Everyday AI Podcast – An AI and ChatGPT Podcast - EP 402: OpenAI’s AI agent revealed: What you need to know about ‘Operator’ agent from ChatGPT

Episode Date: November 14, 2024

Newsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion: Ask Jordan questions on 'Operator' from OpenAIUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:Technology Landscape with OperatorRapid Development and Impact of AIRole of 'Operator' Agent from OpenAITransformation in the Tech IndustryDiscussion on New Software ToolsConcerns and Risks in the Field of AIPotential and Future of Agentic AICompetition and Industry Movement in AITimestamps:00:00 Operator details from OpenAI03:40 AI News: iPhone gets Google AI, AMD cuts jobs.08:52 Agentic AI prioritization for OpenAI's long-term future.12:49 GPT models remain unmatched; anticipate future advancements.14:36 Powerful AI: performs tasks autonomously, manages workflows.20:42 Google leaked AI assistant Jarvis, then retracted.24:13 Large language models' affordability: immense value accessed.25:03 Companies collect data to improve agentic AI.28:29 AI is taking jobs faster than creating.33:34 Establish guardrails for autonomous AI in enterprises.36:56 AI will transform work; Microsoft leads integration.39:40 Magenta 1: polished, easier setup, multi-agent workflow.43:40 Agentic AI risks: rogue behavior, significant threats.44:33 Exploring new advancements in generative AI.Keywords: Jordan Wilson, agentic AI systems, tech companies, Anthropic’s AI system, Google Jarvis, Microsoft's autonomous AI agents, Copilot Studio, Salesforce Agent Force, release intervals, large language models, transformative shift, AI job impact, AI-related skills, corporate interests, AI's impact on jobs, OpenAI Operator, AI automation, Google Gemini app, AMD AI focus, NVIDIA, Everyday AI show, workplace automation, AI safety, AI governance, AI agents, business preparation, Microsoft AI solutions, AI risks, AI developments, Anthropic’s computer useSend 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 in 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. Details about open AIs, agentic AI, have just been released.

It is reportedly called operator. And I think that this new offering from Open A.I reportedly set to release in January, coupled with a lot of things that have been going on in the last two months, are going to completely change. how we all work. And I know I say that now and then here on the Everyday AI show, but I think this time you really need to pay attention. And I'm going to tell you why. Well, if that sounds exciting to you, if that sounds a little scary to you, don't worry, you're in the right place. We're going to demystify it all today on Everyday AI. What's going on, y'all? My name is Jordan Wilson, and I'm the host. And welcome to Everyday AI. This is your daily live stream podcast and free daily newsletter,

helping everyday people learn and leverage generative AI to grow their companies and their careers. So if that sounds like you, like, oh, I'm going to grow my company and career by learning AI with the latest updates every single day. You are definitely in the right place. If you are joining us on the podcast, thank you. As always, make sure to check out your show notes because you can always find a link to our website there. That is where you go to take action at your everyday AI. because you can listen every day. You can try to keep up, but we do all that hard work for you.

So make sure you go sign up for our free daily newsletter where we recap not just every single episode, every single day when we bring on great guests from the biggest companies and kind of great minds growing AI. But we just keep you in the loop with everything else that you need to know in the AI world. So make sure you go do that at your everyday AI.com as well, let's go check out. I don't know. We got like 400 plus free episodes. You can go learn about AI's impact on.

any sector. It's all on our website. All right. So I'm excited to talk today about open AI's new AI agentic system called operator. It's not out yet. Details were just released. We're going to get into that. But before we do, let's first go over as we do every single day, the AI news. So a standalone Google Gemini app has just been spotted as in minutes ago on iOS, bringing a new live AI assistant into the mix. So a standalone Google Gemini app has been spotted on the iOS app store, marking a shift from its previous integration within the Google app. So the development could signal a new phase in Google's AI accessibility on iOS devices. So previously, Gemini was accessible through Google's app, Gemini tab, or via a home screen widget that was updated in September to include

direct shortcuts. But the standalone app features a Gemini icon on the home screen, promoting the tagline, unlock the power of Google AI on your phone, and offers access to Gemini Live, a feature not yet available in the Google app. So the app makes use of live activity allowing users to interact with Gemini while continuing to use their phones, enhancing the user experience with real-time capabilities. What all that means is, you know, hey, even as iPhone users, which so many of us are, I'm an iPhone user. You could kind of use Google Gemini AI on your iPhone, but the big feature, the live neural voice that you can talk to in real time, like ChatGBT, GBT's advanced voice

mode was not available to iPhone users until of literally a few minutes ago. So speaking of the newsletter, I'll probably be reviewing the new Gemini live feature today. So make sure you go check that out. Our next piece of AI news, AMD has cut a thousand jobs to focus what they say is on AI growth opportunities. So chip maker, advanced micro devices or what everyone knows them as AMD, announced it will lay off approximately 1,000 employees, which represents 4% of its workforce to strengthen its position in the AI processor market. So the layoffs are part of AMD's strategy to align resources with its largest growth opportunities, particularly in the AI sector where it faces stiff competition from

Nvidia Corporation. So AMD is the second largest manufacturer of graphics processing units, those GPU chips that make AI work. Yet it remains significantly behind Nvidia, which dominates according to estimates anywhere from 80 to 90% of the global GPU supply. Yikes, no one's catching Nvidia. So despite AMD's efforts, its stock has decreased. by 5% this year, which basically anyone working in and around AI has grown over the past year,

except AMD because their biggest competitor, InVIDIA, is running away with everything. So, yeah, AMD's stock has gone down 5% while Nvidia has surged by over 200% reflecting the competitive gap between the two companies. And our next piece of AI news, kind of not really a personal one, but a little bit. I will be, well, the show will be joining you live next week from Microsoft's Ignite conference happening in Chicago. A lot of big news expected. Make sure you tune in because I will be live. And it just so happens, y'all, that an embargo is lifted at the exact same time our show goes live. Look at that. So make sure you tune in next week. And I think that's,

on my birthday as well next week, right? So we will be bringing you literally some of the biggest news from Microsoft live from the conference the second it is released. So make sure you tune in next week for that. All right. Let's get into it. Let's talk about Open AI's operator in what this means. And hey, good morning, live stream audience. It's good to be back. I took a little time off That's why, yeah, if it looks like I fell asleep in the sun to our live stream audience or maybe our podcast audience, it sounds like I'm a little rasp. You know, I guess I'm used to talking about AI every single day, but when I take a couple of days off, you know, I get a little rusty. So, hey, our live stream audience is good to see you back. Everyone from Michael and Jackie and Brian, Marie, Christopher, David, everyone else.

Thanks for tuning in. So let's talk a little bit about operator. So this is a, it all started just a couple of hours ago, a Bloomberg report saying that Open AI is nearing the launch of its AI tool to automate task for users. So right now, this operator is a codename. All right. So, you know, don't expect a tool called operator. Similarly, you know, the 01 model was first Q-named QSTAR, then it was co-named Project Strawberry. and then it obviously was released as OpenAI01,

because kind of the new reasoning model. So this operator is probably not going to be what it is going to be called. This is just right now the reported code name for operator. So what the heck is it? What is this operator? Well, according to reports, it is an AI agent that acts more like a personal assistant. It uses a computer to perform multi-step processes with minimum human intervention, and it performs actions on your behalf, right?

That is the key mark distinction between an AI chatbot and an AI agent. So an AI agent usually will perform multiple steps with one of those steps being an action taken on your behalf. Okay. And I think early agent. are going to be much more limited, right? And we're going to talk just here in a second about the broader landscape, because I think that's extremely important to put into context.

But the biggest thing here is Open AI, they were technically the first one. We saw reporting on Open AI's investment in belief in agents about a year and a half ago, right? Some of the initial reporting, even before the new 401 model or the 01 models were even released. We saw a lot of early reporting in late 2023 and early 2024, just how big of a priority, agentic AI and AI agents were both to OpenAI CEO Sam Altman and to the company's long-term future at large. Right. So, but the biggest thing, if you are confused and you're like, what, what's with all these agents? What is an AI agent? Isn't that just a chatbot? No. So the biggest, like I said, the biggest distinction is an agent can perform multiple steps and ultimately take an

action on your behalf. So that means that this agent will have, in many cases, either access to a browser, access to a virtual computer, access to your third party data, right? It depends on what company you're using, what agentic AI that you're using. But that's the biggest differentiator, right? Because we're going to go into some details here. But I wanted to just. kind of draw that line in the sand because I think as large language models become more robust, more powerful, more features, more integrations, it's going to be kind of not difficult, but I think that the average person can kind of get confused, right, when we're talking about large language models, generative AI versus agentic AI. So that is the big differentiator.

an AI agent can take multiple steps with minimum human interaction and ultimately perform an action on your behalf. And you might be wondering, well, Jordan, can certain AI chatbots kind of already do that? Yes, in theory, they can, right? Even in open AI, you can make very limited or you can take very limited actions inside of chat GPT as an example, right? So if you're using GPTs, right? So there's millions of GPTs, 99% of them are garbage, but one percent of them are great. Right. So as an example, a very flexible GPT that you can use right now is Zapier.

Right. So right now, with a lot of duct tape and a little bit of MacGyvering, you can actually do a lot of kind of agentic workflows right now by using different GPTs. And if you really have a grasp hold on marketing automation, you can already do a lot of these things. But this is moving there by default, right? All right. Let's keep this thing going.

What is this coming? Well, according to reports, January 2025 is the date to look out for. And it will be released as a research preview. Again, this is according to reports and through OpenAIs API for developers. All right. So similarly, I do expect it to be. released in a similar way as Claude's computer use was released. So more on that in here in a minute. And the announcement for operator reportedly was made during an internal open AI staff

meeting on Wednesday. All right. So this is all, you know, very new, very fresh, but it is moving very fast because January is like what, a couple of weeks away. So if this is actually released to the public in January, we're talking weeks away. We're talking weeks from how your business does business, y'all. The fastest movers will win. That's how it always is. Also, take that January, not deadline, take that January date with the world's largest grade of salt, okay?

Because let me just say this. Open AI is the leader in the AI space. they are the leading large language model maker. Their researchers are by far, and their technology is by far the most powerful in the world. It is not even close, right? I know people like to say, oh, you know, what about Claude, 35, Sonnet, New, and all the new offerings from Gemini. Yeah, they're great models, but still, no one is able to touch GPT40 in the O1 model. and GPT40 is based on technology that is almost now two years old, right?

Just wait until we get a new model, right, from OpenAI, a new GPT series model, whether that's 4.5 or the reported GPT5 or Orion, right? Wait until we get that with agentic, you know, with an agentic AI agent, maybe called operator. And I'm sure it's going to have some of that 01 kind of reasoning, kind of model powering it as well. Wait until we get all that under one roof. This is going to fundamentally change the way that we all work,

whether you want to realize that or not. And hey, live stream audience, if you do have any questions, it's good to be back live. It's good to be back live. Yeah. If you're listening on the podcast, you might not know this, but we do this live. This is unscripted, unedited, right?

So if you do have any questions, make sure to get them in now. I've missed answering your questions live. All right. So let's talk about some of the capabilities and example use cases of operator. So it can write complex code. So that's the thing you also have to keep in mind. Do not expect this operator or whatever it is ultimately called to, you know, be an archaic product. It is powered by OpenAI, right? So again, we don't know if it's going to be running GPT4O, if it's going to be running 01, if it's going to be running a combination. But regardless, this agentic AI is going to be powered by the most powerful technology in the world.

So it can do all the normal things that a large language model can do, but it is giving it that multi-step ability, kind of that chain of thought reasoning that happens under the hood. And then it can execute and perform tasks on your behalf. So as an example, maybe you can book flights, book hotels, or transportation based on your preferences. It can manage your schedule or finances. It can integrate with existing tools and your existing workflows and handle complex labor intensive tasks autonomously. That is the big thing.

Think of it like this, y'all. Airplanes, right? I was on a lot of airplanes recently on, yeah, like I said, it was great to get away for a week. or two with my wife celebrating our 10-year marriage anniversary in Australia. So I was on a lot of flights. And I actually used to be, you know, way back in the day, I used to be roommates with a couple of pilots. So one thing I didn't know, I don't know, until like 15-ish years ago, which is kind of common sense now. Maybe I didn't know because I was early, you know, early in my 20s. But I didn't

know that essentially pilots, they don't do a whole lot of manual work. work, right? They kind of assist during takeoff and landing, but for the most part, much of a flight is autonomous, right? It's kind of human in the loop. And if I'm being very honest, that's what I see the future of work as. I see it kind of like a pilot. We are all going to be pilots with many autonomous AI agents being our kind of co-pilots, right? Which, by the way, I hope there were reports Microsoft was going to stop using or stop leaning into the co-pilot naming, which I hope they don't, by the way, but that's what I think the future of work is ultimately going to be, right? Us humans are going to have many different agenic AI systems at our fingertips, and we are going

to kind of help during takeoff, and we're going to help during landing. So we're going to give access to the right AI agents with the right tools, with the right guardrails at the right time for the right project. We're going to kick it off like a pilot. We're going to sit back, you know, kind of look and watch and make sure that our agetic AI does the job. And then when it's time to land the plane, we're going to check and make sure everything's good. We're going to put our kind of hands back on the wheel or back on the keyboard and the mouse,

right? Or automating it with your voice. That's the future of work, y'all. And I know it might seem a little weird, right? And it's more complicated than that. But agenic AI is the future of work. All right. Let's talk about some competitors.

Because this is not just open AI. All right. This is not just one company saying, all right, we're doing this agent. AI thing. We're going all in on agents. It's already available, right? Anthropic released their computer use in October, so a couple of weeks ago. Pretty, I'll say it's pretty promising. We'll say that. Right. Excuse me, y'all. Like I said, my voice is a little, I got to re-you know, re-warm up my voice apparently. Geez, you don't talk for like two weeks and then your voice stops working, apparently.

So we have Anthropics computer use, which we actually did a dedicated show on. We showed you guys, you know, for the non-technical user, how you could install computer use. But this is the first agentic AI from a quote unquote name brand company that is released in the wild. Agentsic AIs is not new either, right? There's been some great early, kind of early companies like Langchain. There's a lot of them, but a lot of them were, you know, a little less robust at the time of release. With Anthropics computer use, it's a working agentic AI. It's very limited right now, right?

In terms of what you can do, the guardrails are super heavy, but it's out. It's working right now. Go watch that episode. That's one you might want to watch, right? Yeah, maybe you normally just listen, but we did literally a step-by-step tutorial. So if you want to see kind of that archaic version of an AI agent, so first of all, you got to tip your cap to Anthropic for putting it out in the wild before it was super polished, but agentic AI is here.

You can literally sit here and type to a large language model and go watch it do work. Watch it browse the web autonomously. Watch it summarize articles. Watch it create, you know, new content, new documents, you know, based on research that it is doing, right? Multi-step executing tasks for you. It's already out with Anthropic, right? So you have to do a little bit of technical work, right? It's not something you just log into claw.

dot AI or you know the Claude's uh you know desktop app and is there it's it's a little bit of technical work but we walk you through it another competitor Google Jarvis yeah yeah Jackie we'll see uh Jackie said uh you know joking about the co-pilot naming she said how about Jasper because there was a company called Jarvis already in the AI space they faced a little bit of a lawsuit from Marvel and had to change their name to Jasper from Jarvis but Google kind of released slash leaked slash took back their kind of agentic AI assistant that's called Jarvis. So Google kind of did confirm that they have a kind of working agentic AI that was released via a Chrome extension.

But they pulled it. So we don't know if it was supposed to, you know, be a leak if it was supposed to be an early release and was just pulled because of issues. but it was up for a very short amount of time last week. But again, there we go. Google reportedly will be releasing this before the end of 2024. So presumably in December, they will be releasing, similar to what Claude has,

an AI agent that will work in its Chrome browser and complete multi-step tasks on your behalf. 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 pro-grade tools across Adobe Creative Cloud apps, including Photoshop. Illustrated 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.adobie.com. And then, and this is y'all, this is everything except OpenAI's operator. I wanted to give you a little bit of the landscape here. And last but not least, Microsoft's autonomous AI agents within Copilot Studio. So we saw this in October as well. Yo, this is all happening all at the same time. I should have thrown in Salesforce's agent force, which is now out and live right now.

But that's more niche, right? It's just for, you know, Salesforce customers. But I know that's probably hundreds of millions of companies or hundreds of million of people that may have access to agent force from Salesforce. But now you also have Microsoft's autonomous AI agents, which are set reportedly set to be released November. 19th in public beta. That's my birthday. Yay. So this is all happening at once very, very quickly. And this release, y'all, the release of all these agentic AI systems from the big branding companies, they are all smack dab in a couple of week window where that's not how it was with large language models, right?

I think we got what Google Gemini like 10 months after chat. ChbT was released to the public, something like that, similarly with Claude, right? You had this for chatbots for large language models. You had this multiple quarter kind of weight between all these big releases. AI agents are coming so quickly and they're coming at the end of 2024, all at the same time. Weird. I think I said that last year. Oh, wait, I did. Because even a year and a half ago, I said, we are going to all be using AI agents on a daily basis by the end of 2024. Did that come to fruition? Kind of. I mean, we'll see, right? But there was no released, no confirmed AI agents.

And a year and a half ago, I'm like, y'all chatbots are just the beginning. Large language models are just the beginning. This is a precursor to autonomous AI agents, right? We always wonder, how are these large language models so cheap? If you really know what you're doing, you are getting immense value out of large language models. And you're like, why? How are these so cheap?

I've said since day one, I would pay hundreds or thousands of dollars for something like chat GPT if that was the only thing to do, right? If it was zero large language models or paying. hundreds or thousands of dollars a month for a very capable state-of-the-art model, I would still pay that because it's worth it if you know what you're doing. Most people don't. Most people don't know how to use large language models. Most people don't know how to prompt.

They're extremely valuable. So why have they been so cheap, almost free, right? And even the free accounts are great. Well, it's because those companies need all of our data. They need all of our training data on what we like, what we don't, right? If you're a heavy, large language model user, you probably see sometimes, A model will give you two different choices, ask you to pick one. You know, there's always the kind of, you know, thumbs up, thumbs down when you're using AI chatbots.

These companies need all of our data, all of our feedback to make their agentic AI, right? Because once agentic AI is actually good, I think the first iterations, that's, that's why they're releasing it, you know, for kind of developers. They're releasing it via APIs. and that part's huge because that means hundreds or thousands of name brand pieces of software that are already using this technology on their backend, right? We say, oh, everything's powered by AI. Everything's powered by AI. These companies aren't building their own AI.

They're just using either Claude from Anthropic. They're using Google Gemini or they're using the GPT from Open AI. So once agenic AI is very good. changes how we work. And I think that starts November 19. That starts in a couple of days when Microsoft's AI agents are reportedly supposed to be making a beta appearance to the general public. So what's the impact? What's the impact of these AI agents?

Well, pretty fascinating study that was just released in the Harvard Business Review. on the impact of chat GPT on automation prone jobs versus manual intensive jobs. Maybe we did share about this in the newsletter the other day. So that's why you got to be reading it daily to keep ahead, even when I'm on vacation, you know, making sure our newsletter goes out with the most up-to-date information. But a pretty telling study, pretty telling study. All right. So this study said essentially,

genitive AI is causing significant job displacement, especially in writing and coding by automating tasks. It also is talking again, this study, and I'm going to put it into perspective here. It's talking about the impact of chat GPT on jobs, essentially, right? Automation pro jobs versus manual intensive jobs. So demand for these jobs has dropped sharply, sharply with increased competition and complexity. However, AI-related skills are becoming more valued, leading to new job opportunities, and companies must invest in upskilling to kind of get with this shift. All right.

So let me put it in real words. New study published in Harvard Review Journal said AI is taking jobs. It's automating jobs, right? Everyone all along has said, AI is going to create more jobs than it takes. No, it's not. Sorry, right? I'm not an AI pessimist, right? I am an AI realist.

And I'm not going to go into a deep dive on AI and jobs. I do that, you know, maybe every six months. It's not what today's episode is. But a very reputable study just came out that said, oh, you know, everyone all along, you know, economists and the World Economic Forum and the International Monetary Fund, All these big organizations said that AI is going to, you know, create significantly more jobs than it takes away. I've always said absolutely not, right?

I think organizations, research institutions, you know, they like to paint that picture because they want people, especially here in the U.S., they want everyone to be upbeat and optimistic about the economy, right? They want people to feel secure in their jobs. The reality is AI is taking jobs faster than it is creating them, which if you are a long time listener of this show, I've been saying this literally since day one. The day one thing is is not an exaggeration. Literally, that was the first episode that we did nearly two years ago was how AI is going to take jobs faster than it is going to create them.

And all the studies out there were essentially wrong. So now we are finally, you know, more than two years into this large language model journey, we are getting reputable, hard, concrete information that, yes, AI is taking jobs away faster than it is creating them. ChatGBTGBT is essentially automating jobs that, and we are not creating new jobs fast enough. Hint, I'll save you an hour. You can go back and listen to the episodes, but corporate greed. That's why AI is not going to, you know, create more jobs than it can replace.

it's like, oh, we can automate a bunch of jobs and our stock's going to go through the roof and we can lay off tens of thousands of employees and we're going to continue making money. Sounds great. All right. So why do I tell you that? This is a shift in how we all work. Is the first version of Open AI's operator going to be a seismic shift? Absolutely not.

In the same way, computer use right now from Anthropic, it's not that good. Right. It will become better as more companies start using the API, as these companies start getting all of our training data, as they start getting feedback building on their roadmap, it will get better. But if you think large language models where the shift that changes the future of work, you are wrong.

It is the seed, all right? That we've all been providing water and we've been providing sunlight to this seed. But it will grow. All right. A couple, couple hot takes, y'all, and then I'm going to wrap this on,

wrap this up. So if you do have a question, I'll get to, I see you with a couple. So I'm going to get to them here at the end here. All right. So what is the impact? Well, I think agentic adoption

will actually be slower than large language model chat bots, even though they're much more powerful, right? Ultimately, for the past two years, you know, even bigger companies that have invested decently in large language models, they essentially say this. They're like, okay, well, how do we measure the return on investment? And hey, sometimes, you know, this just creates more work for the human.

Well, number one, those companies don't always know what they're doing. They're listening to bad advice out there. They're not listening to everyday AI where we bring you the leading experts in the world that tell you how to get return on investment on AI. initiatives, but they don't always know what they're doing. But ultimately, what the market has wanted for two years, whether they know or not, is agenetic AI. Because everyone's like, okay, cool, these large language models are great, but then what, right? We still need humans, you know, to intermittently be inserting themselves into all of these different tasks, right? And

some companies that have actually found success with AI are like, well, now are humans that are great at A.I. Teams and companies that have, you know, made great efforts into large language models, investing their data, you know, governance, security, guardrails, all these things that are like, wow, now we're just more productive. We're just busier, right? This AI implementation is a growing our company, absolutely, but it's growing the tasks that are required from these humans. Well, guess what? Agentic AI is not going to be like that. You got to train it. You got to get it set up. But once you do, again, think of that analogy. We're going to be like pilots.

Launch, keep your eyes on it, land. All right. Kind of hot take number two. Safety and guardrails are going to become increasingly more important. If you think safety, guardrails, governance, if you think those are important just by connecting your data to generative AI systems to large language models by giving your company's access, yeah, wait until agentic AOP, right? When you are handing the keys over to, you know, open AIs operator or to autonomous AI agents

inside Microsoft, inside of their co-pilot studio or, you know, Jarvis, right, from Google, or once Jarvis gets rolled out into their suite of products, I think guardrails are important now. What about when you literally hand the keys over to agentic AI and it's executing actions on your behalf, it's making decisions, it's starting to land the plane and take off and you're just assisting it. I think these things are important now. This is where companies need to be investing time, money, and resources, right? You need to set your company up like pilots, create new positions. for humans in the loop because how we work, the very nature of work is changing.

And enterprises, like I said, you have to really emphasize human in the loop and rethink what jobs even mean, right? Yeah, I think the majority of jobs are going to be impacted. I'm not saying replaced. But the majority of just jobs, if you are a knowledge worker, if you sit in front of computer most of the day, right, and perform a task using your brain, your job is going to be greatly impacted by Agentic AI. So C-suite boards, business leaders, you need to be thinking right now, redefining what jobs

even mean, right? Because if your organization wants to be competitive, if you want to stay ahead, you're going to be using Agentic AI in 2025. So you need to start rethinking. now because Microsoft is reportedly releasing something in days. You've got to be ready. All right. Let me get to a couple of the questions that we had. So as bowels or blows, sorry, joining us from YouTube says, what happens when someone tricks an agent into thinking? It's you giving them instructions. Large language models have proven pretty easy to gaslight with a little

creativity. Yeah, absolutely. Right. I know I talked a little bit more about kind of some example use cases of agentic AI. I talked about, you know, use cases, examples, capabilities, right? I didn't talk about the dangers. Yeah. There's extreme dangers of agentic AI, right? What's it mean for your company, for your brand? What happens when and if AI agents go rogue? Because they will, And that's ultimately why right now, at least in early iterations, they're working in virtual sandboxes, right, like Anthropic Clause computer use, right? It doesn't have kind of full capabilities to your actual machine, to your data, right? Because you need full control over the ecosystem, which is why I think keep your eyes in your ears

open for Microsoft's announcements next week at their Ignite, because that is autonomous AI agents working within the ecosystem, right? You're not going to have anything as robust, I think, early on from companies, even though they're going to have impressive, impressive AI agents. You're not going to see, I think, high usage from companies like Claude, even chat GPT with their operator, even Google with their Jarvis. I think you're first going to see it within a copilot studio within Microsoft, right? because that's where organizations run.

And those organizations that go all in and that really elevate security, guardrails, data compliance at the highest levels and take advantage of agenic AI that will soon become available, those are the ones that are going to win. But yeah, it's going to change how we all work because there is inherently much more risk now using cutting edge technology than there has ever been, right? You remember it took some companies, you know, quarter. or years to adjust to storing all of your information on the cloud, right? But now you can't get by without it.

It's going to be a lot more impactful when we talk about agentic AI. Jackie says, I'm curious actually how this will work, like Zapier, where it just goes when you give it a workflow. Yeah, we'll see. We'll see, Jackie. Right now, the report is it will, either have access via the browser or a virtual environment. Like I said, I think aside from co-pilot studio, I think early iterations are essentially

going to live in a browser or live in a virtual environment. I don't think necessarily, I mean, we'll see. I could be wrong. I don't think when this operator from OpenAI is first released, I don't think you're going to go to chat gpt.com or, you know, access it via the desktop app. I don't think the first iteration will be like that. I think it will be in a more controlled environment like a virtual desktop because it's easier to pull the plug on something like that. It's easier to have agentic AI that is less likely to go off the rails, so to speak, right, off the guard rails that you set up when it is in a sandbox, when it is in a more controlled environment.

Michael, similar question says, do you think it'll be similar to computer use? Yes, I think early on operator will be similar to computer use. And just FYI, right? This is just according to all the reports that are out there. Yes, I have contacts at open AI. No, I'm not getting at least right now, you know, any exclusive information that's out. But I've talked to a lot of people from all the big different companies. So I do think the first iteration will look like a slightly more polished version of computer

use by Anthropic. I think it'll work better. I think it'll require less. technical setup. But I don't think at least the early iterations are going to be kind of in the full chat GPT ecosphere, if that makes sense. Good, good question, Michael. So Michael's asking, what's the difference between Magentic 1? Is this easier to set up? So yeah, Microsoft actually released a product called Magentic 1. Magentic 1. You have to download it from a GitHub repo, similar to Claude Anthropic.

So I think Magentic one is kind of a multi-agent workflow that works similarly to Claude's Anthropic computer use. I actually think it's a little better. I haven't gotten the chance to personally play around with it. So again, you're going to see a lot of these companies put out very similar software in different use cases, right, to collect different data or to try to capture users in a different way. A great example is Google, right?

Look at them. They have their Gemini chatbot. They have their Gemini for developers. They have Notebook L.M, which is essentially, you know, a state of the art large language model with built in Rack, right? Retrieve long minute generation. Then they also have their new learn about, which is something in between, right? I think you're going to see from all the big companies, you're going to see both kind of large language model products,

Agentic AI products, different names, different features, different functionalities, because they're trying to capture different data or users in different ways. So Microsoft did just last week release something called Agentic 1. It is much more developer focused from being honest. Aside from research and data, I don't know what role Magentic 1 from Microsoft will have in the long run. Right. Maybe it's just a way to collect data to help, you know, better tune into better train its co-pilot studios. We'll see. Someone S. Blows also asking, can you talk about the risk of agents? Yeah, absolutely. The risk of agents is they can go completely off the guardrails, right? Yeah, it's not, it's not difficult to, I mean, you talk about things like prompt injections. right, you talk about, you talk about things like jailbreaking, large language models. This technology is still relatively new, right?

Think of computer viruses. Think of malware, right? Think of Trojan attacks, right? Things that the IT world has been dealing with for decades that are still huge concerns, ransomware, cyber attacks, right? you're going to have your own kind of agentic AI versions of these new viruses malware that pop up, right? You are. That's why I think early iterations of these are in very controlled environments.

But I think we are going to see essentially a whole new breed of problems and issues that we don't yet understand when it comes to agentic AI. And yes, that's where I think a lot of new jobs and new roles within enterprise. companies are going to be created, essentially you're going to have your guardrails team, right? And your guardrails and what that means is going to be much different for agentic AI than it is for large language models. Because again, when all of a sudden now, whether you're just using agentic AI internally, whether you give that access to front end customers, potential clients, right? Maybe they're interfacing with an agentic AI on your website and they can change things in their

account, right? It can go rogue, right? There was a something, I didn't get a huge chance to look into it, but there was a report yesterday that a Gemini's model kind of went off the rails and kind of told the user to die. Yikes, not good, right? Large language models, AI can go off the rails. So when you take it a step further and you go into agentic AI, the risks are even greater because Agentsic AI is actually executing tasks on your behalf or your customer's behalf or your team's behalf or your organization's behalf. So is there a promise that is potentially unimaginable by using agentic AI? Absolutely.

But there's two sides to that coin. All right. Tackle the couple of questions. I don't want this to turn into a two-hour episode. I hope this was helpful. Taking a look at Open AIs. AI agentic, kind of some of these reports just coming out, very fresh called operator. So we're taking a big step in a big shift in the generative AI world in the large language

model, you know, kind of sphere over the last month. In the next month, I think is going to be just as exciting. So again, according to reports, this is going to be coming out in January. We already have a very primitive, but again, tip of the cap to Anthropic for releasing computer use. We have Agentic AI from Salesforce. We're presumably going to be seeing some announcements from Microsoft next week. Agentic AI is here, not just from Open AI, but from everyone else. All right, I hope this was helpful, y'all.

If so, please repost this, share this with a friend. If you're listening to all the podcasts, please click that subscribe or follow button on the podcast and leave us a rating. we would super appreciate that. But maybe you listen every day. Maybe you read our newsletter every day. And maybe now you're the go-to person in your organization for Everything AI. Don't be a jerk, y'all.

Share it. Share the wealth. Share this with your co-workers. Share this with your friends. And please go to your EverydayAI.com. Sign up for that free daily newsletter. Thank you all. It's great to be back live.

And I hope to 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.adop.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.