Everyday AI Podcast – An AI and ChatGPT Podcast - EP 361: AI Agents - What they are, and why they’re suddenly all the buzz

Episode Date: September 18, 2024

Everywhere you turn.... AI agents. This past week more than ever, the biggest companies and hottest startups have seemingly gone bonkers for AI agents. Why? And, what the heck are they? We'll give you the 101 on all things agents. Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on AI agentsUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:1. AI Agents Explained2. Future of AI Agents3. Corporate Adoption4. Economic Impact5. Ethical ConcernsTimestamps:02:30 Daily AI news06:30 What are AI agents?10:03 AI agents have become mainstream in companies.11:37 Human-AI collaboration saves billions for companies.15:26 Microsoft's contextual memory aids business interactions greatly.18:16 Salesforce uses autonomous agents, deep CRM integration.23:14 Develop AI for complex language tasks understanding.25:03 Defining an AI agent and core functions.29:02 AI agents surpass large language models due tool access.32:35 Recent AI advancements by Meta, Google, OpenAI, Microsoft, Salesforce36:53 Chatbots now effective using natural language processing.39:27 AI agents manage tasks using natural language triggers.44:02 Challenges, opportunities in agent interfaces and tools.45:28 Insufficient guardrails can make autonomous AI unsafe.51:29 Microsoft Copilot Studio excels over Zapier Central.54:18 Future work demands creativity; AI integration inevitable.Keywords:Natural Language Processing, AI in Customer Service, AI Tools, Microsoft Copilot Studio, Zapier, AI in Sales and Marketing, Coding with AI, Replit agents, Cursor AI, Devon from Cognition, Future of AI Agents, Salesforce, Benefits of AI Agents, Challenges of AI Agents, Bias in AI, Ethical Concerns in AI, AI Agents Discussion, Corporate Adoption of AI, AI Accessibility, Economic Impact of AI, Productivity and Efficiency in AI, AI News, AI in Workforce, AI in stock market, Future Trends in AI, Business strategies for AI, Workplace fairness, AI in everyday work, Agentic models of OpenAI, Ethical Decisions in AI.Send Everyday AI and Jordan a text message. (We can't reply back unless you leave contact info) Start Here ▶️Not sure where to start when it comes to AI? Start with our Start Here Series. You can listen to the first drop -- Episode 691 -- or get free access to our Inner Cricle community and all episodes: StartHereSeries.com Also, here's a link to the entire series on a Spotify playlist. 

This is the Everyday AI 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. Meet Firefly AI Assistant, now live and Adobe Firefly, the all-in-one creative AI studio. Just describe what you want to create and the assistant handles the rest, orchestrating multi-step workflows across Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome. The assistant accelerates execution. Picture this.

Billions of AI-powered agents doing work autonomously. 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 the last week, AI 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 and implement AI agents. I'm excited for this one, y'all. What's going on? My name is Jordan, and this is Everyday AI. 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.

You got to click it. It is our website, your EverydayAI.com. On that website, literally hundreds of episodes, hundreds of hours of whatever you care about in generative AI, it's all on our website for free for you to learn. It's like a free generative AI university. And this show, it's for you doing it every single day, the live stream, the podcast. and the free daily newsletter. So make sure you go to your everyday AI.com

and sign up for that free daily newsletter. And you know what? We're not dragging this on anymore. We're finally, today is the absolute blast day. Right. Yeah, we've been tracking this on. But our thanks a million giveaway, if you haven't signed up to celebrate a million downloads.

I think we're ending it at midnight tonight, central standard time. So giving away a year, whatever your favorite large language model is, the base subscription, the $20 to $30 a month price. whether that is chat GPT, Claude, Gemini, et cetera, go make sure, sign up. It takes 10 seconds. You get a link, refer your friends.

It's not too late to enter the contest. All right. I'm excited to talk about AI agents. It's a topic that we don't talk about maybe enough on the show. So before we dive in, we're going to start as we do every day by going over the AI news. So California has taken some bold steps. against AI-generated election misinformation.

California Governor Gavin Newsom has just signed three significant bills aimed at combating the use of artificial intelligence to create misleading images and videos and political advertisements, especially with the 2024 election approaching. So the new law makes it illegal to create and publish deep fakes related to elections within 120 days before election days and 60 days after, imposing civil penalties. for violations. So courts will have the authority to halt the distribution of misleading materials enhancing legal recourse against deceptive practices. Large social media platforms will be required to remove deceptive content under a pioneering law set to take

place set to take effect next year positioning California as a leader in regulating AI usage and political context. So political campaigns must disclose if they are using AI-altered materials in their advertisements, promoting transparency to voters. So Newsom's actions come in response to recent incidents of manipulated content, including a controversial video featuring altered images of Vice President Kamala Harris, shared by Elon Musk, as well as AI generated images shared by former President Donald Trump. In related news to that, California Governor Gavin Newsom also signed legislation to protect actors and performers. from unauthorized use of artificial intelligence,

allowing them to back out of contracts that may permit studios in California to create digital clones of their likeness. So that's a pretty big one as well. All right. Our next piece of AI news, we barely snuck this in the newsletter yesterday as it went out a little late. But BlackRock and Microsoft have united for a $30 billion AI infrastructure fund. In a significant move to bolster AI capabilities,

BlackRock and Microsoft have announced plans to establish. a fund exceeding $30 billion, yeah, billion with a B, $30 billion aimed at investing in AI infrastructure, including data centers and energy products. So the newly formed global AI infrastructure investment partnership, that's a mouthful, will focus on enhancing AI supply chains and energy sourcing to support the growing needs of AI technology. The fund is expected to mobilize up to $100 billion. in total investment potential factoring in debt financing,

which could significantly impact the AI landscape. So also worth noting that MGX in Abu Dhabi-backed investment company will serve as a general partner, while Nvidia will contribute its expertise in AI chip technology further strengthening the fund's capabilities. All right, last but not least, the Biden administration is set to host a global AI safety summit amid regulatory challenges. So the Biden administration is making headlines this morning by

organizing a global safety summit focused on artificial intelligence, highlighting the urgent need for regulation in this rapidly evolving field. This initiative comes as Congress struggles to establish effective oversight for AI technology. So this conference will be held in November of this year in San Francisco and the summit aims to foster global cooperation for the safe and trustworthy development of AI with participation from member countries including Australia, Canada, the EU, Japan, and obviously the United States. All right, we're going to have much more on that and all other stories in our newsletter. So make sure if you haven't already, please go to your everyday AI.com. All right, that was actually a lot. I'm sorry, that took a hot minute and a half,

but let's just jump into AI agents. You know what? I'm going to start here because when was this? A year ago? Yeah. At the end of last year, at the end of 2023, I came out with 24 bold AI predictions for 2024, right? So I think this was probably in November, so about 10 months ago. One of the things that I claimed that would happen in 24, 2024, 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, you know, not a lot of people agreed with me, but I did kind of a midterm show in June. And strangely enough, almost all of them had either come true or were on pace to come true. And one of them that had yet to kind of come to

fruition is I said in 2024, we may see more AI agents than humans. So here we are in September 24 and the last week has been absolutely bonkers for AI agents. It's like every single big trillion dollar company in the world got together and they said, all right, let's go ahead, agents on three, one, two, three, break. And then everyone went out and started either announcing or pushing or highlighting or updating and improving their agentic capabilities. All right. So on today's show, we're going to go over some of the basics. And I also, hey, I want to hear from you. So I want to hear from our live stream audience. What questions do you have about agents?

So everyone joining us, you know, Colby and Michael and Antonas and Marie, Rolando, everyone, Cecilia, Tara, Brian, thanks for joining us. What questions do you all have? But also podcast audience, I always leave my LinkedIn URL and my email in the show notes. So make sure you go check that out. Reach out. Let me know what questions you have about AI agents. But one thing that I want to first talk about, because we're not going to be going over this over the course of today's show, is there's different kinds of AI agents as well.

So there are AI agents that will either be triggered manually, right? So I can sit here, essentially click a button and have that AI powered agent essentially go execute a series of tasks. Right. So there are manual AI agents. there are autonomous AI agents. So those are AI agents that are essentially running around the clock. Okay. And then there's something in between. These are called semi-autonomous agents. And am I making that term up? Yes. I am making up these three classifications. So if you go Google them, you're probably not going to find very much. But I think we have to separate them in those three

different categories. So keep that in mind as we talk. And kind of the middle, these semi-autonomous agents, that's something in between. So that could be triggered by a workflow that could be triggered by something, a customer inquiry. It could be something that is scheduled to happen, right? So it's those three different types. So something that is manually triggered by a human, something that is fully autonomous running kind of around the clock, and then something that is maybe triggered by an event, a web hook, a customer inquiry, Zapier, et cetera.

All right. So keep that in mind as we talk about everything that's going on in the world of AI agent. 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 a timeline later. But, you know, AI 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 generative AI in large language models that have thrust AI agents not just into the conversation, but also they've created.

their way into every major company. The biggest companies in the United States, and therefore the world, have all highlighted and emphasized their investment in AI agents. So this is no longer one of those fringe discussion topics. Like I think two years ago, it kind of was, right? You had to kind of be a large language model or generative AI or an open AI dork 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. 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. But if not, you are going to be hearing about AI agents a ton. So I think it's important to set that groundwork. And here's the other thing. Like I said, they're not science fiction anymore. They're not even a fringe idea anymore. They are here. They are locked. They are available. And they are the other thing that's wild is they are both no code and low code. So even if you want to deploy these in your organization, it's not like you need to have an army of software developers.

If you have someone that can understand human language and type on a keyboard and click a mouse, that is all you need to connect AI agents, whether those are semi-autonomous, autonomous or manual. if you have a human that can speak to essentially an AI system and click a button, you can start deploying these for your organization today. And like I said, I think this has the potential to be both very powerful and exciting, right, as this can lead to billions of dollars, literally in saved revenue for companies, right? You have huge companies that are already using AI, agents as an example. I believe Amazon is spending nearly $100 billion a year in research and development, right? So when you talk about the potential to save billions of dollars for companies, that's not hyperbole. That is actually what's happening. But productivity obviously can skyrocket, right?

When you think of the manual task that you do over and over and then essentially saying, hey, why don't I train an agent to do 80% of this? Well, it's possible today, right? So there's great promise in terms of productivity, business growth, new opportunities, et cetera. But then there is obviously the downside, right? The sobering reality, y'all, and I never lie to you here on the everyday AI show, because what you hear, I think, is a false narrative because it's a comfy security blanket. When we say things like, oh, AI won't take your job, someone using AI will, that's BS, because that person using AI, especially if they're using AI agents, right? In theory, the work of that one human can do the work of two, three, ten, twenty, thirty,

humans, right? So there's obviously a great challenges and pitfalls when we talk about the safety and the ethical aspects of AI agents. So are they both 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. All right. So let's just jump straight into it.

And I want to talk about some of the latest breakthroughs and why this is especially timely now, right? I've been thinking about having this conversation about AI agents now since day one, since I started everyday AI, you know, more than a year and a half ago. But I think now is the right time because literally over the last 10 days, three of the most, either the largest or some of the most consequential companies, that control how we work have gone all in on AI agents. All right, let's talk about it a little bit. First, Microsoft. All 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. So we talked about kind of Microsoft copilot in their co-pilot Wave 2. So part of this Wave 2 of Microsoft co-pilot was talking about the Microsoft co-pilot studio. Essentially, it is a drag-and-drop AI agent builder. All right, so these are customizable agents for Microsoft Copilot 365,

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 data, right? Because what good is an agent if it's done? 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 this is rolling out, I believe it should be available by the end of September here 2024. So having that contextual memory for personalized business interactions is huge, right? And I think that's one of the things that's been, you know, that has created this kind of gap, you know, over the premise and promise of agents over the last two years when combined with large language models to actually seeing them in production. It has been this ability to, number one, is it technically feasible?

Right. Because two years ago, it was pretty difficult, right? You had to have a bunch of dorks, right? People dorkier than me. You had to do a lot of duct taping in McGi-Ring to make this work. Not anymore with tools literally. Like Microsoft co-pilot studio, the ability. with no code and low code.

So what that means is typing something to an AI, and that AI helps you build an AI agent. And you might need to click a couple of buttons here and there to connect your database, right? So maybe from SharePoint, OneDrive, et cetera, Microsoft's suite of Microsoft 365 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? I think the wave two announcement from Microsoft, it got some good play. But I also think that's wave two is kind of what closed the gap, I think, from the original hype around co-pilot, right? When it was first announced, you know, 18 months ago. And so where we are today, I think this wave two announcement really closed that gap.

and one of those things is bringing this combination of large language models that can do autonomous work via AI agents and tap into real-time data. All right, that's not the only one, right? This one is extremely important. Salesforce. Yeah, literally this week is the Salesforce Dreamforce conference. And we've seen sales, the Salesforce CEO essentially say, hey, we've been a CRM company for decades. We are doing a hard

pivot. We are an AI powered agent company now. Literally, that's what he said, not me. 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 y'all, I mean, Salesforce is one of the largest companies in the world, right? One of the leading tech companies. And if you're a large enterprise organization that sells something ultimately to customers, clients, other businesses, right? So whether you're in B2B, B2C, there's a high likelihood that you're using Salesforce, right? And there's a high likelihood, whether you are on a sales team or not, there's a high likelihood that you spend a decent amount of time in Salesforce going through all of this data, you know, to help you better manage your customer relationships. So with agent force,

these are autonomous agents for sales, customer service and marketing with a deep CRM and data cloud integration with Salesforce, as well as a low code and almost no code environment for quick deployment. And also here's the other thing, the Nvidia collaboration, right? That piece is huge there for Salesforce.

All right. And kind of the third piece that I wanted to talk about is OpenAI. Now, obviously, Obviously, Open AI is not a company like Microsoft and Salesforce that has been dominating 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, right?

I just mentioned, as an example, Microsoft powered by OpenAIs, GPT40. Another big company that we probably use every day, right? Apple. If you're a 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 OpenAIs GPT40, right? The Apple intelligence. Yes, they have their own kind of edge AI small language models handling certain queries locally. But then for other queries, they're going to be sending. you to Open AIs GPT40, right? And we talked and we've heard even from Microsoft their willingness to integrate Open AI's newest model, Open AIO1. Okay. And this is a, you know, it was formally codenamed Q-star, then it was codenamed Strawberry, right? So if you've heard about Q-Star or strawberry, this is a model that is completely different. All right. And also, hey, can we stop calling this GPT 01. That's not its name. All right. So there is the GPT class of models, right? And then

there is the reasoning class of models, which is the O1. All right. But so many companies, yeah, I mentioned Microsoft and Apple are actually using Open AI's technology, but there are, I wish I had an official 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 tons of thousands of them, but there are thousands and thousands of companies that we all use all the time that are actually powered by OpenAI's technology.

So when you're using, you know, you think you might be using, oh, this AI powered real estate app, guess what? They're using GPT40, right? So we also have to pay attention to what Open AI is doing in this agentic or AI powered agent space. And for that, I think we have to look at their

recent model, this strawberry Q-star 01, right? So last week, OpenAI released O1 preview and O1 Mini. These are, again, a new class of models, and we don't even have access to their most powerful model, which their most powerful reasoning model, which is O1. Okay, so we essentially have O1 Preview and O1 Mini. But this is an agenetic model that is capable of reasoning. It is capable of thinking under the hood. And that is one key aspect that can bring this agentic workflow to thousands of the softwares and services that we all rely on every day. So think of how I said, right, as an example, Open Microsoft has their new copilot studio agents. Salesforce has gone all in with agent force, right? This is not some one-off trend. We are going to be. We are going to

seeing this probably on every single big piece of software. You're going to see agenetic capabilities and there's a good chance they might be powered by Open AI. So hey, this is fresh off the press, right? A couple hours ago, Sam Altman tweeted, incredible outperformance on goal three, even though it took a while. And then he left a link to an Open AI technical goals blog post. And here's what goal. three is. Build an agent with useful, natural language understanding. All right, I'm going to read this quickly. We plan to build an agent that can perform a complex task specified by language and ask for clarification about the task if it's ambiguous. Today, there are promising algorithms for supervised language tasks such as question answering, syntactic parsing,

and machine translation, but there aren't any more, but there aren't any for more advanced linguistic 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 he is clearly hinting that with Open AIs, O1 model, they have essentially outperformed this goal, right? where they said, we expect to develop new algorithms and paradigms to tackle these problems, which is the ability for an AI agent to understand a document and follow complex instructions in natural language.

All right. So we're going to be talking more about open AI here in a bit. All right. And hey, live stream audience, if you didn't already get your question in, try to get it now. I'm going to tackle them, try to tackle them at the end. All right. So let's talk about what actually makes an AI 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 nuttier than a squirrel on keto that all of these things from, you know, Microsoft, Salesforce, and OpenAI have happened over the course of like six days, right? That can't be a coincidence on where the industry is heading. But I think we also have to just talk about what makes an AI agent. All right. This is my list, y'all. This is six core functions of an AI agent.

AI agent. There's other lists 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 AI or generative AI or large language models, right? They try to spit out those buzzwords as quickly as possible on their earnings calls and quarterly forecasts, right? Now you're going to hear the same thing with AI agent. So what the heck is an AI agent? And what's the difference between an AI agent and a large language? model. Well, first of all, 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 an AI agent. So you've got to check all of these boxes. All right. Number one is it 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 AI agent in plain English or plain whatever language you speak, and it needs to be able to understand human language. That's number one, right? If you have to write Python on the front end for something to happen, in my opinion, that is not an AI agent. Can it be an agent? Sure, right?

But not by my definition. It needs to be powered by a large language model, and it needs to have NLP or natural language processing. It needs to understand humans. Number two, it needs to have tool interaction. It needs to be able to use outside tools. All right. Number three, it needs to have the ability to plan on its own, right?

To handle complex tasks. An AI agent needs to be able to plan how they plan to do that, right? And sometimes you might see this chain of thought reasoning, which is kind of what we see with, you know, strawberry or open AI's O1 model. Number four, it needs to be able to have memory and or access to company data, right? An agent is not useful or even, I wouldn't consider it an agent if it 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 it needs that.

Number five, this is a big one. It needs to be able to actually execute tasks on your behalf. again, whether that is a manual trigger, autonomous trigger, semi-autonomous trigger. All right, it needs to be able to actually execute something, not just, oh, here's how you would execute this in theory, right? That's one of the things that, you know, one of the lines in the sand, so to speak, that differentiates a large language model with an AI agent. Can it actually execute a task, right?

And now we're seeing that with, especially right now with Agent Force and with Microsoft. co-pilot studio. There's other offerings that we'll be talking about here in a bit from Google, meta, et cetera, but it needs to actually be able to execute a task, which is one of the new things that we saw from Microsoft 365 copilot in their studio, essentially their AI agent builder, is it can now execute tasks on your behalf. You have to give it access or you have to kind of, you know, click, yes, you can execute this task, but it can. And then number six, learning and adaptation, right? That's the other big one.

If an agent cannot learn and improve, right? And normally, if this is powered by a large language model, it can learn and improve, but it needs to do that. All right, let me recap those things quickly. One, large language model and natural language processing. Two, it needs to have tool interaction or outside tools. Three, the ability to plan or chain of thought reasoning. Four, memory and or access to company data.

five, the ability to actually execute tasks, and six, the ability to learn and adapt to become better. All right, those are the six core functions. And that is what differentiates AI agents from large language models. Because right now, large language models, with the exception, I think, of what Open AI's O1 model will do in the future. Because right now, O1 does not have tool access, right? if I'm being honest, if it had access to all the tools that GPT40 has access to, for example,

code interpreter slash advanced data analysis, if it had access to, you know, Dolly, even though I don't think Dolly's that great, if it had access to the ability to upload files, if it had access to the ability to browse the web via the browse with Bing integration, if the O1, if the O1 model had access to that right now, it would be an A1. agent, right? It would be an AI powered agent. Right now it doesn't have access to those things, although Open AI said that should be around the corner. So, you know, essentially, you have the two different pieces with chat GBT and Open AI right now. You have the GPT4 model, which doesn't have, you know, number three on this kind of the ability to plan chain of action, chain of thought,

reasoning and tax execution number five. But, you know, GPT4 models don't have that. But the new kind of reasoning model does. So Open AI has all the pieces, which is why, you know, I think that not so cryptic tweet from Sam Altman means a lot more than we think. All right. So like I said, the difference, large language models, text generation, right? And agents are decision making, execution, completing task, real world, being able to learn and adapt.

All right. Let's go over a very, very brief history. Adobe just introduced an entirely new way to create, bringing the power and precision of its creative suite into one conversational experience. Meet Firefly AI Assistant, now live in the Adobe Firefly app, the All In One Creative AI Studio. Powered by Adobe's Creative Agent, Firefly AI Assistant lets you start with your vision, just describe what you want, and shape the outcome as it takes form with the Assistant.

The Assistant orchestrates multi-step workflows, drawing on 60-plus Propherson. grade tools across Adobe Creative Cloud apps, including Photoshop, Illustrator Premiere, Lightroom Express, and more to help bring your ideas to life. You can also get started with creative skills, a growing library of pre-built workflows for common creative tasks, like batch editing photos, creating mood boards, portrait retouching, and creating social variations. Every step the assistant takes is visible so you can refine, redirect, or take over at any time. You stay in the driver's seat as the creative director. Adobe Firefly AI assistant now in public beta. See it today at firefly.adopi.com. Very brief history. I don't want this to accidentally be an

hour long podcast. All right. You can't really talk about modern day AI agents without first shouting out Langchain. All right. Langchain was very early to this game. So in October of 2022, Lank Chain launches launched and you know this was essentially you know very ahead of its time. Don't get me wrong. But think of this as a way it was kind of like duct taping in McGivoring, right? So you could tap into different large language models and then, you know, kind of stringed together, creating a workflow to create a sort of agent. So again, it was very ahead of its time. But you can't not talk about Langchain.

So then in November 2020, Open AI launched GPs, right? Oh, no, wait, that was 23. Sorry. So first, LangChain in quarter three introduced L-CEL for, that's essentially their kind of language for flexible agent creation. Then we saw in November from OpenAI the ability to create GPTs. So again, that's not an AI agent, but that's laying the framework, right? So with custom GPTs, that was the ability for essentially agentic task, right? Not an agent, but more of an assistant, right, where you could, you know, kind of make a custom.

version of a large language model, upload some of your data. It has access to all of those tools, and it can complete singular tasks, right? It can't really do that chain of thought reasoning and, you know, run autonomously or semi-autonomously, but GPs were definitely a step in that. All right, and then we can fast forward to early 2024. So, Invita showcase their AI agent hardware acceleration, so you can't skip over Nvidia's involvement in this. And then we go to April 24 meta AI with Lama 3 has started to integrate and slowly tease out its agentic workflows. The same thing with Google at their I.O. conference kind of teased 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 Open AI announce the O1 model, a preview of agentic reasoning once it has. access to all the things that the GPT models have access to. Then we saw the Microsoft co-pilot studio Wave 2 with these new enhanced agent capabilities. And then we saw Agent Force from Salesforce that is marking a shift from one of the largest software tools in the world going from a CRM company to an AI agent company. It's a very brief history and just a very brief recent history.

all right but AI agents have been around for a very very long time all 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 under change work how they could change work it's like y'all you got it you got to think you got to look at the writing on the wall right and also look at the largest companies in the world right so i i already talked about Microsoft right I already talked about Apple, right? Apple with their Apple intelligence. So Apple and Microsoft, they control the devices that we use. And Microsoft has already all out said, yes, AI agents. They're here. 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 you will start to see some type of autonomous or semi-autonomous workflows in the future with Apple. Invideo, right? I'm going over the largest companies in the world. InVidia, they create, they are the engine, right? They are literally the engine driving AI agents and how we all work in the future. Google, like I said, Google at their, at their I.O. conference announced AI agents, right, in their Vertex AI agent builder. So as they continue to approve, they're very capable Gemini models.

I think Gemini models 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 AI agent builder, again, one of the largest companies in the world. Then you have Amazon, right? Amazon is investing billions of dollars into large language models. They have their own large language model platform, Amazon Q. They're working on simple agendic workflows. And then meta as well, right?

Those are literally the six largest companies in the United States and six of the largest companies in the world. And they are investing their dollars and their people into AI agents in some way, shape, or form. So you have to see the writing on the wall. You have to always follow the money. And the money, the time, the attention is all going toward AI agents. So this will greatly impact how we all work. And it is unfolding now before our very eyes.

All right. Let's talk about a couple example business use cases, right? I hear y'all when you reach out to me on LinkedIn and send me emails. I always appreciate that. And you know, you always say, hey, we got to hear, we got to hear more business use cases, right? So I'm just going to give some examples. So Salesforce agent force, right? So they did a video on this.

We'll leave that video in the newsletter. It was short video, five minutes that kind of shows how this kind of no code or low code agent building works in their agent force. So in this example, you can use your CRM data, you know, know, build the parameters, you build the guardrails. 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 split.

So on the left side, with natural language, you kind of set the parameters in the rules of how your AI agent can respond. And then what happens is it is connected to your live Salesforce data. And then you can essentially create a chat, right? It's funny. Chatbots were so ahead of their time, but they were so useless, right? Because with chatbots, pre-large language models, you had to set all of these defined conditions, which maybe, you know, accounted for like 1% of conversations that might actually happen on a chatbot.

But now it is flipped, I think, with natural language processing in large language models, now you can probably hit on like 99% of all customer inquiries, right? But with agent force, you can essentially tap into your Salesforce data, create a simple agent that you can then put on your website. And in this example that Salesforce had, you know, essentially a customer is asking like, hey, what, you know, they had a question about their order. Salesforce, the agent answered it. And then they said, hey, I need, 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, September 27th? 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. nuanced conversation and translate that to data and then connect it to your database, right? And then the Salesforce agent gives the options and they said, hey, we have the following available times on this date. And then they gave a couple options. Again, the person just says 230. And then that's one of the options. And then the agent force agent schedules it and then

updates the CRM accordingly. Right. So this is huge. Something's, 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, 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 of salt, but then you have similar offerings from, you know, other tools, softwares, like, right, I said Microsoft Copilot Studio. Same thing. But y'all, that is going to completely change how customer service gets done.

Let's talk about sales, you know, kind of sales and marketing. Zapier. Zapier has, they call them AI assistance, but they're really great. Zapier Central, we've talked about it on the show before. Similarly, these are more like semi-autonomous, but with Dragon Drive. So you have to connect to other third party software. So maybe as an example, you don't use Salesforce. Maybe you use something else, right?

But you can build kind of these semi-autonomous agents drag and drop with natural language tapping into a large language model via Zapier, right? So, oh, when we get 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 write it in a way that takes into account all the information they put on the form, right? So you can essentially have a combination setting guardrails, setting conditional triggers, and then, you know, tapping into literally almost any software or service on the internet.

So when you have this, you know, kind of AI powered agent tapping into a large language model and your data, your services, right, you can see the future. Coding. Coding is another one, right? We're talking about real business use cases. I know I asked you all, I got mixed responses, right? Do you want to see more on AI coding, AI software development, right? Because this is changing as well.

And I think that this obviously makes the job easier for people who are already, you know, in software development or engineers, people doing coding, Python, right, et cetera. These tools like Replits agent, cursor AI, right, being able to code with AI, Devin, you know, the AI software engineer from Cognition, right, which OpenAI prominently featured in its O-1 announcements, right? So those three companies alone, there's a lot more, but these are probably three, you know, Cursor AI, Replit agents and Devin 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 more niche and targeted AI agents. But then also, so yes, this changes what's possible for software developers, engineers, et cetera. But then this also gives new capabilities, right? That's the other thing with AI agents. They bring 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 or five prompts, 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 drastically change. So yes, there are general purpose AI agents, right?

But then there are niche or skill specific AI agents as well. And I think that Replit agent, cursor AI, and Devin are great examples of those. benefits, right, here as we're wrapping up. You got to be able to see the benefits, right? This is 24-7 non-stop work, right? If we're talking more on the autonomous agent side. Salesforce itself said they see billions in the next year. Billions with a B.

Y'all, I'm not crazy. People think I'm crazy when I come off with these hot takes. There's going to be more AI agents than human. And people laugh and they're like, this guy's dumb. And then Salesforce says it. Salesforce says within a year they see billions of their agents. Is that part marketing, part, you know, dreamer vision? Sure, right?

This happened at Dreamforce after all. Is it a reality? Absolutely. Could it or sorry, could it be a reality? Yes, heck, yes, it could be. Right? Because now I can go create today 10, 20, 30, 40 agents. You literally have people and softwares that are agents that are creating,

other agents. So literally autonomous agents working around the clock creating other autonomous agents. You have agents interacting with each other. Again, no longer science fiction. Two or three years ago, you know, us dorks were sitting around on Reddit and quorum being like, wouldn't this be cool? And now it's reality, right? So the benefits, 24-7 working with your data on guardrails you set up using your up-to-date knowledge. 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 generative AI, it democratizes. It lowers the bar. So you no longer, you know, 10 years ago, yeah, there was artificially intelligent agents, 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?

All right, we can't talk about this without talking about the challenges and limitations. All right. So the quality of reasoning is huge. Really defining this agent computer interface or ACI, as it's sometime called, in designing effective interfaces for tool utilization, that's a challenge as well. But I think what we're seeing with Agent Force, and co-pilot studio really is huge. And we'll see with Vertex AI 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 AI powered agents. So that's a huge challenge. And also ethical and safety, right? especially as AI agents start to learn and adopt, or sorry, learn and adapt, right? And then as we start talking about multi-agent environments, again, I'm not a doomsdayer out here talking every day about Skynet and Terminator, but you got to think about that, right? Hey, you set up some maybe weak guardrails and then you, you know, set out a system of, you know, 50 autonomous AI agents that can all talk with each other, powered by as an example,

maybe OpenAI's O1 model, right? Bad things could in theory happen, right? If you don't have enough human in the loop, which I know, y'all, let's be honest. Sometimes that is an exaggerated way of saying, hey, us as 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. All right.

If you don't constantly have humans overseeing them, if you don't have strict guardrails in place and constantly doing QA on these agents, the future can also be very scary, right? So I don't want to skip over bias, stereotypes, safety, ethics, right? Not even talking about job loss because I don't care what anyone says, AI is ultimately going to take away way more jobs than it creates.

Sorry. It's not me being a pessimist. That's me being a realist. That's why the largest companies in the world have all invested billions of dollars into generative AI, into GPUs, into AI agents, right? Yet they're laying off tens of thousands of employees with record high profits, right? I don't know why people don't, you know, I know we need to be optimists as human beings when we talk about AI 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.

And Wall Street is really going to start to love AI agents, right? Keep an eye on Microsoft stock, Salesforce stock in the next year or so. You'll see what I mean. All right. Then the future and what's next. Well, I don't know what's next. All I know is it definitely has to do with AI agents. Like I said, out of the top six companies in the U.S.

So aside from Apple, every other company, Microsoft, Nvidia, Alphabet, Amazon, and Meta have all come out and publicly said that they are either investing in or investigating AI agents or they are all in. Right. And then we talked about Salesforce as well, one of the largest companies in the world that touches so many other Fortune 500 companies, right? Like the majority of large enterprises are using Salesforce. I don't even know who Salesforce's biggest competitor is, 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. All right, let's wrap this up, y'all. So some key takeaways here. We are on the cusp of this big shift from, yes, AI to generative AI to AI agents, right? And, you know, you can't really talk about that shift without talking about AGI and, and, you know, our more capable models and AI agents, kind of one of this next step that gets 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 AGI or not, it doesn't matter. But we have to talk about that. So we also have to talk about the duality of AI powered agents. If I'm being

honest, I was shocked over the last week, that all of this is happening all at once. And that's why I said, we got to do a show on this, right? I think a year ago was still too early. But now the writing is on the friggin wall, y'all, you got to pay attention to it. All right. And we have to talk about the duality. And we have to business leaders and decision makers. You have to do AI agents and kind of this implementation in the right way. You have to prioritize humans. You have to to prioritize safety, guardrails, data, et cetera. You can't skip over this in the mad rush to be the first big company to implement something from Agent Force to implement, you know,

the biggest, you know, AI agent from Microsoft Copilot. You don't have to do that. You should be acting now, right? If you don't already have generative AI in place in large language models, you're kind of screwed, right? You should already be having that conversation about these autonomous workflows, AI 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. All right. So let's wrap. Got a couple of questions here. Let's see. Let's see.

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 AI and we're spending so much time now talking to AI, I think it's important to have a human conversation. So, you know, podcast audience, you might get kind of annoyed when I, you know, ask questions so much of our live 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-to-human conversation, right? I want this to be a place where we can explore and learn together. So a question here from Kobe. Thanks for joining us. He says, I'm curious about your thoughts, Jordan, on building agents inside of Zab Central. Kobe was, you know, reading ahead. have you built any and have you found Zapp effective? So when Zapier Central first came out, I went in and

played around with it. So I did build one or two basic workflows. To tell you the truth, I haven't gone back there since. And I'm kicking myself because I really should be. Right. We pay for Zapier every month and I'm barely tapping into it. But I do think, especially for companies that maybe don't have Salesforce 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 $2 per conversation. I'm not sure about that go-to-market strategy, but that's why they probably have a bunch of smart people telling them that that's the way.

But I do think Zapier Central and Microsoft Copilot Studio, their AI agent builder, are probably the two most robust kind of AI agents that are ready to go. I do think end-to-end Microsoft Co-Pilot Studio is obviously a little more capable because it works with your real-time data out of the box, whereas in Zapier Central, if you're building kind of these agentic flows, you know, you are, I won't say duct taping it, right?

It's a little more seamless when it is built into the software that you're using on your computer versus when you're having to kind of put the pieces together. All right. Marie asking, devil's advocate here again, if AI 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. and I kind of wrapped the show as much, right, talking about that.

What is the future of work? And I think it is a more responsible use of humans in the loop. 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 large language models across their organization, top to bottom, you know, as they figure out AI agents, I think so many companies are going to rush to just fire employees. by the thousands. We've kind of already seen that, 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 pessimist. I'm not being a pessimist. 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 this being commonplace having, you know, four-day work weeks as an example, right? Who knows, right? But I don't know what the future of humans role in this as large language models get smarter and more capable and as we see AI powered agents. But I do know what we've traditionally done in the past decades where you are rewarded for your domain expertise, right? You're rewarded for all of these facts and decision

making that you have in your head around, you know, your specific industry. I'm not saying that that's going to be gone, but that is going to be greatly deprioritized. 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 company's data, when they can access the, you know, entire history of the world, in theory, right? And they can adapt and access your company's data and learn and change, right? That domain experience you have up in your brain all 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 30 years ago that essentially every single knowledge worker here in the U.S. would be working, quote unquote, 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 co-pilot, whether that's, you know, agent force or, you know, from Google, vertex AI agents, whether you're building your own AI agents, I will assume the majority of workers in the coming years will be working around generative AI in large language models in some way, shape, or form, just like we're all right now working around the internet. Last question, and we're going to wrap it, Tara asking, how can we ensure fairness in detecting

AI agent use in the workplace and schools and prevent false accusations of relying on AI without proper evidence? That's a great question, Tara, and I have no clue. I have no clue. Maybe this goes to the question that Marie just said, what are humans going to do? Well, humans, we humans, when we need to prioritize ethics, safety, getting rid of bias, right, fairness, equity, inequality, right? These are the types of things as AI agents and large language models take away maybe the majority of our manual mundane knowledge work tasks. These are the big issues we have to cover.

Great question, Tara. And don't worry, we're going to be here every single day helping all of us uncover those next steps of how do we work in the future? Well, at least you 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 Everyday AI. Thank you for joining us. If you haven't already, please go to Your EverydayAI.com. Sign up for the free daily newsletter.

We're going to be recapping today's 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 this 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 Salesforce as well. But all I know is the future of work is generative AI. And the largest companies in the world are going all in on AI agents.

Thanks for tuning in. We'll see you back tomorrow. And every day for more everyday AI. Thanks, y'all. Meet Firefly AI assistant. Now live in Adobe Firefly, the Allman One Creative AI Studio. Just describe what you want to create in your own words. and the assistant handles the rest, orchestrating multi-step workflows across Adobe Creative Cloud

apps, including Photoshop, Premiere Express, and more in one conversational interface. You direct the outcome while the assistant accelerates execution. Stand control with the ability to step in and refine at any time. See it today at firefly.adobie.com. And that's a wrap for today's edition of Everyday AI. 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 EverydayAI.com and sign up to our daily newsletter so you don't get left behind. Go break some barriers and we'll see you next time.