Everything You Know About AI Agents Is Wrong

Publicado em

Vídeo em Inglês, publicado pelo canal Duncan Rogoff | Learn Claude Code.

About this content: The summary, key points and topics below are generated automatically from the video transcript. They reflect the content, opinions, claims and predictions presented in the video and do not represent a position, validation or fact-check by IA Dots.

Video summary

The video discusses the current state and limitations of AI agents like OpenAI's Dots, Meta's Muse, and Grockbots. The speaker argues that many features marketed as advanced AI agents are similar to existing automations or scheduled prompts, and highlights concerns regarding costs, memory management, and time spent setting them up. However, the speaker also acknowledges potential use cases for small businesses, continuous app testing, and future scenarios where agents communicate directly with one another.

Key points

  • OpenAI invited local creators to test Dots, their new AI agent, prompting an analysis of the current market landscape for AI agents.
  • Many functions attributed to AI agents, such as morning briefings or lead forms, closely resemble traditional automations or scheduled jobs built with no-code tools.
  • Key differentiators of AI agents include the ability to pick steps toward a goal dynamically, adapt preferences over time, and operate independently on cloud servers.
  • Major concerns with current agents include high pricing, lack of clarity regarding long-term memory management, and the risk of wasting time on setup instead of high-leverage activities.
  • Potential high-impact use cases exist for solopreneurs handling repetitive tasks, continuous app testing, and automating communications like handling specific social media updates or email replies.
  • Future developments may involve agents learning personal preferences over several years and communicating directly with other agents to handle transactions and logistics.

Topics

AI agentsOpenAI DotsMeta MuseAutomationSmall businessSub-agentsArtificial intelligence

What you will learn

  • AI agents differ from basic automations by dynamically choosing steps to achieve a given goal rather than strictly following a rigid, pre-drawn sequence.
  • Current AI agents often come with significant costs, such as $100 per month for OpenAI Dots, which can make them prohibitive for many users.
  • Setting up and training AI agents can sometimes consume excessive time, acting as a distraction from foundational business activities that truly move the needle.
  • Advanced use cases for agents include running and testing applications around the clock, handling bug fixes, and managing customer feedback on autopilot.

Questions and answers

What is the main difference between an automation and an AI agent according to the video?

In an automation, every step is drawn ahead of time in a fixed order. With an agent, you give it a goal and tools, and the AI picks which steps to take to reach that goal.

How much does OpenAI Dots cost to use based on the video?

OpenAI Dots costs $100 a month to use.

What is one of the coolest app-related use cases for agentic AI mentioned in the video?

An AI agent can run and test an application around the clock, look for bugs, handle customer feedback, and write fixes into the app without human intervention.

Video transcript

Transcript from available captions; it may contain errors.

AI agents actually suck. This week, OpenAI invited 25 local creators in to see Dots, which is their new AI agent, and I was fortunate to be one of them. Okay, let me get this out of the way real quick. Obviously, the title is a little clickbaity, but let me tell you why I picked it. And to be clear, I genuinely really appreciate OpenAI having me in, and I learned a lot, and I had a great time talking with the team and meeting other creators. But the longer I listen, the more I realized that there were several problems. A lot of the things we're talking about really just kind of sound like features we already have with a new name. So here are the five problems I have with AI agents. And at the end I'll tell you who I think agents are worth it for and where I see them going. So if you've heard about Meta's Muse, OpenAI's Dots, Anthropics Claw Duet, or XAI's Grockbot and wondered whether you actually need one, this video is for you. So let's just start with a simple example. Say that you run a small business and a sales call gets cancelled. You tell an agent, "When a call gets canled, check my calendar, take the next lead on my list, and book them into that spot." A chatbot will tell you how to do this, but an agent will do it for you. So, right now in the market, there are basically two kinds of agents, Muse and Dots, which are basically one agent for everything. And agents like Grockbot, which is a boss agent. So one lead agent hands jobs to helper agent and each helper agent gets a single job. So like one for leads and one for invoices. So think of one assistant who does everything versus a manager with a team. So there is one thing that kept coming up in these conversations and when companies talk about their agents, it's this word proactive. this idea that these agents are just out there doing things for you, which sounds really great, but it is a little bit problematic to me or for anyone who's already been building with AI. The example that people give is like, hey, I want a morning brief every morning when I open my computer and the agent tells me what I need to work on that day, which sounds like really great, but to me, this is just a scheduled job. Like, we can already do this. It's not actively thinking or doing things for you. It's literally just saying, "Hey, every morning at 9:00 a.m., like look at my email and look at my calendar and tell me what's coming up." This is things that we can already build with Claude or with Codeex, right? This isn't necessarily a new feature. It's just a new wrapper for it. Or say in the example of these leads, like say a lead fills out your contact form and it answers a couple of qualifying questions and whether they're qualified or not, they get a slot on your calendar. Now, an agent can handle all of these things, but really this is sounding an awful lot like automation, which we could do way before with like no code tools like Zapier or NAND or make.com. So, it's really this distinction between something that is truly a gentic and like thinking for itself versus something that's just like an automated set of processes, which again we sort of already have. There is a difference in an agent and an automation that I think is worth talking about. So, in an automation, you actually draw every step ahead of time. If this happens, then that happens. There's like one order for everything. With an agent, instead you give it a goal and the tools like your calendar and your email and your invoices, and your AI picks which steps to take. So, to be fair, agents can be really, really useful like for all of the edge cases, right? Like say like a lead responds to your email and says like, "Hey, can we move this time or can we push this to next week, right?" And so now you don't have to build in that set of rules to your automation pipeline. Your agent understand what's going on. and it understands who's making the request to you and it can just move their call on the calendar and send them a follow-up email. So, there are some real advantages here. There is one more big distinction that agents have a huge lead on. So, basically an automation stays fixed until you edit it manually. Basically, the idea behind these agents, whether it's Dots or Muse or any of them, is that they learn over time your preferences and how you like to work. and they should in theory get better the more you interact with them without you having to do things manually. The other part that is nice about agents is that they generally have their own computer in the cloud so they can work all the time whether your laptop is closed or not. And so people have been finding ways around this for some time. Like a lot of people with open call were like rushing out to buy a separate Mac mini that is just always on or you can spin up like a VPS or virtual private server in the cloud to run your automations on all the time, right? So there have been ways around this, but it is nice that you're not paying for server space with these AI agents. They take care of it for you. One other big question that came up when it comes to like DOT or Muse, like the single agent idea is this idea of memory. How does it keep track of everything it knows about you? Everything that's going on in your business, all of your contacts, all of your processes. And to be honest, like they didn't really answer this and I don't know that they have a full solve for it yet. So that's definitely something to watch out for to see if your agents start to drift or hallucinate or make things up or just kind of actually get worse over time the more memory that they build up or if they actually genuinely get better. So problem number three is this idea of helpers or sub agents. And I have this problem a lot with like things like Grockbots, right? When you have like your core main operator agent and then you have like sub agents for a specific task. Like say one manages my email inbox and one like researches new trending topics for me and one writes my posts or whatever, right? And to me what's weird about it is this kind of just sounds like clawed projects or custom GPTs, right? Cuz basically all that's happening under the hood is clever prompt engineering, right? It just has a very specific optimized prompt for any given task. And then it has access to again a specific focused knowledge base of information that it needs to know to do that one single job. And again, we really already sort of did this. What's nice about the agents is that you don't have to like go into a custom GPT and manually type something in, but still you can like talk to the agent directly or you talk to your operator Grockbot agent who then decides which sub agent is right for the job. But still, you still have to proactively communicate with the agents for them to do the specific work. So, is this really better uh than a quad project or a custom GPT? I'm not entirely sure yet. The other thing is generally speaking across the board, I haven't seen a lot of really compelling use cases. A lot of the examples people use in the room were basically based off of personal life. One example that was given is that this person really liked to have flowers in their home and they had a flower subscription that was coming every week, which sounds really nice, right? But their agent noticed that actually in 2 weeks time they were going out of town. And so the agent was like, "Hey, do you want me to cancel your flower subscription?" Okay, that's like all good and well and nice, but like is it worth me really setting up an AI agent to monitor that sort of thing? Like again, I'm not entirely sure. The other thing about agents is that they are expensive. OpenAI dots costs $100 a month to use. There was a person in the room that actually says they have two dots. One for their personal life and one for their business. And for most people, this is too expensive. So, is this price going to come down? Are more people really going to be able to access these tools? And for me, the one use case that I really wanted to see if they could do like the platforms don't allow. So, I really wanted to see if I could have an agent grow my ex account for me. So, that means posting for me, writing articles, commenting, replying. I know I'm contributing to the like sloppification of the internet. Uh my bad, but it is what it is. And so um you know, personal feelings aside, like that's a really compelling use case for me as somebody who primarily creates content on YouTube, but I do understand that having a presence on these other platforms is really impactful. So can I have an agent manage that for me? The answer still is kind of no. And that's at the platform level. Like, you know, LinkedIn and X don't generally want just kind of like AI robots communicating with each other, which is understandable. But then for my specific use case that I was interested in, I'm not really able to do it. But problem number five is really the biggest problem and kind of why I say AI agents suck. They inherently aren't bad. They can do a lot of incredible things, but I actually think for most people, they're not really the most impactful thing. And I've seen people in my community who are like, "Hey, I set up Muse and I set up Dots and I set up Grockbot this week and I spent a whole bunch of time teaching them and setting them up and training them and all of these things." And I think generally speaking for most people, they're just kind of a waste of time and it is a way for you to feel productive without doing the foundational things that truly move the needle for your personal life or your business. I do think there are some use cases like if you are a soloreneur and a small business owner and you are prospecting leads, you are booking sales calls, you are managing invoices and payments and taxes and all of those things. I really do think a proactive AI agent can be super beneficial for you there. But I actually think for most people, they're not really that helpful yet. And so I think a lot of people actually spend a lot more time trying to set them up because they think they're cool and they'll spend less time focusing on the high leverage activities that actually move their business forward. So therefore, agents actually just become a waste of time and a way for you to feel productive and like you're doing more things. And while you may be doing more things, I don't know that those things will actually be that impactful because I generally hear two complaints a lot. People say, "I don't have enough time or I'm not seeing results fast enough." And wouldn't it be great if these AI agents can take all this work off my plate? That is like the promise and the dream, right? But now with AI, what people aren't realizing is that AI can do a lot and it takes a lot off our plate and it should be freeing up our time, but we're actually just asking the AI to do more things. So we are all doing more. So we don't actually have more time. AI is more efficient and it's making a lot of our work faster, but we're also asking it to do more tasks. And so really it all kind of evens out in the end. So for me in my business, the biggest change and the biggest way to scalability actually came from simplifying, cutting out the noise, picking less things to work on. That's why I mostly primarily focus on creating content on YouTube and running the community. And I know if I just do these two things, I will continue to grow. So I encourage everyone to really look at all of the tasks they are working on and decide first what is actually important and moving the needle for you. And if there is anything that feels like it's really high leverage that's taking a lot of your time, then you can hand that off to an AI agent. But if there is something that really isn't that impactful and is taking a lot of your time, you shouldn't be doing it and an agent shouldn't be doing it either. That being said, if you are running a small business, an agent can replace some of what you'd otherwise hire a virtual assistant to do. One of the strongest examples I've seen comes from Grockbot. Basically, you have an AI agent that watches every AI company on X every 15 minutes, and it alerts you when news drops. That way, you can stay ahead of all the trends and create content when it's the most relevant. You can have another agent that will check your socials and when you post and it will turn posts that you've already created into drafts for other platforms. So that's really impactful for anybody who's creating content or trying to run a business kind of growing on social documenting their journey. You can have another agent that drafts replies to every email so you are not stuck monitoring and living in your inbox which is absolutely crazy. One of the coolest use cases I saw was an AI agent that actually runs and tests your app around the clock. So, if you're building product or you're building a platform, right, it's an agent that will look for bugs. It'll handle customer feedback. It'll actually write the fixes into your app for you without you having to touch a thing. It can continue to improve your website, improve SEO and AIO, answer the customer feedback, all on autopilot. So that is really an interesting use case for agentic AI and I'm curious to see if more people start bringing something online like this. So here's where I think AI agents are going and where I think they become actually really compelling. So for me, I've been building with AI for a couple years. I know how to build my own skills. I know how to write good prompts. I know how to ask for what I need. I know how to schedule jobs. All of these things. But most people don't. Most people are new to AI. An AI agent handles all of that setup for them inside like a clean, simple little chat window, a little app, right? So, hundreds of millions of people are coming online to AI in the next few years. I think agents are a great way for them to get things done without having to go down the technical rabbit hole. And I do think this idea of agents getting better the more you use them and they learn over time again is really interesting. The example I've been giving is this idea of using an AI agent to plan a trip or a vacation, right? Maybe the first time you use it, it's not all that different than going directly to cloud or going directly to Chat GBT, right? But say you plan 2, three, four, five trips with it over the next couple of years. Over time, it will start to learn like which airlines you prefer, which hotels you like to stay at, what type of places you like to eat, which experiences you like to have, right? And so now it can plan a more personalized trip experience for you the first time with very little direction from you. So that's where I think they start to get really interesting. And I think the biggest idea is this idea of AI agents communicating with each other. So if each person or each business has their own AI agent, then they start to talk to each other. So that flower subscription example, right? Say I tell my agent, hey, I want to order some flowers. I want them to be delivered every week. It goes, it finds the company with the subscription, it pays for me, it talks to their AI agent. So now on the business side, it receives that inquiry, it processes it, it puts it into the system, it manages the inventory, makes sure the shipment goes out, it collects the payment, and now these agents are just communicating with each other. And so I think that gets really, really interesting. Or say you go out to dinner with a couple of friends and you say, "Hey, I'm out to dinner uh you know, with Tim, Eric, and Bob and we want to split the bill. You guys figure it out and you just hand off the work to the agents and they basically have access to your payments and it all gets sorted, right?" So, I think that is where these start to get really interesting over time. So, I do think we're still early. Most of these agents have been released in the last few months. I do think in the next like two to five years, most people will have some sort of AI agent for them that really knows them well and takes care of their life and their business. And that's really interesting. I'm very curious to see where this goes. So, if you are someone who wants to set up your own AI agent, I don't necessarily know if Muse or Grockbots or Dots is like the best one for you. I encourage everyone to just like explore and try things and see what feels right. But I do think the most important thing is to get a little bit of clarity on what you want them to do. Because telling an agent like, "Hey, I want you to run my business," like is just not going to do it. So like, write down everything that's on your plate this week. Circle the two to three tasks that actually grow your business and cross out any of the ones that don't. And then think about which one of those jobs an AI agent can handle and just start. Talk to it. Say, "Hey, here's the task that I have on my plate this week. I want you to handle this one. This is my desired output. Let me know what you think about this. But I would just be careful falling in the trap of spending too much time playing around with AI agents and not enough time doing the actual things that move the needle. If you want to go deeper on things like this, I run the Claude Code Club where you can master Cloud Code in 45 days and get access to all my resources. I'll leave the link in the description. If you want to see how I turned Claude into my video editor to create videos like the one you just watched, check out this video right here. I'll see you over there.

Assistir ao vídeo original no YouTube

Assuntos: Agentes de IA, Dots, Muse. Publicado no YouTube em 2026-11-12.

Seleção automática por assunto e idioma. O vídeo pertence ao canal de origem.

Comentários

Postagens mais visitadas deste blog

How to build a JARVIS AI Voice Agent for FREE | Full Livekit Tutorial (2026)

How Does AI Actually Work? 🤖 | AI, Machine Learning & Generative AI Explained