Everyday AI Podcast – An AI and ChatGPT Podcast - EP 613: AI Agents: From automation to super agents. 10 AI Agents you should know in 2025

Episode Date: September 18, 2025

There's hundreds of agents. 🤖Most you should ignore. These 10 though.... you've gotta know them. Today, we go over Agents 101, how they work, and the 10 you should be paying most attention to. AI Agents: From automation to super agents, 10 AI Agents you should know -- An Everyday AI Chat with Jordan WilsonNewsletter: Sign up for our free daily newsletterMore on this Episode: Episode PageJoin the discussion on LinkedIn: Thoughts on this? Join the convo on today's LinkedIn stream and connect with other AI leadersUpcoming Episodes: Check out the upcoming Everyday AI Livestream lineupWebsite: YourEverydayAI.comEmail The Show: [e-mail] with Jordan on LinkedInTopics Covered in This Episode:AI Agents Market Growth and AdoptionDifferentiating AI Agents vs. ChatbotsReal vs. Fake AI Agent IdentificationSeven Categories of AI Agents ExplainedEnterprise AI Agent Use Cases & RisksObservability and Traceability in AI AgentsDetailed Review: Top 10 AI Agents 2025Pros and Cons of Major AI Agent PlatformsAI Agent Implementation Best PracticesFive-Day Plan for AI Agent AdoptionTimestamps:00:00 Navigating AI Agents Dilemma06:23 "AI Agents: Opportunities and Risks"07:44 Evolving Multi-Agent AI Systems11:32 AI Agents vs. Agentic Browsers14:32 "Empathetic AI Agents with Goals"17:53 "AI Oversight and Control Challenges"19:48 "Chat JCPT: Versatile AI Agent"23:12 Microsoft 365 Copilot Permission Challenges28:17 Agent Force: Salesforce Automation Tool30:13 Explore and Master AI Agent Tools33:35 "Replit Agent 3: Pros and Cons"38:23 "Genspark's Rapid Update System"39:40 "Manus AI: Autonomous Project Executor"43:36 Identify Automatable Problems First48:34 AI Podcast Recommendations49:21 "Subscribe for Daily AI Updates"Keywords:AI agents, agentic AI, autonomous agents, super agents, AI automation, AI-powered workflows, generative AI, agent washing, agentic browsers, AI chatbot, AI agent vs model, large language models, enterprise AI adoption, agentic model, self-correcting AI, planning AI, AI agent categories, autonomous software developers, general purpose task agents, enterprise workflow automators, specialized research agents, foundational platforms, agSend 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. There's literally hundreds of options out there when it comes to AI agent options.

It's overwhelming. Someone that covers generative AI every single day. I almost like get tired of seeing new AI agents. Part of that is, well, one of the reasons is Gartner came out with some recent research going over the agent washing epidemic, where they found that 95% of companies even promoting AI agents weren't actually AI agents. So what do you do? Do you need to use a bunch of them and find out what's good for you? How do you separate the real from the fake and how do you actually use them?

Well, we're going to be hopefully answering those questions and a lot more today on everyday AI as we go over AI agents from automation to super agents. 10 AI agents you should know. I'm excited for today's conversation. Hope you are too. What's going on, y'all? My name's Jordan Wilson. Welcome to Everyday AI. This is your daily live stream podcast and free daily news that are helping everyday business

leaders like you and me cut through the fluff, understand what's going on, not just learn what's real and what's not, but how we can all actually leverage that to grow our companies in our careers. If that's like, yo, that's what I'm trying to do. You're in the right place. Starts here with the unedited, unscripted live stream podcast. But if you want to take it to the next level, you got to go to our website, your EverydayaI.com.

There, go sign up for the free daily newsletter. We're going to be recapping all the highlights from today's show in case you missed anything and while you're out walking your dog or wishing I would shut up so you could finish your walk on the treadmill, right? But we're also going to keep you up to date with everything else that's happening in the world of AI news. There's a lot over the past 24 hours. So make sure you go check that out in the newsletter.

So AI agents, what's real, what's not? What should you be paying attention to? it seems like almost an endless onslaught. Like agents are being shoved down our throats. But what's the difference between an agent in chat GPT? Should we be using agents? How? And for what purpose is.

That's what we're going to be talking about on today's show. So we're going to be going over and diving into the explosive growth of the agentic AI market. We're going to uncover the hidden challenges and potential pitfalls in agentic AI adoption. Tell you the real reasons you should be paying attention. mentioned AI agents in 2025, not just because they're shiny and new, but the real reasons. And we're going to break down the 10 most popular and widely used AI agents across different disciplines. Y'all wanted this. All right.

So, yeah, if you listen to the podcast, when I say, go sign out for the newsletter, it's for reasons like this. I work for you guys, right? In the newsletter, I said, hey, are we talking about agents too much or not enough? And you guys overwhelmingly said, feed us more agent content. So if you're a listener of the podcast and you haven't subscribed to the newsletter, that's the only way I can ask you guys what you want to know more of. But you said you wanted more agents. And at the end, I've been doing this.

I've been putting together these little kind of bonus guides because I can't fit everything into, you know, 30, 40 minute episode. This one's good. All right. I've, I'm not going to brag. I've pulled out some banger, like extra bonus content this week. So if you want, if you want access to it, uh, There's 20 total agents.

There's 10 that were really good, but I couldn't just make the cut. So go repost this show on LinkedIn. The link is always in the podcast show notes. We put the link in the newsletter. And so make sure you go repost that and I will send it to you. It is done. It is ready to go. It is a comprehensive guide on 20 AI agents.

So let's talk. What the heck is an AI agent? Well, an AI agent can plan, act, and also self-correct. That's the biggest thing to get work done. So an AI chat bot, you're just chatting with it. An AI agent can plan on its own. It can reiterate. And ultimately, it takes actions on your behalf.

The whole human in the loop thing that I hate. I think human in the loop is bad. I think expertise driven loops are what we should be focusing on. Anyways, you've got to have humans overseeing AI agents because they can make decisions and actions on your behalf. which is both really cool and extremely useful for enterprises, but also extremely dangerous. Because if you don't have the right humans driving that loop, right?

The feedback self-correcting loop, yeah, agents can be bad, right? And I've talked about that plenty of the show. But the AI agent or sorry, the AI agent market is projected to explode past $7.5 billion this year. I would expect it goes well beyond that. And this represents, I think, a pivotal shift from AI. right that just answers our questions to get things done right and we've seen whether they're advanced or simple offerings right because you know at this time last year there weren't really no AI agents available from the big players to the general public now there is

Microsoft co-pilot studio chat GPT their agent mode Claude Anthropics Clod has multiple different agents, a lot of them for coding, but they have a computer using agent as well. Google, they have their Project Mariner, Project Astra, hopefully they release their agent space to the masses soon, right? But every single big player has an agent that's available today. Are they all great? No.

Are some of them really, really good? Absolutely. But whether you want to know. know what or not. The future of work in AI native workplaces isn't talking to a chatpot. It isn't just like, all right, well, now we have a chatbot plus our business, you know, info. We have rag. We have, you know, these connectors. No, it's smart humans like you and me, overseeing agents, right? And right now, studies show that over 80% of enterprises are adopting agents right now. So there's massive opportunities.

for all of us tuning in, because if you're tuning in and listening to this, you're ahead of the curve. But it's all about what are you going to do with this information? And you have to be able to understand the risks. So, like I said, what does, how is this getting to the point now? Because it's more than just talk. Like I said, we've been talking about AI agents on this show for multiple years. But the good thing is, I got the receipts two years ago. I'm like, nope, we're not ready.

24, too early. Now we're ready because agents are already showing up where we work. That's the thing. Maybe 18 months ago, two years ago, you had to really go out of your way and try really hard to make anything on the AI agent side work. You had to have a lot of duct tape, hope and dreams and maybe you got a little juice out of it. Now agents are showing up in the actual enterprise systems where we've worked. So that's in our documents, email, CRN, browsers, Ide, et cetera. And the tooling has obviously matured because the models, right, most AI agents are powered by one or a series of underlying models, right?

A lot of them have five, ten plus models running kind of the agented capabilities and a main agent will break off a complex multi-step tasks to multiple sub-agents. And all of those sub-agents are running. You know, maybe one of them is running a coding model from Anthropic. Maybe one is running a reasoning model from Google. maybe another one's running like a, you know, like, you know, GPD 4.1 from open AI or something like that. So these models themselves are much more capable, much more robust than they were six to 12 months ago, especially now that we have these reasoning models and, you know, all of the scaffolding that comes along with these models. And that's another reason why they are not just explosively growing right now,

but they're actually explosively useful if you know what you're doing, right? The tooling has matured and big vendors have shipped platform agents as well. Focused startups have shipped specialized agents. Everyone is shipping them. And they are now more than just co-pilots. They are pilots. They can go out and do work on your own, right? Very simple use case.

I have chat GPT, their agent mode, probably one of the easiest to use, not the most useful unless you are decent enough at prompt engineering. But once you get it, I have my agent. it's on a scheduled run. I don't do anything. It just goes out, execute tasks for me every single day at a certain time. So again, what's the difference? What's the difference between an AI agent versus a model?

All right. And hey, live stream audience, love to see you. Tune it in. Sorry, she gave you a shout out a little more. But let me know if you have any questions. Someone said, I would love to see an episode on Perplexities, Agenic Browser Comment. Already did it, LinkedIn user.

So go go check that out on our website. It's free. Go go check it out. But yeah, live stream audience, if you have any questions, please, please let me know. I'll try to answer them. Just put a question in your, in your comment. So I can make sure I can see it just because we have comments rolling in from all the different platforms. But what's the difference between an AI agent in a large language model?

I've done an entire episode on this, but to put it simply, the lines are blurring because now the base models are agentic in nature. whereas nine months ago, they weren't. So essentially, we had our, let's say, rewind the clock a year ago. The models, the most powerful models in the world were transformer models. They didn't really have a bunch of useful tools, right? All the scaffolding that make models really good. And they couldn't, they weren't reasoning models.

So they couldn't plan and think and retrace their steps and come to a fork on the road and try a couple of things on the left side of the road and figure out, oh, that's wrong. I need to go back to the middle and use some different tools, right? That's an agenic model with tool use. We didn't have that, right? So the lines between what is an AI agent and what is an agentic model, they are kind of blurring, but for the most part, if you think of traditional old school AI models, large language models, they're chatbots, right?

They're obviously becoming much more robust and they're becoming entire operating systems now. But, you know, you are still having as the human to take that. information to do something with it. Whereas now AI agents, you give them a goal. You make sure they have access to the data that they need and then they go figure it out. Right. Whereas I'd say more chat box kind of self-service. Right. AI agents, autonomous cars, they do it for you. An agent holds a goal, chooses its own tools, takes its own action, and checks its own work. And a lot of times, it's not just taking the best available road. It is really in real time building its own role, right?

So let's talk about categories of AI agents. Because there's a lot and this is also where it starts to get a little confusing. Because like someone already said, you also have your agentic browsers now, right? Which I actually think, at least for, hey, what has more utility today? I would say it's almost in some use cases, agentic browser. right? If you look at top to bottom, if you pick a middle of the road, AI agent versus an agenic browser in the short run, I think agentic browsers have much more utility for the average business leader right now. But we're not talking about agendasic browsers. We're not talking about

agentic models. We are actually talking about AI agents, complete systems. And I would break them into these seven categories. So number one would be autonomous software developers, such as Devin. right so these are agents designed to handle end-to-end software engineering tasks from writing and debugging code to testing and in deployment then you have your general purpose task agents which is like chat gpte agent mode so these are platforms that act as a versatile digital assistant that can plan and execute a wide variety of complex multi-step tasks for professionals and consumers you have your inner enterprise workflow automators that would be kind of the third category uh these are like Microsoft Copilot Studio, right?

They can automate company-wide workflows, and they're all connected to your dynamic enterprise data. Then we have specialized research and analysis agents, such as Gen Spark, all right, in a kind of an AI agent startup. So these are agents that focus more on gathering, synthesizing, and analyzing information from various sources to generate comprehensive reports, summaries, or insights. All right. Then you have your foundational platforms in frameworks. This is more like Amazon Web Services, Bedrock agents.

So these are the underlying toolkits, cloud services, and open source libraries that developers use to build, deploy, and manage their own custom AI agents. Then you have kind of your UI and web automation agents. There's a lot of more of these recently than there were, you know, maybe six months ago. So these are agents like, you know, the UI Path, agenic offering. So these are kind of like GUI. They work in a GUI way. So they work on the graphical user interface,

but they're agents that specialize in interacting with software through its graphical user interface enabling automation of applications that lack modern APIs. So technically, right, this is kind of where agentic browsers would fit in, but I do think they're a different category. They're not agents. They're agentic browsers. Then you have your conversational, last but not least,

you have your conversational companion agents. This is more for agents who kind of like personal agents. So it's almost like a blend between a standard AI chatbot, but can also go out and do things for you. Right. So this category is for agents whose primary interface is empathetic dialogue, but they still perform simple, goal-oriented actions like setting reminders or finding information to assist you personally.

An example of that would be inflection AI's pipe. All right. 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, 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. What do all these AI agents have in common? Right. I broke down kind of seven different categories because they're all very different. Right. A lot of people think an agent,

all agents are just general use case agents. And that's not the case. Right. And when you're choosing, hey, what should our company be doing or what should I personally be doing when it comes to AI agents, you have to think of, well, what's your goal? But one, some core functionality that most AI agents share. They break big goals into smaller steps. They call on and use different tools. And then they click through or use some type of user interface and then fix mistakes as they go along, right?

That's the other thing. AI agents are iterative. A lot of them will test and deploy things on their own, go back, which I know it sounds weird, go back and fix the own errors that they created, right? Just like a human would or a team of humans. They pull live info from your data sources, from connectors, from search, from your company knowledge, sometimes from rag databases, right? But they also keep humans in the loop for risky steps and leave a clean paper trail.

Right. So almost every single AI agent we're going to be talking about, there is a level of traceability, right? And that's so important when we talk about AI agents, right? because let's be honest, when we think of, oh, human in the loop, I think we think of, you know, kicking our feet up on the beach with a margarita and our AI agents are doing everything for us. That's not good, right? But you always have to keep observability and traceability in mind.

That's number one, can you observe the agent as it works? But traceability is after the fact, can you go back and see step by step what the agent did, right? So as an example in chat chit-tie agent's mode, you can. You can click the three little dots. You can either watch it in real time, observability or traceability. After the fact, you can always go, click on kind of the reasoning data. It'll extend it. And then you can literally see step by step, a screenshot and a description of what the agent did.

That is a very simple example of traceability. So most AI agents do have those things. But it's not all funny games. The pitfalls are enormous, right? I think especially, and this is maybe a, another conversation for another day. I think AI in general, because so many people have become over-reliant on it,

it's taken our normal human guards way down, which actually makes these hidden challenges and pitfalls way more consequential than they, in theory, should be. Right? Sometimes when you're using an AI agent, what you start to do is, right, kind of like what I'm doing. You're just sipping the coffee and you're like, oh, yeah, good job.

Good job. This would take me five hours. You did in five minutes. I'm going to spend five seconds looking over it. Perfect. You can't do that. You have to be vigilant. But some of the most common pitfalls is something just looks done, but it isn't actually done.

Until you go through and go through the step level logs and, you know, pass failed checks. Permissions. And sometimes in some cases are too wide. AIA agents sometimes have too much power and humans aren't doing enough job to rein that. in or control it on the front end. Sometimes there's no per agent identity. There's no accountability. And but you need to be able to add approvals and change control. Also costs, right? Costs can creep up when agents keep looping forever. Right. So a very simple example of this if you've ever gone into chat GPT and you have the wrong mode selected. Right. And you give it a very simple question.

And you know, now there, you know, as of last night, there's even new, you know, thinking levers inside chat gpt right there's the light standard heavy and happy you know light standard extended and heavy right so in the same way a normal large language model that can reason you might choose the wrong one and it's thinking for 12 minutes on like you know when was the last time the chicago bowl's won an NBA championship right the same thing the same kind of mistakes in terms of these endless loops uh can happen for a i agents the difference is depending on how you set them up you might be paying for usage right so you might accidentally be like, oh my gosh, it's this, this never-ending loop that you're not paying attention to. And then you go look at your API bill and you're like, oh, my gosh, this, it would have

been cheaper to have a human do this, right? You see that all the time. So you need to have clear SOPs, ownership, and fast rollback as well. All right. Enough to chat, top 10 agents that you need to know. Here we go. Open AIs, chat GPT, agent mode. This is more of a generalist virtual computer, right?

Chat ChitpT agent mode has access to a virtual computer, a sandbox. It can run code like it would on a computer, a browser, etc. So what it does, it runs multi-step tasks end to end. It can research, draft, analyze, and build files inside of chat. How it works, it uses a secure cloud computer with a browser, code, and adds files, narrates each step and ask before high impact actions. You can also log in to different websites on a browser. You can open, you know, you can temporarily take control, open up different browsers,

log into different services if you like and then hand it back over. So at any point, you can take over, which a lot of these AI agents, you can do something similar. The pros and the cons of Chad GPD agent, well, it's one of the easiest to use and it's extremely versatile. It's in one place, right? 700 million weekly active users to chat GPT have access. Well, not all of them because you have to be on a paid plan. But, you know, tens of millions of people right now have access to an agent that is the learning curve is essentially nothing. is it the best?

Absolutely not. If I'm being honest, I'd say out of these 10, it's probably only better than maybe two of them. Right? But you have to pay attention to chat GPT's agent mode

because it's going to get way better in the future. Obviously, they have their agents SDK, so companies are building on top of it and you should be using it anyways. I think it's one of the easiest ways, and that's why I put it first on the list. It's one of the easiest ways

to understand how an agent works. You have to go back, look at the, you know, the observability, traceability that I talked about. You're going to learn so much just how an agent calls its tools, how you should be talking to an AI agent versus an AI chatbot. So much can be learned when you go back on the observability, traceability side and, you know, track it. So some of the unique features, it chains multiple tools in one persistent session. So context and outputs stay together. So, yeah, you can keep the context going inside of one agent chat.

just like you would a normal conversation inside Chachieviti. This is best for busy professionals who want one agent to go from question to finish deliverable. It's not great right now. Let me just say that first and foremost. It's not great, but it is the easiest to use. It is the most widely available. And I think out of most of these, it probably has one of the higher ceilings. Next, probably one of the most used here.

would be Microsoft Copilot Studio agents. And these are more, the category would be kind of a governed enterprise orchestration. So what does, and this is different than Microsoft co-pilot. People get confused. They're like, oh, Co-Pilot studio. I use that. And I'm like, okay, show me how.

I'm like, no, that's co-pilot. You're just chatting with copilot. Or you're using co-pilot in one of the 62 places that it exists within Windows 365, copilot. Microsoft Copilot Studio is a dedicated autonomous AI agent builder that you can build something with no code. So it lets non-technical people as well build agents in natural language for Microsoft 365. So it taps into the Microsoft Graph, Intra Identity, and DLP. Admins can add approval in guardrails. If they want, they should. And agents live where office users

already work. That is the big key. Pros and cons. It has enterprise grade governance and deep app hookups and it's tied to Microsoft's ecosystem and it needs setup time. That's the thing. Everything, like one of the biggest downsides to Microsoft 365 co-pilot in general is it's extremely difficult to handle permissions, access, things like that. The learning curve is actually, I think, way steeper than it needs to be. I hope Microsoft improves on that in the future. What's unique about this?

Well, each agent gets an Azure Entra identity with full audit. So IT orgs or IT departments. can treat agents like real users. They can go track anything an AI agent does across their entire organization as if it was an actual employee because different people within your organization can use the same agent, right? Sometimes people are collaborating with the same agent at the same time. You know, one agent might be used by just one person.

It might be used by an entire team. It might be used by thousands of employees across the organization. So that's a good, unique feature for the co-pilot studio agents. and it's best for organizations, obviously, who are already deeply ingrained inside the Microsoft or co-pilot ecosystem. Next, Claudecode. So this is kind of safety first coding execution agent.

So what it does, it plans at its runs and test code changes, turning clear tickets into reviewable patches. So if you're in software engineering, anything on the dev side, you probably use ClaudeCode. It has been one of the more popular agentic coders and one of the first now. I think there's maybe some that might end up being more useful in the long run.

But Claude Code is one of the most popular and most widely used coding agents in the world right now. And it's really, really good. As much, yeah, I know people like realize and I get messages all the time. People are like, oh, Jordan, you always, you know, poo-poo on Claude. Claude's great for coding, software engineering. I say it all the time. That's not the majority of our audience, right? The majority of our audience is non-technical.

So if you're using Claude. dot AI in the front end chatbot, not very useful, not very helpful. If you're software engineer, developer, Claude, absolutely amazing, right? And Claude code also has subagents where it can, you know, break these tasks up. It can work for, I think they said, up to 90 minutes on its own. So how it works, it modifies files and runs tests and loops, logging each step for easy review. Some of the pros and cons, it's safer, test guided changes in explainability.

But on the downside, it needs a lot of testing QA, A lot of times you'll ask for one small change, right? If you're working across an entire repo, you know, an entire app, you ask for one change and it might change absolutely everything. In at least over the last couple of weeks, Claude has had some problems. The service has not been as good. Everyone's kind of jumping ship over to OpenAI's codex, which did not even make the list. But regardless, Claude code is one of the best.

So some of unique things, it's a purpose-built coding loop for small. and save edits and clean diffs, and it's separate from the chat interface. Obviously, Claude code is not going into Claude. A.I and coding in there, it is a separate product, you know, like I said, with the agent, sub-agent architecture. And this is best for teams automating refactors, upgrades, and fixes where automatic build and test pipelines are already in place. All right.

Next AI agent, you need to know, AWS Bedrock agents and Agent Corps. This is more on the governed enterprise integration side. So what is this? Well, it's a cloud runtime. So this is an agent that runs in the cloud for building and scaling custom enterprise AI agents. How it works, agent court adds session isolation, memory identity, identity, and observability. It's kind of like Lego blocks for agents.

So obviously, if you're a large enterprise using AWS, AWS bedrock agents, right, where do you start? You always start where your data lives. So for, I don't know, how big of a market share, AWS has. in the cloud, right? They're one of the three big. You could say Oracle now is in the big four of cloud AI. But you know, you're always like, where do I start? Well, so many, I think so many organizations are on AWS, at least maybe on the smaller side and aren't fully investing in bedrock agents, which you probably should. So pros and cons, it's highly modular, highly modular and framework agnostic. Another good thing about

AWS is they have so many platforms. Yes, they have a strong partnership, obviously, with Anthropic and their investment there. So using the, the claw models, but you can use just about any different models, proprietary, open source, closed source, et cetera. So what's unique? Well, it's plug and play to run many different agent patterns securely on AWS. And this is best for builders who want company-wide standard agents on top of their current existing AWS infrastructure. Next, another company that sometimes I poo-poo on, Salesforce. But if you are a dedicated Salesforce organization. If you're using Salesforce for your CRM, which so many people in the world are, I'll say this. Asian force, in some cases, if you don't care about costs,

it's better than not using it, right? It's better than probably manually having to go through all of your Salesforce information. So what is agent force? Well, it automates sales and service steps inside of the Salesforce CRM that teams already live in. So it uses the CRM context to play plan actions. It has a command center that provides approvals, oversights, and outcome tracking. It can send emails to, you know, prospects that have gone cold based on your data, based on your conversational history. So some of the pros and cons, well, it's trusted, data grounded actions, and it shines most if you're already deep in Salesforce. So like I said, if Salesforce is not, you know, obviously for sales teams that are very dedicated and use Salesforce all day,

Salesforce agent force is a no-brainer. If you're a company that, you know, let's say you use Microsoft 365 co-pilot, you're just in Salesforce every so often. Agent Force might not be the one that gives you, might not be the agent that gives you the best ROI. So who is this best for? Well, revenue and support teams that want measurable impact inside their system of record. Next, Google Project Mariner. This on the list is probably the least used AI agent. But Google is with OpenAI.

They're running ahead of everyone else. So even though their Project Mariner, not that great, right? It's a Chrome extension. You have to be on a higher tiered paid plane to use it. You still have to pay attention because eventually Google will release agent space to the matches. This is their kind of premier AI agent platform. However, it's been hardly released to anyone, right? So I was at Google Cloud Next when it first started rolling out back in May.

You know, here we are five or six months later and still very few organizations have access to agent space. I know we have a lot of people, you know, listening at Google. So Google, please start rolling this out to SMBs and, you know, enterprise companies in general. I think it's a fantastic product, but hardly no one has used it. So I can't really talk about agent space because I haven't even personally used it. But I personally use Project Mariners. I can talk about that. So this is more of a lives in the browser, right?

But you like if you have access to Project Mariner or if you can get access and you are a leader on your team when it comes to AI, this is one of those things. You got to get the reps in, right? Because Google and Open AI are going to continue to run away with everything. So you need to get used to their agent tools with where it's available now, even though most people, right, Google's Project Mariner might get rolled up with Project Astra and agent space. Who knows if it's going to be its own dedicated offering in the future, but you should get use of the technology and how it works. Like I said, one of the easiest ways to learn AI agents right now is to go use chat GPT agent mode and go observe its steps. You're going to get better at prompting AI agents.

It's completely different than prompting a large language model. The same thing can be said for using Project Mariner. So here's what it does. It clicks through real websites for you at scale without manual scripts, right? So it's essentially a browser using agent. It's not a computer using agent where chat GPT's agent mode has use of it, essentially a virtual computer, including like things that you can do outside of the web. So Google's Project Mariner is more of just web-based. So it runs isolated cloud browser sessions, plan steps,

fills forms, and can work in parallel for speed. So pros and cons, it has great, obviously, real web execution, but the packaging and scope are still evolving. So what's unique here? Well, it can run up to 10 concurrent browser sessions to finish web tasks faster, which you don't have that in Like chat GPT agent mode, you're kind of restricted to one run at a time. So you have this kind of parallel advantage here with Google's Project Mariner. And this is best for teams that need repeatable web workflows done in handoffs reliably. One thing I like about Project Mariner is it has this teach and repeat thing. So if you have a process that you manually do over and over, I should have used this for a project

that I was working on yesterday. I was trying to do it inside both perplexities comment and check. IGPT agent mode, they weren't doing the best. I probably should have done it in Project Mariner because it has this teach and repeat pattern, right? If you have, you know, five different websites that you all use concurrently, you're grabbing something or some information from one, carrying it over to two, making changes, bringing it over to three.

That's a great use case for Google's Project Mariner. But come on, Google, give us agent space. Next, and this is a more recent one. And one, at least on the list that I've probably been the most impressed by recently. And that's Replit Agent. And this is an autonomous software creating agent. So here's what it does. It turns plain English into working apps, talking full scaffolding, coding, running tests, fixing everything.

Front end, back end. It is a absolute workhorse. So you'll notice that some platforms didn't make the list. I didn't put cursor on here. I didn't put lovable. I didn't put both all great, all great tools. But I think Replit agent. three outshines them.

All right. So it also uses a reflection loop to execute tasks, tests in a browser, and auto repair issues inside of Replitt's IDE. So some of the pros and cons, super fast idea to app, but complex code still benefits from human review and controls. This is one of those.

Can a non-technical person use Replit Agent 3? Sure. But your outcomes are going to be much better if you know what the heck is going on, right? It's kind of like watching. a movie in a different language you don't speak without subtitles, right? You're like, okay, I think I know what's going on because I can see everything. And I think I'm following the storyline of turning plain English into a working app with a database and a backend authentication

and payment processing, right? What Reflet Asian 3 can do is extremely impressive. But it probably makes more sense if you speak the language, right? So it's almost like some of its most, I would say appealing use cases are for non-technical people because technical people have plenty of AI IDs they can go work with. Like I said, cursor, lovable, bolt, you know, clawed code. I think Replit is one of those that's probably going to separate itself over the next year and become wildly even more popular than it already is. So what's unique about this?

Y'all, three-hour runtime. An AI agent can go run for three. three hours. And I've seen some videos on this, right? Unedited before and after of what it can do in two to three hours. Wildly impress it. So yes, some of the other platforms, the lovable, the bolts, right? It can accomplish the same things, but with a lot of back and forth iterative prompting. Right. So if you get it right on the front end and if you know what you're talking about. Again, think of the movie analogy. If you can direct the movie, Replit Agent 3 can go do it. Three hours. So this is best for founders, educators, and developers who want prototytes. types live and fast. Next, Zapier agents.

I've been very impressed. Zapier agents are really good. So what it does, this is no code agents that automate work across the more than 7,000 enterprise apps that Zapier connects with. So I'm not even going to get into like the difference between, well, I'm, I will shortly, actually, right? I think a lot of people when they think AI agents, they're thinking AI powered workflows. And yes, Zapier has that as well.

but their agent's offering is completely different. So yes, you can still have your, you know, kind of your visual builder, building different, you know, marketing automation workflows and enter in AI. Because a lot of people think of, you know, like, oh, N8N, that's an agent. I'm like, no, that's an AI powered workflow. So Zapier does have AI powered workflows, but they have their actual agent as well. So completely different product. And it's extremely impressive. So all you do is you describe the outcome you want.

The agent selects the app actions. it runs them and that it can hand certain things off to other agents. So some of the pros and cons, well, there's massive app coverage more than 7,000. So you don't even have to worry. There's support for MCP, but you don't even have to worry about creating your own custom servers because Zapier connects to the internet. It is the glue that holds the internet together. So that is a huge advantage of Zapier agents.

So some of the downsides, the cost and complexity can grow, right? Especially with more open-ended flows. What's unique? Well, there's agent-to-agent calls. live knowledge sources with drive box drop box dashboards and ready to use templates and this is best for ops teams marketing sales and support teams wanting practical cross-app automation without code right so i think zapier agents um have a place it's not like you choose one AI agent i think zapier is ultimately uh going to continue to rise uh in relevance outside of those in the marketing

world, right? I've been using Zapier for more than 10 years. I love it. It's traditionally been for marketers. I think non-marketers are eventually going to catch on to Zapier because of Zapier agents. Miguel said, would you say Zapier is better than N8N? Yes, absolutely. NAN is free. It's open source. Zapier's better, right? Don't be wrong. N8N's great. A lot of people, right, stop seeing it, right, a lot of people see like what's on social media and that, you know, it's all these N8N workflows, you know, comment, follow me, repost this and do these 30 things, and I'll send you this agent that prints money. It prints a million dollars, right? No, it's just an AI-powered workflow.

You know, part of the reason I think N gets a bad name because people are abusing it and selling it as snake oil. So, but no, Zapier's better. All right. Next, we have Jen Spark, super agent. And this is an orchestrated research and creation. So here's what it does. It produces research pages, summary, summary, slides, and media fast. The research and creation process in GenSpark, extremely impressive. And both GenSpark and Manus, to their credit, keep churning out updates a lot faster than as an example, co-pilot studio or chat GPT agent mode. They're really shipping fast. So here's how Jen Spark's super agent works. It routes subtasks across multiple models and tools. I think the last time I

checked, it uses nine different large language models and then it stitches together at clean results. So essentially, you know, you have a routing agent. And then there's specialized sub-agents and they all kind of work together. They all go off and do their own thing, report back to the main agent and then give it back to the user. So the pros and cons, well, is broad and quick. You can verify key outputs before publishing in a professional setting. So what's unique about GenSpark? Well, it has these thing called Spark pages.

Think of them is essentially like almost a cross between like a website and a PowerPoint, right? But they are shareable pages. that you can go in and update, and this is where all of this information can live. So instead of just having a bunch of text or some random, oh, it spits out a PDF, it's like a way for your content that you create with the Jen Spark Super Agent to live.

And this is best for analysts and marketers who need synthesized deliverables without heavy lit. All right. Manus AI, and this is more of a hands-off autonomous executor. So what does Manus do? So GenSpark and Manus are both out of China, which is why maybe you've heard of them and maybe you haven't, right? But they are obviously, I think, starting to creep their way into the U.S. ecosystem as well. So what does Manus AI do?

Well, it handles long, multi-step projects and end, then shows you every step it took. So how it works is it works in a cloud desktop and browser environment similar to chat GPT's agent mode, but it plans action and it logs everything and keeps going even if you disconnect. So I will say it is much more robust right now than chat ChbT's agent mode. So, but they work very much in a same way in terms of the user interface and user experience. So some of the pros and cons, well, there's deep autonomy with transparent trace, but also there's mind privacy and access policies. So what's unique? Well, persistent cloud sessions work continue even when you go do something else,

right? So that's nice. Chad GPT's agent mode is hit or miss with that. So essentially you can log in or keep these instances kind of live and running across multiple days, multiple weeks inside of Manus AI. So this is best for, I'd say smaller teams, creative teams or consultants and researchers who want an agent that just kind of does the work. Right. So I'd say Manus is a great kind of starter agent for a lot of people.

All right. That's our 10. We're going to keep this thing going. A couple of things to wrap up here. So what's the difference? Well, I gave you the different categories, but let me start breaking these now because each platform offers different runtime primitives, including identity, memory, isolation, observability, and parallelism capabilities. I told you. Some of these agents don't run in parallel.

Some of them do. Some of them have persistent cloud instances. Some of them don't. So, you know, every single thing, like a car, right? Some have Ford War, some have two. Some are hybrid. Some are electric. Some are four-seaters.

Some are vans, right? So it depends on what you need and what you want. But the platforms vary in domain grounding depth. You know, some are CRM native, some are IDE native, some are browser native, et cetera. And the delivery approaches range from no code studios like Microsoft copilot studio to framework run times to embedded application agents, right? Some are great at certain tasks right now. And I do think, you know, as we go on, there's like marketing specific agents, there's sales specific agents, there's HR specific agents, there's legal AI agents, right? Harvey, right? There's there's so many different agents for different categories as well, but then

there's also different genres, even within those categories, which is why it is sometimes hard to keep up. But cost structures also significantly differ. Some of them are subscription models and have pretty generous usage. Some of them, you're just paying for the actual API costs in addition to your usage. So you also have to keep cost in mind. Also, the integration philosophies span from ecosystem lock-in, more like AWS bedrocks, to universal connectivity, right? Like Zapier. So it's also like, what systems do you ultimately need these agents to pull data from and talk to?

That's the other thing. You have to think of agents, you know, as like a mind map, right? It has to be able to pull in dynamic data from multiple places, and it has to be able to execute tasks into multiple ways. All right. So what's your agent use case? Here's what I don't want you to do. I don't want you to look back at these 10 AI agents that I told you, say,

oh, this one's my favorite, this one sounds good, and go get started. Think of it the exact opposite. You need to think about your automatable problems. Is that a word automatable? That's what you need to do. Think about your automatable problems. Because again, if you have a hammer, everything looks like a nail and you're going to get screwed.

Think. Like so much of getting the most out of large language models, out of AI powered workflows, out of AI agents, is really doing an honest audit of your day-to-day responsibilities or your team's day-to-day responsibilities and how you actually fulfill those. I know it sounds boring. I tell people, use time trackers. What are you doing? What is your team doing?

What are you actually doing? What is the work you're actually performing? hands-on keyboard, hands-on mouse, inside of large language models today. And then you need to find the automatable problem first, not the shiny tool. Here's what I would do. Start small, pick one workflow people hate, and then define what done looks like. A lot of people don't define what counts as agetic success first. And they're just like, oh, this is good enough.

You need to lock it down. Right. So you need the least privilege access, a person. approvals, logs, and fast rollback. You need to make sure that you're using it in a safe way. If you're giving an AI agent access to your company's dynamic data, you need to make sure it's auditable, traceable, guard rails, etc. You also need to be able to measure weekly, not just on a per task completion,

but you need to be able to measure completion rates across the board, fixes needed, time and cost safe. You need to understand how much does it take humans to do this? Guess what? Humans make errors. Humans go through revisions. I think people think of large language models and AI agents as something they're not. Nothing will ever be perfect.

Just like there's never been a perfect, you know, human worker. No worker has gone through their entire career, never made a mistake in that spreadsheet, never made a mistake in that presentation. You know, they've no, no one's ever killed absolutely, you know, 10,000 out of 10,000 sales meetings. Human workers make mistakes. So do AI agents, right? So you have to be able to calculate. the cost from start to finish.

And then once you have and you see the positive ROI, right? Maybe you think, oh, Microsoft 365 Copilot Studio, this is where we're going to go. Maybe you're going to get great ROI out of it. Maybe that's for six months from now. Maybe you're going to get something better with using, I don't know, chatypte agent mode or Manus or, you know, Zapier agents. So you need to pick one that solves your automatable problem, be able to measure it, and then what can scale with you.

Five days. This is what I want you to do. Ready? Write this down. Get out your pencils. Take a screenshot. Whatever. Day one, you have to define the win, right?

The automatable problem. And then pick a lane. One lane outcome. So whether that's Agent Force, co-pilot studio, et cetera. Don't test one thing across five different platforms. It's not going to work. Okay. Day two, you need a short list and smoke test.

So try two platforms. All right. So not five, two. All right. Eventually, get in one lane. We need to test. Try two platforms on one real world example. You need to note the setup, the accuracy, the workflow, go back, look at the observability,

traceability. Day three, build the MVP. So connect only the needed data and apps. Okay? And then confirm usable by owner. Make sure has the right access to the data. Make sure the inputs are correct. Make sure it can output or create whatever deliverable you actually need.

All right. Day four, run five cases. Time the runs. Note the fixes and the limits and then compare it to the baseline of what your humans do. That's how you calculate the ROI. Then day five, you decide and expand. You confirm the platform. You write a one page, how it works.

And then you start sharing with the rest of the team, the rest of the organization. and then you can add adjacent use cases across different teams or across your organization based on what already works. Very small scopes. There it is, your five-day plan. We covered a lot. I went over 10 AI agents. I talked about some of the problems and opportunities as this space continues to blow up.

All right. But I will tell you this. Even though this is a longer episode and I apologize for that, but hopefully you all found some value. There's more. All right. So if you found value in this episode as, we wrap up here, please share this with your community. Go click the repost button on the LinkedIn post.

So if you're listening on the podcast, always put the link to the LinkedIn post. Go repost this. And I will share. I put together a crazy useful guide on 20 different. There are some other AI agents that I really like that didn't make the top 10, right? So make sure you go repost this. I will share that with you. Also, we've had a lot of great agent content and some good interview. use over the last year or so. So go check out episode 590, which is agents, LOMs, or algorithms, a playbook for choosing AI. Go listen to episode 452, which is AI agents, the future of enterprise work. And then go listen to episode 422 licensing AI agents.

I like this episode. What is it and do we need it? All right. So learn from other people, not just me, some great experts that I've had on the show. All right. I hope this was helpful. helpful. If so, please go to your everyday AI.com. Sign up for the free daily newsletter. Thank you for tuning in. 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.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.