JESUS, this is so helpful, dude, I wish more people on RUclips knew how to teach like this guy does. n8n went from overwhelming to super-exciting in a matter of 2 short videos. I was "seeing" all of this stuff in my n8n instance but just had no idea what I was looking at. And now, it all suddenly makes perfect sense with just a little intelligent explanation. Well done, and seriously, much thanks.
These videos are awesome. I've looked at about 20 different people teaching N8N. They're all good but your delivery and how you simplify is everything. So on point. I would love to see you connect this to 11 labs Voice AI and the Google calendar for booking. Thanks much.
Thank you, Max, for this super clear video about the AI agent in N8N! 🙌 Your content is a goldmine for anyone looking to automate and integrate AI into their workflows. 🚀 Keep it up-it’s always great to learn something new from your videos! 👏💡.
Thanks a lot for the awesome content. I think lots of people might want to see a stress test of workflows with supervisor agents delegating tasks to other tool agents. I've seen some pretty examples on yt, but nothing really reliable. Or maybe a robust assistant, deciding and using multiple tools seamlessly.
Great video!!! Part 3 could explore the possibilities of using OpenAI's o1 model (and how to set it up). It might also be interesting to mix it with a tutorial on how to use ReAct Agents and the Plan and Execute Agent. :D If there’s a video about this elsewhere, please let me know! :)
Thanks for the ideas :) FYI It's likely those other agent modes will get deprecated soon - things change fast in AI space and overall performance from tools agent (and roadmap for it) should be great for tool and non-tool usecases (and I suspect would then just become the "AI Agent" without a mode). So don't invest too much time researching those within n8n context :)
@@theflowgrammer Perfect Max! Thanks for sharing that :). Have you used the o1 model for any AI agents yet? I don't know how to get it to work. I know that it's not possible to use system prompts for that model yet, so how do I set up a an AI agent up that get's around this issue? I get the error even though I'm not using any system prompts 😄
@@future-of-tomorrow I have not used o1 yet for AI Agents. Part of that reason is o1's benefit is billed as "thinking more deeply" which is something you already have control over with AI agents (since it makes multiple calls to complete a task and outputs final answer), or via the workflow (multi-agentic or serial agentic approach), or both! Other than "thinking longer", o1 is slow and expensive. I have reports from users that the new Gemini 2.0 models are working fine in n8n (on my list to tinker with them this week). And since they are fast, affordable (+ great free tier), and seemingly capable; I'd recommend trying that out instead :)
Awesome video Max, the prompting tips definitely help. 1) I see you're using # in your system prompt e.g. #role, any reason for that or just breaks it up for you visually? 2) would love to see you cover tools agents calling other workflows with specific inputs & manipulating the outputs - as there's no $fromAI( ) there's definitely more ambiguity in the prompting
Hopefully releasing part 3 that focuses on prompting. 1) # Role is for my benefit as reader but also for the AI. How much that helps will probably reduce over time as models get smarter (and less helpful the shorter your overall prompts are). 2) Try 1.74.1 (latest beta atm) it overhauled subworkflow calling and supports fromAI now on subworkflow tool. Should be covering it in this week’s episode!
Please make video on how you would configure your agent to make use of files in google drive for context to subject matters. Also some prompt engineering would be great...
Max here, working on prompt engineering walkthrough as we speak! Might do a full-agent build out after that, will keep your gdrive suggestion in mind for that. Cheers!
Great suggestion! If you're looking for RAG solutions, we collected quite a few videos on that topic on our Tutorial Hub: community.n8n.io/c/tutorials/28
Hey Max could you go over what are the benefits of using an AI agent vs similar workflow that prompts a model directly and the outputs are mapped to the nodes you need?
Sure I’ll try to cover that in future! Tl;dr is it’s about autonomous action: deciding to do n actions in a specific series (or not so specific if that’s the task) to accomplish something. Simple example: if I’m creating an article summariser; it might be easier to have my agent to find 15 relevant articles, vs a predefined process. “Relevancy” includes keyword identification in request, ideating related keywords, searching, filtering out unrelated. More and more I’m seeing tasks like that expressed more easily as an Agent, than a multi-step predetermined flow.
For part 3 I would be interested in knowing what to do if I need to transform the data before using a tool. e.g. Running the AI and then formatting the result through a code node before sending it to a telegram node.
Great point! Yes, most likely you would want to add some filtering before (recommend the Filter node there; and starting with filtering by category). Doing this makes sure you don’t have an LLM call for non-drafts. My aim was to pick a generally relevant usecase when showing how tools work. In future I’ll clarify better, and give folks ideas on what they could do next.
Great question! I show how to set up AI tools in n8n using a simple gmail draft example; but you can apply these concepts to other usecases (that gemini does not have). Here's a great example from my last Studio Update where Oumnyaa shows me her multi-agentic workflow (AI Agent consuming other AI Agents as tools, which use a constellation of tools themselves to get a hell of a lot done): ruclips.net/video/63Nw2Pt1VFw/видео.html
community.n8n.io is the best place to make a feature request (or upvote existing ones). The n8n product team is checking in there constantly to help inform the roadmap.
![Building AI Agents: Prompt Engineering for Newbies [Part 3]](http://i.ytimg.com/vi/77Z07QnLlB8/mqdefault.jpg)
![Building AI Agents: Prompt Engineering for Newbies [Part 3]](/img/tr.png)
![Building AI Agents: Chat Trigger, Memory, and System/User Messages Explained [Part 1]](http://i.ytimg.com/vi/yzvLfHb0nqE/mqdefault.jpg)
![Felix "Unfair" | [Stray Kids : SKZ-PLAYER]](http://i.ytimg.com/vi/Oswujxm2Ag0/mqdefault.jpg)
![Blox Fruits Dragon Rework Update [Full Stream]](http://i.ytimg.com/vi/EqDAp8udhm0/mqdefault.jpg)
JESUS, this is so helpful, dude, I wish more people on RUclips knew how to teach like this guy does. n8n went from overwhelming to super-exciting in a matter of 2 short videos. I was "seeing" all of this stuff in my n8n instance but just had no idea what I was looking at. And now, it all suddenly makes perfect sense with just a little intelligent explanation. Well done, and seriously, much thanks.
You just made my FREAKIN day man, thank you!
These videos are awesome. I've looked at about 20 different people teaching N8N. They're all good but your delivery and how you simplify is everything. So on point. I would love to see you connect this to 11 labs Voice AI and the Google calendar for booking. Thanks much.
Thank you, Max, for this super clear video about the AI agent in N8N! 🙌 Your content is a goldmine for anyone looking to automate and integrate AI into their workflows. 🚀 Keep it up-it’s always great to learn something new from your videos! 👏💡.
Incredible tutorial! Thanks for putting this together. Looking forward to seeing more!
Thank you for this series
master class ! AI for ALL, thank you!
Please add a tutorial that uses a web scraping tool and one that allows the app to use imported pdf (and other file formats) as context.
Thanks a lot for the awesome content. I think lots of people might want to see a stress test of workflows with supervisor agents delegating tasks to other tool agents. I've seen some pretty examples on yt, but nothing really reliable. Or maybe a robust assistant, deciding and using multiple tools seamlessly.
after a long time of searching for good n8n videos. i completely watched and loved it.. i really should make a tatoo of n8n ;)
we want more tutorials ❤
Working on it!
Love the video❤️‼️
FOR PART3 HOW ABOUT YOU SHOW US HOW TO AUTOMATE GOOGLE CALENDAR BOOKING😎
Great !!, thank u very much
Best Video Ever❤❤❤
Great video!!!
Part 3 could explore the possibilities of using OpenAI's o1 model (and how to set it up). It might also be interesting to mix it with a tutorial on how to use ReAct Agents and the Plan and Execute Agent. :D
If there’s a video about this elsewhere, please let me know! :)
Thanks for the ideas :) FYI It's likely those other agent modes will get deprecated soon - things change fast in AI space and overall performance from tools agent (and roadmap for it) should be great for tool and non-tool usecases (and I suspect would then just become the "AI Agent" without a mode). So don't invest too much time researching those within n8n context :)
@@theflowgrammer Perfect Max! Thanks for sharing that :).
Have you used the o1 model for any AI agents yet? I don't know how to get it to work. I know that it's not possible to use system prompts for that model yet, so how do I set up a an AI agent up that get's around this issue? I get the error even though I'm not using any system prompts 😄
@@future-of-tomorrow I have not used o1 yet for AI Agents. Part of that reason is o1's benefit is billed as "thinking more deeply" which is something you already have control over with AI agents (since it makes multiple calls to complete a task and outputs final answer), or via the workflow (multi-agentic or serial agentic approach), or both! Other than "thinking longer", o1 is slow and expensive.
I have reports from users that the new Gemini 2.0 models are working fine in n8n (on my list to tinker with them this week). And since they are fast, affordable (+ great free tier), and seemingly capable; I'd recommend trying that out instead :)
Part 3: RAG both Cloud and Local
Great video!
Thank you Max 💜
Catch part 1 here if you missed it: ruclips.net/video/yzvLfHb0nqE/видео.html
Awesome video Max, the prompting tips definitely help.
1) I see you're using # in your system prompt e.g. #role, any reason for that or just breaks it up for you visually?
2) would love to see you cover tools agents calling other workflows with specific inputs & manipulating the outputs - as there's no $fromAI( ) there's definitely more ambiguity in the prompting
Hopefully releasing part 3 that focuses on prompting.
1) # Role is for my benefit as reader but also for the AI. How much that helps will probably reduce over time as models get smarter (and less helpful the shorter your overall prompts are).
2) Try 1.74.1 (latest beta atm) it overhauled subworkflow calling and supports fromAI now on subworkflow tool. Should be covering it in this week’s episode!
make a tutorial on how to use IA to learn and receive guidance to operate n8n
Please make video on how you would configure your agent to make use of files in google drive for context to subject matters. Also some prompt engineering would be great...
Max here, working on prompt engineering walkthrough as we speak! Might do a full-agent build out after that, will keep your gdrive suggestion in mind for that. Cheers!
Great suggestion! If you're looking for RAG solutions, we collected quite a few videos on that topic on our Tutorial Hub: community.n8n.io/c/tutorials/28
Hey Max could you go over what are the benefits of using an AI agent vs similar workflow that prompts a model directly and the outputs are mapped to the nodes you need?
Sure I’ll try to cover that in future!
Tl;dr is it’s about autonomous action: deciding to do n actions in a specific series (or not so specific if that’s the task) to accomplish something.
Simple example: if I’m creating an article summariser; it might be easier to have my agent to find 15 relevant articles, vs a predefined process. “Relevancy” includes keyword identification in request, ideating related keywords, searching, filtering out unrelated. More and more I’m seeing tasks like that expressed more easily as an Agent, than a multi-step predetermined flow.
For part 3 I would be interested in knowing what to do if I need to transform the data before using a tool. e.g. Running the AI and then formatting the result through a code node before sending it to a telegram node.
One approach would be to use the subworkflow tool to handle it. You'd have full control over the data that way.
Great tutorial, but this flow will clutter your gmail with a new draft every minute, right? Adding a label and filtering would solve it
Great point! Yes, most likely you would want to add some filtering before (recommend the Filter node there; and starting with filtering by category). Doing this makes sure you don’t have an LLM call for non-drafts.
My aim was to pick a generally relevant usecase when showing how tools work. In future I’ll clarify better, and give folks ideas on what they could do next.
gemini already has built in email drafting and inbox analysing feature natively....whats the need for these types of tools...
Great question! I show how to set up AI tools in n8n using a simple gmail draft example; but you can apply these concepts to other usecases (that gemini does not have). Here's a great example from my last Studio Update where Oumnyaa shows me her multi-agentic workflow (AI Agent consuming other AI Agents as tools, which use a constellation of tools themselves to get a hell of a lot done): ruclips.net/video/63Nw2Pt1VFw/видео.html
How can we get a google drive to download two files and attach it to a email
Please post this question in our Community forum - it's the best way to get help on detailed questions like this: community.n8n.io/
Can we expect an AI Agent tool for firestore like we have for supabase and mongodb? Thanks.
community.n8n.io is the best place to make a feature request (or upvote existing ones). The n8n product team is checking in there constantly to help inform the roadmap.
Great but hope to add the system message texts plz
hey so didn't you say the coupon code is 12 months off?
Correct! It will work both for our monthly and yearly plans though.
Slack integrations pls
how did you get dark mode?
Select Settings > Personal.
Under Personalisation, change Theme to Dark theme.
Cooool
OR at least a shareable link