models, tax generation, a right. And agents are decision making, execution, completing task, real world, being able to learn in a debt right.

Let's go over a very, very brief history, very brief history. I don't want this to absolutely be an hour long podcast. You can't really talk about modern day AI agents without first shouting out line chain.

I lane chain was very early to this game. So in october of twenty twenty two, lang chain launched, launched. And you know this was essentially, you know very ahead of its time.

Don't give me wrong, but think of this is way IT was kind of like duck tapping in the guides ing, right? So you could tap into different large language models and then you know kind of string together, creating a workflow to create a sort of agent. So again, IT was very ahead of bits time, but uh, you can't not talk about long chain.

So then in november twenty two, OpenAI launched GPT. right? H on no wait. That was twenty three, sorry.

So first line chain in quarter three uh introduced l cel for that essentially their kind of language for flexible agent creation. Then we saw um in november from OpenAI the ability to create GPT. So again, that's not in A I agent, but that's a lae the framework, right?

So with a custom G P S, that was the ability for essentially uh, agented task, right, not an agent but more of an assistant right where you could you know can make a custom version of a large language model up, you know, upload some of your data. IT has access to all of those tools, and IT can complete a singular tasks, right? Can really do that chain of thought reasoning and you know, run autonomously or semi autonomous.

Ly, but gp s were a definitely a step in that. Then we can fast board to early twenty twenty four. So in video h showcase their AI asian hardware acceleration.

So you can skip over nvidia's involvement in this. And then we go to uh April twenty twenty four, meta AI uh with lama three uh has started to integrate and slowly tease out its agents workflow. The same thing with google um at their I O conference, kind of a tease and previewed agent building capabilities there.

And then that brings us to current day in september in the last, like I said, seven days. We've seen OpenAI announced a one model, a preview of agents reasoning once that has access to all the things that the GPT models have access to. Then we saw the microsoft co pilot studio wave two, uh, with these new enhanced agent capabilities.

And then we saw agent force from sales force that is marking a shift from one of the largest uh you know software tools in the world going from A C, R, M company to an A I agent company. It's a very brief history and just a very brief of recent history, right? But AI agents have been around for a very, very long time, right? Let's talk about how they can change work.

Well, if you're still listening to this podcast and you don't see the potential for how they could undertake work and how they could change work, it's like you you got you got to think you got ta look at the writing on the wall, right? And also look at the largest companies in the world, right. Um so I I, I, I already talked about microsoft, right? Um I already talked about apple, right?

Apple with their apple intelligence. So apple and microsoft, they control the the devices that we use. And microsoft has already all outside, yes, AI agents, they're here.

They're they're a big part of what we're doing. Apple is not there yet because they're like two years behind literally every other company. But we have apple intelligence coming out.

So presumably, uh, you will start to see some type of autonomists uh or 3i autonomists workload in the future with apple。 In video, right? I'm going over the largest companies in the world.

Uh, in video, they create, they are the engine, right? They are literally the ancient driving A I agents and how we all work in the future. Google like I said, google at their um at their I O conference um announced A I agents right in their vert tax, A I agent to builder.

Uh so as they continue to approve, uh, they are very capable gemini models. I think g mini metals are great on the back end for developers in their AI studio. Not so great on the front end for the average user, right.

But with the vertex, uh, A I agent builder, again, one of the largest companies in the world. Uh, then you have amazon, right? Uh, amazon is investing billions of dollars in two large language models.

They have their own large language model platform. Uh, amazon q they're working on simple agenda. Work flows are and then mea as well, right? Those are literally these six largest companies in the united states and six of the largest companies in the world.

And they are investing their dollars in their people into A I agents in some way, shape or form. So you have to see the writing on the wall. You have to always follow the money in the money the time the attention is all going toward A I agents.

So this will greatly impact uh, how we all work. And IT is unfolding now before our very eyes, the talk about a couple example business use cases, right? Um I hear you all when you reach out to me on linked in and semi emails, I always appreciate that.

And you know you always say, hey, we ve got to hear we got to hear more business use cases, right? Sometimes going to give some examples. So sales force, agent force, right? Ah so they did a video on this.

We'll leave that video in the news letter who's short video five minutes that kind of shows um how this kind of no code or no code agent building works in their agent force. So in this example, um you can use your C R M data, uh, you know, build the parameters, you build the guard rails again, you don't have to be super technical. IT is drag and drop.

So on my screen here again, I have like a left side, right side splits on the last side, a with natural language, you kind of set the parameters in the rules of how your AI agent can respond. And then what happens is that is connected to your life sales force data. And then you can essentially create a chat, right? It's funny chat.

Botts were so ahead of their time ah but they were so useless right because with chat box pro ge language models you have to set all of these defined conditions uh which maybe account for like one percent of conversations that might actually happen on a chapter t but now IT is flick, I think with natural language processing in large language models. Now you can probably hit on like ninety nine percent of all customer inquiries, right? But with agent force, you can essentially tap into your sales for data, create a simple agent that you can then put on your website.

And in this example that a sales force had um you know essentially a customer is asking like, hey, what a know they had a question about the order sales force, uh, the agent answered IT and then they said, hey, I you know installation, I need installation and then they said, what about next friday right so they weren't saying, hey, what about you know friday, uh, september twenty seven they just said next friday, right? So you have to have that natural language processing in a large language model to be able to take nuance conversation and translate that to data and then, uh, connect IT to your database, right? And then the sales force agent gives the options and they said, we have the following available times on this dates.

And then they gave a couple options. The a again, the person just says two thirty and then that's one of the options and then the uh agent force agent schedules IT and then updates the C R M accordingly right? So this is huge, something I mean, I know this is a super simple example, right? But customer service is I don't you know what I don't see how humans in the future in the very near future are going to be the driving force behind customer service.

IT doesn't make sense anymore ah right IT really doesn't. When you have you know if this works right, this is this is all very new. You got to take this with a big grain assault but then you have similar offerings from uh, you know other uh, tools, software like right? I said microsoft copilot dio same.

But yes, that is going to completely change how customer service is done. Uh, let's talk about sales, you know, kind of sales and marketing a zap ier zap ier has they call them A I assistance? But they're really great zaire's central we've taught you about on the show before.

Similarly, these are more like semi autonomists, but with dragon drops. So you have to connect to other third party, no software or maybe as an example, you don't use sales a sales force, maybe you use something else, uh, right. But you can build kind of these semi autonomous agents dragon drop with natural language, tapping into a large language model VS API.

Er right. So oh, when we get a ah you know an inquiry on our website, you know you can kind of set some simple rules with natural language. If someone chooses option a, you know you should send them this email but writ in a way that takes into account all the information they put on the form right, you can essentially have a combination setting guard rails, uh setting conditional triggers um and then you know tapping into literally almost any software or service on the internet.

So when you have this, you know kind of A A I powered agent tapping into a large sandwich h model in your data, your services, right, you can see the future. Uh, coding. Coding is another one, right? We're talking about real business use cases.

Um I I know I asked you all. I got mixed responses, right? Do you want to see more on A I coding AI software development, right? Because this is changing as well. Um and I think that this obviously makes the job easier for people who are already you know in software development or engineers.

Are people doing coding python right at sea? These tools like rap lets agent uh cursor A I right being able to cope with the A I uh davin you know the A I software engineer um from cognition right, which OpenAI a prominently featured in its one announcements right. So those three companies alone, there's a lot more um but these are probably three you know curr A I replant agents um in devon from cognition are probably three of the more prominent.

These are AI powered software agents, right? So yes, you have these agents that will in theory be able to do a little bit of anything and everything, but then you also have a more niche and targeted A I agents. But then also, so yes, this h changes.

What's possible for software developers, engineers is saturday. Then this also gives new capabilities, right? That's the other thing. With A I agent, they bring a new capabilities to anyone else, because I can go right now on cursor AI as an example. And I can, with natural language, I can say, hey, build me a program that does this, I need IT.

And then probably after you know, three or four five problems, I actually have a piece of software that I created that solves one of my own problems, right? So the capabilities in specific categories of work are about to dress will change. So yes, there are general purpose A I agents, right? But then there are uh niche or skill specific AI agents as well.

And I think that replay agents, cursor A I and defend are great examples of those benefits right here. As we're wrapping up, you gotten be able to see the benefits, right? This is twenty four seven non stop work, right? If you if we're talking more in the autonomists agent side, sales force itself said they see billions in the next year, billions with A B yeah, i'm not crazy.

People think i'm crazy when I come off with top takes is is gonna be more A I agents than human and people laugh and I like this guys dumb. And then sales force says IT sales four says within a year they see billions of their agents. Is that part marketing part? Um you know, dreamer vision, sure right has happened at dream force, after all.

Is IT a reality? Is forget lutely could have or sorry, could have be a reality? yes.

Hacks, yes, IT could be right. Because now I can go create today, ten, twenty, thirty, forty agents. You literally have people and software that are agents that are creating other agents. So literally autonomists agents working around the clock, creating other autonomists agents. You have agents interacting with each other again, no longer science fiction.

Two wor three years ago, you know, as dorks were sitting around on on edit and quora ming like wouldn't be cool and now it's it's reality, right? So the benefits, twenty four, seven, working with your a data on guard rails, you set up using your up to date knowledge. And this also helps non technical users.

That's the other big thing, right? Because again, technically, you had artificial intelligence powered agents now for probably more than a decade. But with generate A I A democratized IT lower the bar so you no longer you know ten years ago yeah there is artificially in intelligent asian ts um you know probably more on the manual or semi autonomous side right but now anyone can do IT.

You could have started this process at the beginning of this podcast and probably could have already created five by now. It's also probably a key for me to go a little faster, right? We can't talk about this without talking about the chAllenges and limitations, right? So the quality of reasoning is huge um really defining this agent computer interface or A C I as IT sometimes called um in designing designing effective interface for tool utilization.

That's a chAllenge as well. But I think what we're seeing with agent force and copilot studio, uh really uh is is huge and will see uh with vertex A I agent builder from google how that goes. But also there's model limitations, right?

There's biases right now in these large language models that are in theory powering these A I power agents. So that's a huge chAllenge in all. So ethical and safety, right, especially as A I agents start to learn and adopt, as I learn in adapt. Uh right. And then as we start talking about multi ancient environments, again, i'm not A A domes dyer out here talking every day about skynet in terminator.

But you ve got to think about that, right? Hey, you set up some maybe weak guard rails and then you you know set out a system of you know fifty autonomous A I agent that can all talk with each other of other power by, as an example, maybe uh, open a eyes of one model, right? Bad things could, in theory happen, right?

If you don't have enough human in the loop, which I know you're less be onest. Sometimes that is an exaggerative way of saying, hey os's, humans are babysitters for AI. But if you don't have enough human in the loop at the right point, a series of multiple autonomous agents working together in the future could obviously be very unsafe alright.

If you don't constantly have humans, uh, overseeing them, if you don't have, uh, strict guard rails in place and constantly doing QA on these agents, the future can also be very scary, right? So I don't want to I don't want to skip over bias, stereotypes, safety, ethics, right, even talking about job loss because I don't care what anyone says. AI is ultimately going to take away away more jobs and IT creates sorry, it's not me being uh a pessimist that me being a real alist right.

That's why the largest uh companies in the world have all invested billions of dollars into a general AI, into GPU, into A I agents right? Yet they are laying off tens of thousands of employees with record high profits, right? I don't know why people don't, uh, you know, I know we need to be optimists as human beings when we talk about A I and our jobs and career and the meaning of work, right? But you also have to be a realist.

Wall street hates employees. Wall street loves profits. In wall street is really going to start to love A I agents, right? Keeping eye on microsoft stock, salesforce stock in the in the next year.

So you'll see what I mean right then the future in what's next? Well, I don't know what's next. All I know is IT definitely has to do with A I agents, like I said, out of the top six companies in the us. So aside from apple, every other company, microsoft, video, alhamid, amazon and meta, have all come out in publicly said that they are either investing in or investigating A I agents or they are all in right.

And then we talk about sales force as well, a one of the largest uh companies in the world that touches so many other fortune five hundred companies, right? Like the majority of large enterprises are using sales force, I don't even know who sales force is. Biggest competitor? St, right? Who knows? I mean, I probably know a couple of them, but you get what i'm saying.

The biggest companies that dictate how we work all going all in on AI agents. Let's wrap this up. Yep, so some key takeaway here. We are on the cusp of this big shift from, yes, A I to generate A I to A I agents, right? And you you can't really talk about that shift without talking about A G I.

And and you know are are more capable models and A I agents kind of one of this next step that gives us to this artificial general intelligence or when AI is much smarter at every task than any human being in the world and can do all of these tasks without really much human intervention, right? And I think agents are a step to get us there. Whether you are rooting for H.

G. I or not, IT doesn't matter. But we have to talk about that. So we also have to talk about the duality of AI power agents. Um if i'm being honest, I was shocked over the over the last week that all of this is happening all at once.

And that's why I said we gotta do a show on this, right? Um I think a year ago was still too early, but now the writing is on the frequent oyle. You've got to pay a attention to the right and we have to talk about the dwelling and we have to business leaders and decision makers.

You have to do uh A I agents and in kind of this implementation in the right way, you have to prioritize humans. You have to be prioritize safety guard rails. Data is such a you can't skip over this in the mad rush to be the first big company to implement something from agent force to implement in the biggest you A I agent from a microsoft copilot.

You don't have to do that. You should be acting now, right? If you don't already have generated eye in place in large language models, your kind of screwed, right?

You should already be having that conversation about these autonomists workload, A A I powered automation and AI agents. You have to be having those conversations whether you're implementing IT tomorrow. That's not what i'm telling you to do.

You have to be planning for IT now, right? So um let's rap got a couple of questions here. Uh let's see. Let's see. Um that's part of doing this as a live stream.

I know I ramble on sometimes, but I think even with everything that's going on with A I and we're spending so much time now talking to A I, I think it's important have a human conversation. So you podcast audience, you might get kind of annoyed when I know ask questions so much of our life stream. Audience, but I want this to be a place.

And hey, podcast listeners, come join us. I want this to be a place where we have real human and human conversation, right? I want this to be a place where we can explore and learn together.

Uh, so a question here from kobe, uh, things we're joining us, he says i'm curious about your thoughts, Jordan, on billing agents inside hob's APP central. Uh, kobe was, uh, you know, reading ahead. Have you built any and have you found this APP effective? So when zp ier central first came out, I went in and played around with IT.

So I did build one or two basic workflows to tell the truth. I haven't gone back there sense, and i'm kicking myself because I really should be right. We pay for up here every month and i'm barely tapping into IT.

But I do think especially for companies that maybe don't have sales force um or don't want to go all in on this agent force because the thing I didn't mention is I believe the pricing for agent force is two dollars per conversation. I'm not sure about that go to market strategy, but that's why they probably you know have a bunch of smart people telling them that that's the way. But um I do things up central um and microsoft copilot uh studio their A I agent bilder are probably the two are most robust kind of A I agents that are ready to go.

Uh I do think and and a microsoft copilot o is obviously a little more capable because IT works with your real time data out of the box. Where is in APP central of your building kind of these uh agenticity lows um you know you are I won't say duck taping IT, right? It's a little more seamless when IT is built in to the software that you're using on your computer versus when you're having to kind of put the pieces together.

A mary asking devils advocate here again, if A I is going to quote unquote takeover, what are humans going to be able to do? Good question, right? Yeah, I think that's important to talk about.

I kind of, uh, wrapped the show as much, uh, right, talking about that. What is the future of work? And and I think IT is a more uh, responsible use of humans in the loop.

Uh, I think hopefully, right. Who knows? Maybe we will actually see some companies make the ethical decision.

I think unfortunately, many companies, as they figure out largely with models across their organization, top to bottom, uh, you know, as they figure out A I agents, I think so many companies are going to rush to just fire employees by the thousands. We've kind of already seen that, uh, right from big tech companies over the last year. I think that's going to happen a lot.

It's going to happen in mass. Unfortunately, again, i'm not being me. I'm not being a best that's following the data that's following the writing on the wall.

So what are humans going to do? Well, hopefully that more creative and strategic thinking, you know who knows maybe we'll see a this being common place having you know four day work weeks has an example right um who knows right but I don't know what the future of humans role in this as large language models uh get uh smarter and more capable and as we see A I powered agents. But I do know what we've traditionally done in the past of decades where you are rewarded for your domain expertise, right?

You you're rewarded for all of these facts in decision making that you have in your head around you're specific industry. I'm not saying that that's going to be gone, but that is going to be greatly deprived tizer. IT doesn't matter what you know about marketing.

IT doesn't matter what you know about shipping and logistics, right? IT doesn't matter because these AI agents, when they can access your comedies data, when they can access the um entire uh history of the world in theory, right and they can um adapt and access your company's data and learn and change right that domain experience you have up in your brain of a sudden is much less valuable. So I do know the future of work.

Requires us to think more creatively and strategically. And I think almost all of us in the same way that you know now, who would have thought thirty years ago that essentially every single knowledge worker here in the us would be working quite, quote, on the internet all day. I think in the very near future, we are going to be working in and around AI all day.

So whether that's microsoft copilot, whether that's, you know you agent force or you know from google, a vertex AI agents, whether you're building your own A I agents, I will assume the majority of workers in the coming years will be working around generate A I in large sandwich models in some way, shape, form, just like we're all right now working around the internet. Last question, we're ongoing to rap IT terror asking how can we ensure fairness in detecting A I agent use in the workplace and schools and prevent false accusations of relying on A I without proper evidence? That's that's a great question.

Tera, and I have no clue. I have no clue. Maybe this goes to the question that marriage is said, what are humans going to do? well? H, humans, we humans, when we need to prioritize ethics, safety, getting rid of bias, uh, right? Uh, fairness, equity and inequality, right?

These are the types of things as A I agents and large language models take away, maybe the majority of our manual, mundane knowledge work test. These are the big issues we have to cover. Great question, tera.

And don't worry, we're going to be here every singing goldy helping all of us uncover those next steps of how do we work in the future. Well, at least know a little bit of how we're going to be working in the near future, but we're going to continue to tackle IT here on every day. Ai, thank you for join us.

If you haven't already, please go to your every day A I dt. com. Sign up for the free daily news letter we're onna be recapped today show yeah I know what was a long one, but I think this was an important discussion to have because I would not be surprised if we have the same conversation at the same time next year.

I wouldn't be surprised if there are billions of AI agents out in the world. That wasn't just my bold prediction. You heard IT from sales force as well.

But all I know is the future of work is generated. A I and the largest companies in the world are going all in on A I agents. Thanks for tuning in.

We will see back tomorrow in every day for more everyday A I. thanks. That's a rap for today's edition of everyday A I.

Thanks for joining us. If you enjoyed this episode, please subscribe and leave us a rating. IT helps keep us going for a little more AI magic. Visit your everyday A I dot com and sign up to our daily newsletters so you don't get left behind. Go breaths and barriers and we'll see you next time.