← все видео

The most powerful AI Agent I’ve ever used in my life

Dan Martell · 2026-02-26 · 11м 55с · 800 487 просмотров · YouTube ↗

Топики: launch-ai-content

Аудио ещё не скачано.

📝 Summary

model=openai/gpt-oss-120b · prompt=summary-v7 · 4 739→1 809 tokens · 2026-05-28 09:02:08

🎯 Главная суть

AI‑агенты уже способны полностью выполнять проекты без постоянного контроля человека: они планируют, пишут код, собирают данные, создают файлы и даже взаимодействуют с другими сервисами. Перейти от простого чат‑бота к такому агентному уровню — ключ к масштабному росту продуктивности в ближайшее десятилетие.

Три уровня искусственного интеллекта

  1. Чат — ChatGPT, Claude, Gemini: пользователь задаёт вопрос, получает ответ.
  2. Автоматизация — интеграции типа Make.com, Zapier, n8n, где AI используется в готовых воркфлоу.
  3. Агент — AI самостоятельно мыслит, планирует, открывает браузер, пишет код, создаёт файлы и проводит исследования без постоянного вмешательства человека. Большинство находятся на первом уровне, небольшая часть уже экспериментирует со вторым, а лидеры переходят сразу к третьему, минуя промежуточный шаг.

Новая роль человека: от исполнителя к режиссёру

Самая распространённая ошибка — использовать AI лишь как «умный поиск» или «партнёра по мыслям». На самом деле человек должен задавать цель, а AI выполнять её. Это «обратное промптинг»: сначала формулируется желаемый результат, затем AI строит план и реализует его. Пример IBM: внедрение AI‑агентов для 270 000 сотрудников дало $4,5 млрд экономии и ускорило задачи менеджеров (например, определение кандидатов на повышение) на 75 %. Успех пришёл не из‑за «умнее» AI, а из‑за того, что люди перестали выполнять рутинные действия и стали только направлять систему.

Метод «обратного промптинга»: от результата к плану

  1. Определить конечный результат — чётко сформулировать, какой проблемой нужно решить.
  2. Задать форматы и требования — предоставить примеры вывода, шаблоны, критерии качества. Чем конкретнее, тем точнее будет работа агента.
  3. Обратная связь как у стажёра — регулярно давать фидбек, сохранять его, чтобы агент «учился» и мог выполнять задачу автономно в дальнейшем. При таком подходе AI перестаёт выдавать «общие ответы» и начинает генерировать точные, готовые к использованию результаты.

Три правила эффективного управления AI‑агентом

Как подобрать подходящий инструмент

Профиль пользователя Рекомендованный агент Ключевые возможности
Владелец бизнеса, нужен исследователь, контент и общие задачи Manus AI Самостоятельно исследует рынок, пишет код, генерирует готовый сайт.
Креативный специалист (копирайтинг, дизайн) Claude Cowork Управляет файлами, чистит папки, открывает вкладки, завершает проекты без вашего участия.
Разработчик Cloud Code Находит и исправляет баги, пишет тесты, работает параллельно с кодовой базой.
Техничный пользователь, желающий полностью автоматизированного личного ассистента OpenClaw Имитирует человеческое поведение, имеет память, но требует сложной настройки (есть риск неконтролируемых действий, как покупка $3 000 курса без разрешения).

Важно сосредоточиться на одном инструменте, стать в нём экспертом и только потом расширять арсенал.

Практический пример с Manus AI

  1. Задача: владелец цифрового маркетингового агентства хочет проанализировать конкурентов в Канаде, собрать цены, ключевые фичи и сильные стороны.
  2. Запрос в Manus: «Research the top three niches, digital agency in Canada, finding their pricing, main features, and what they're doing well. Then create a simple one‑page website that summarizes the findings».
  3. Результат: агент автоматически нашёл три компании, собрал данные, построил список задач, сгенерировал код сайта и создал готовую страницу — всё за < 10 минут.
  4. Итерация: пользователь попросил добавить раздел с отзывами клиентов; Manus собрал отзывы, обновил сайт и сразу же разместил его в Slack, где его увидел креативный директор (Sam Gaudet).
  5. Обратная связь: после получения комментариев в Slack пользователь сказал «apply any feedback automatically», и агент сам внес правки в сайт.
  6. Рассылка: Manus отправил готовый сайт по email Джоэлу Харрисону и опубликовал в общем канале.
  7. Параллельные задачи: такие воркфлоу можно запускать одновременно, пока пользователь занимается другими делами, а агент продолжает работать в фоне.
  8. Мобильность: всё то же самое можно инициировать с телефона, просто проговорив запрос, и агент выполнит его, подключившись к нужным сервисам.

Эта демонстрация показывает, как без навыков программирования и без создания сложных автоматизаций можно получить полностью готовый продукт, управляя лишь конечным результатом и небольшими корректировками.

📜 Transcript

en · 2 524 слов · 27 сегментов · clean

Показать текст транскрипта
AI just changed forever. And if you're still using ChatGPT like a fancy Google search, you're about to get left behind. Most people are still chatting with AI, asking questions, getting answers, maybe writing some emails. But the people who are going to win the next decade, they're not chatting with AI anymore. They're putting AI to work. AI agents that actually do things, build things, complete entire projects while the human sleeps. I build AI companies and tools every day. And let me tell you, this might be the most important technology shift we've ever seen. Bigger than ChatGPT when it came out. Bigger than the internet. So today I'm going to show you what this new AI actually is and why it's different. How to think about it differently so you actually use it right. The perfect tool for your scenario, not just some hyped up crap. And if you stick around to the end, I'll walk you through your first fully agentic workflow step by step. So let's start by understanding what it is in the first place. Point one, what is agentic AI? There's three levels of AI. The first level that we all got introduced to was chat. ChatGPT, Claude, Gemini. I talk to it, I give it context, it replies. It's gotten better over time. The second level is automation. And that's when we start taking the AI and its ability to classify and respond, create workflows that run automatically. And these are tools like Make.com or Zapier or N8N for the more advanced people. Level three is Agent. This is where the AI thinks, plans, reasons, and executes entire tasks on its own. It opens browsers. It writes code. It creates files. It does research. It doesn't need your hand holding. Most people are at level one. Maybe level two, but the right people are moving to level three. And the best part, you don't have to go to level two to get to level three because it'll do it for you. Point number two, the new mindset you need. Here's the biggest mistake that people are making with AI right now. Truth is they're barely using it. At best, they might be using it as a thought partner when it could actually build for you. This mindset shift that I see that I encourage people to consider, which is you are the director, not the doer. You need to be designing, not the task maker. And people think, well, if I don't do it, who's going to do it? AI will do it. Your job is to give direction, to review the work, to course correct, not to do the work yourself, which sounds crazy. It's called reverse prompting. Essentially, you start with the outcome. What is the goal? And then you let AI build a plan to accomplish the goal. If you're influencing it, then you're telling it what to do. And I'm telling you, it knows better than you. So if you haven't heard. IBM rolled out AI agents for their 270,000 employees. It was a $4.5 billion in productivity gains. Managers now complete tasks like managing their team and figuring out who needs a promotion 75% faster. That's not because AI is smarter than them. It's because they stopped doing and started directing. So how do you actually become a good AI director? There's only three things you need to do. The first thing is you have to have a clear outcome. You need to know what's the problem to solve. Before it gets going, because it can do a lot of things, you gotta ask yourself, what is that result? What is that outcome you want it to do? The second thing is we wanna give it clear instructions. I think about what is an example of an output? If I'm gonna ask it to do a task, can I give it a format? Can I give it the requirements? Can I give it the template? The more specific you can be, the better it's gonna work for you. And three, clarify the results. I want you to treat it like an intern. give it feedback, tell it to save the feedback, and it will remember. Because as we build this automation, this agentic workflow, it'll be able to do it on its own. If you do those three things, the AI stops giving you generic garbage and starts giving you exactly what you need. And notice how in none of those steps, we actually tell it how to do the work. We begin with the end in mind. It's its job to figure out how to get there. The CEO of Notion, which is the one of the top productivity tools out there, he calls it the manager of infinite minds. When you have this army, of infinite minds out there accomplishing, researching, feeding back to each other, and then you can sit in your cockpit and being the director, the future is going to look completely different than how you got to this point in your life. And again, I'm going to show you exactly how to do this and it's super simple towards the end of this video. If you're a business owner and you're trying to scale past seven figures and you're kind of struggling figuring out ways to actually integrate AI into your business, I created my AI implementation workbook for a lot of my friends because I wanted to walk them through how we integrate AI into every department in my company step by step. So if you want a free copy, just find me on Instagram at Dan Martell and DM me the words AI business and I'll send it right over. There's a bunch of AI agent tools out there and it can confuse you. So I'm going to tell you which one is right for you. Point number three, which tool is the right tool? If you ask AI to tell you what tools you can use to create agentic systems, it'll give you a bunch of different tools because there's a bunch and they're coming out every week. And every tool is great for one specific thing. So you could go use them all, but that would mean you'd be a master of none. What I'm going to recommend is you go deep on one. Become an expert in that. I'm going to give you the answer to which one is right for you based on who you are. But once you know that, I want you to go all in. So here's how we pick the right tool. If you're a business owner who needs research, content, and general tasks done, use Manus AI. Because it is the best all-around agent for getting real work done at the filming of this video. If you're a creative who writes, designs, or works with ideas, use Claude Cowork. They call it co-work because it actually runs on your computer. It can manage your files. It can clean up folders. It can open up tabs on your browser and essentially can accomplish complete projects for you without you involved. Now, if you're a developer or someone who writes code, use Cloud Code. It can fix bugs. It can add test files. It can do work in parallel to your code base while you focus on other harder problems or talk to customers or test what is built. It is the tool that all the developers at the other AI companies use to build that AI. It's kind of wild. And if you want a completely automated, full personal assistant that runs on your computer and you're a little nerdy, use OpenClaw. The reason why is it runs like a person. It'll understand who you are and it has memory. I think it gives you a glimpse of what the future will look like at scale, but it is a little technical setup and truthfully risky. So be careful. I saw an example where the bot bought a $3,000 course for itself so it could be a better agent with no permission. The point is, is pick one. Don't try to learn them all right now. I know you're excited and you're like, oh my gosh, this is the future. But the more you use the one, the more you ask it to teach you how to use more features, the better you'll get at directing AI, which is the skill we need to develop. Let me actually show you exactly how you can use AI agents because I'm going to go through a real task with Manus right now. I'm going to make it easy for you to understand so you can follow along and see exactly how this works. Point number four, Manus in action. So for example, if I'm the founder of a digital marketing agency and I want to do some research on my competitors so I can see what they're doing and figure out if I got to change my pricing or my offer, I can get Manus to do that. So just go to Manus and type something like this. Research the top three niches, digital agency in Canada, finding their pricing, main features, and what they're doing well. Then create a simple one-page website that summarizes the findings. So notice it's doing the research. It's finding those three companies. It's building a task list so they can follow. It's writing the code of the new website for the output. It's doing it, and I'm not touching the keyboard. And I didn't give it any expert detailed instructions. Even somebody that doesn't know a whole lot about anything could have wrote what I wrote, and they would have got an expert level output. Most people would have took a week plus and they probably wouldn't know how to code this and they would have confused it. This did it automatically in less than 10 minutes. And my favorite part is within Manus, I can now iterate because, you know, there's some things that are missing or maybe I want some changes. So I'm going to tell it. Now add a section to each one that includes client testimonials. Enter. Watch. It's going. It's literally updating itself. And just like that, I now have client testimonials that were pulled from each of those companies, did the research and updated the website in minutes. So now I'm going to ask it to send it to Sam Gaudet, who's my creative director in Slack and ask him to provide feedback. And then also email that link with the details of what I did in the research to Joel Harrison. Now I go check Slack. It posted it as me and it has a little icon that says Manus. So they know that it was from Manus. I can now get feedback from my team and apply it to that website. So I can tell it to monitor that Slack thread and see if there's a reply every 15 minutes, every hour, whatever I want to tell Manus. And if it gets feedback, it's going to either prompt me or I can just say, take any feedback and automatically apply it. And it'll apply it to the website. It did all of that. It did the research. It created the website. It shared it with my team. It took their feedback. It updated it. It took my feedback, updated it all within minutes. And what's cool is you can kick off a bunch of those in parallel and they'll all run while you're doing something else. So now you understand. I went from an idea to a prompt, to sending it out to people I work with, to getting the feedback, to updating the output. And I can even say now email or post on Slack to everybody that's been involved, letting them know the change is made. I do this all the time. It's amazeballs. It's crazy. The fact that without knowing how to create automation, you've just created automation. My favorite part, it's mobile. So on my phone, I can sit there and I can talk to it and I can tell it what I want it to do. Spins up, runs it. connects to stuff, pulls everything. Now, here's a pro tip. I challenge you to stay in the tool. See, we've been taught because of the chat level that we need to chat and it gives us a response and we copy the response and then we go do something with it. Post it in Slack, send an email, create a Google Doc. Don't do it. I dare you not to do it. You'll be tempted to take action. Don't. That way you get better and the AI gets better because it's doing more for you and it's learning how you like to work. So now that you've seen how it works, this is the question. What are you going to do about it? What are you going to change? AI is the opportunity you've been waiting for. You've been asking, like, why wasn't I early to the internet? Oh my gosh, I wish I would have came out with that mobile app idea. This is now. And the next five years is the AI gold rush. And there will be more millionaires created in the next five years than the history of the internet. Just like the dot-com boom. This is bigger. This isn't about being an AI expert. The cool part is AI is the AI expert. It's about being willing to learn. It's about telling yourself, I need to just sit down and do the work. And I know I'm not technical and I know this still confuses me, but I'm willing to read, I'm willing to research and I'm willing to empower the tool to help me in my life. And the best part is, is you just saw how easy it was to do something that seems complicated and it did it in minutes. Stop pretending not to know. I know it's overwhelming. But if you just take action, I promise you it will click. So here's what I want you to do right now. Pick one tool based on what I said earlier. Number two, pick one task that you're doing every week that takes you a lot of time. Three, have the agent do it for you today. Sit down with it, play with it. If you do that, you will start to change the mindset and the belief and understand how easy this is. And you will get ahead of 99.9% of everybody else using AI to do chat. And remember, if you want my free copy of my AI implementation playbook that we use in all my companies, just DM me the word AI business on Instagram and I'll send it right over. And if you want to learn what are the best AI tools to make money in 2026, click here and I'll see you on the other side.

⚙️ Pipeline jobs

StageStatusAtt.UpdatedError
download done 1/3 2026-05-28 08:50:04
transcribe done 1/3 2026-05-28 09:01:49
summarize done 1/3 2026-05-28 09:02:08
embed done 1/3 2026-06-30 06:43:56

📄 Описание YouTube

Показать
✅ Get Your FREE AI Company Operating System here: https://go.danmartell.com/4tVaYZG

👥 Are you building an AI software company? Partner with me: https://go.danmartell.com/4b7mFoS

This might be the most important shift in AI we've ever seen, and it’s happening right now.

I’ll walk you through what agentic AI is, why it’s different from ChatGPT, how to choose the right tools, and how to set up your first fully automated workflow step-by-step.

Tools mentioned:
ChatGPT: https://chat.openai.com/
Claude: https://claude.ai/
Gemini: https://gemini.google.com/
Zapier: https://zapier.com/
Make: https://www.make.com/
Manus: https://manus.ai/
OpenClaw: https://openclaw.ai/

▸▸ Subscribe to The Martell Method Newsletter: https://bit.ly/3XEBXez

▸▸ Get My New Book (Buy Back Your Time): https://bit.ly/3pCTG78

IG: @danmartell