Building AI Agents: Prompt Engineering for Beginners [Part 3]

[@amerrashed6287]

the most important lesson for all n8n tutorials on YT. it should be pinned forever in n8n page :). thanks Max.

[@petrkadlec6060]

I just binge-watched all 3 parts in your "yet-to-become-a-huge" series and it's hands-down one of the best structures of the video covering a theory and practice on any topic I've seen. 3-level prompting for AI Agents is awesome, I'm building a multi-step automation for a client running a bunch of dental clinics, so very useful to get some inspiration. Thank you. As for next topics - multi level workflows - storing things in the database, retrieving, db memory for the agent, using output in multiple steps of the workflow.... I do have basics, but I feel I can learn a bit here and there anyway. Thank you and keep up with the great work!

[@youtubeccia9276]

Max for AI President

[@jorgecantero7101]

Hey Max, thank you so much for the videos you create. They are super illustrative with the use cases for agents. It's great how you separate the system layer, the user message layer, and the action layer. This works fabulously well for agents with reduced tasks, but when we are building a multi-agent structure where there's an orchestrator agent managing area director agents, who in turn manage sub-area managers, and they, in turn, oversee micro-agents performing tasks-creating a team of over a hundred agents arranged in a pyramid-I have many questions about the best way to approach tasks like this regarding the inputs and outputs of the different agents, which become tools for higher-level orchestrator agents. If you could explain or make a video about this, it would be fantastic.

[@EricMooney]

Awesome instructional video, thank you so much. I'm experiencing that Rules section now and my first attempt, I entered all my rules at the beginning and the tools were not being called and wasn't getting the desired output. So I did the opposite on my second attempt and started with just the connections to the tools and the date of today. Then one by one, I started writing in the Rules of the workflow and I felt like I had much more control doing it that way. Thanks again and keep the videos coming!

[@joeymuller4842]

Pure gold. Now I gotta go back and watch parts 1 and 2.

[@aftabdx]

Thank you, Max. Love the way you structure your prompt. Look forward to your next series!

[@ThomasMock-c5n]

Great video! Looking forward to the next in the series!

[@DQ940]

Great video. You have a great ability to communicate. Would be great if you could use your n8n influence to see if you could allow metadata to be accessible and available to use by the AI agent from the vector store tool. I'd imagine this would be a pretty simple thing to do since the data is already there when you look at the logs. Also, in the longer term, support for streaming apis is something that I think n8n is going to have to see if it could manage, even though that might need a lot of work under the hood.

[@theflowgrammer]

Thanks a lot for that. Sharing with the AI squad now and will advocate for it. Can see how that would be useful 100%

[@theflowgrammer]

Just heard back from the AI Squad: the new App vector store tools like "Supabase Vector Store" do by default send metadata. These are relatively new so make sure your n8n is up to date. FYI, the "Vector Store Question Answer Tool" specifically indeed wouldn't include metadata since the output is intended to be passed to an "answer LM" and only then to agent. Hope that helps. Streaming alas won't come overnight but will continue to advocate for it :)

[@musumo1908]

Great vid…would love to see n8n add langraph 😀

[@requesttruth505]

You do a really good job of teaching this material. I find it easy to implement and understand. Thanks.

[@hirunadilmith9457]

need part 4 💟

[@MMMM-ps5ok]

Very helpful, very well presented! Look forward to future videos 👍

[@Superturisto]

Beauty AI prompting 👌

[@SiraEkabutPlus]

this episode is really good. as always ❤

[@mirkonik8788]

Awesome tutorial. I really appreciate your effort. Really nice!

[@rubengouws120]

Very nice video, a lot of information

[@AlfredNutile]

Thanks I was really stuck at the calendar tool and not sure how to limit the date range!

[@guiduv]

Really good thanks!

[@LeonardoPinto]

Great stuff, thanks!!!

[@DARKKLIN1]

Thanks for the very informative video-it was exactly what I needed! However, I’m facing a small problem with my AI agent, which I’m building as a personal trainer for a real client. The issue is with the language: I need the AI to speak exclusively in Hebrew. Most of my prompts are written in Hebrew, but sometimes I use English to define tools or specific instructions. The problem arises when combining Hebrew (a right-to-left language) with English (a left-to-right language) in the prompt editor. The text formatting looks messy and inconsistent when mixing the two languages. I hope n8n will address this issue and provide better support for combining languages in the editor.

[@ScottCLemon]

This is a great series! Where did you share your project? I want to use this as a framework for a flow … is it shared someplace?

[@djselcius]

Getting a few questions around this. Once I ship this week’s show, will try to get it together and share!

[@robertmarquardt]

Awesome.

[@wEBMedJA]

Hey! Quite nice, but was wondering about more elaborate markdown optimisation of the system prompt as well as other frameworks like ASPECTT and so on. Also, better tools for prompt engineering, like iterative teams for AI system prompt engineering implemented in Flowise or Anthropic API Dev tools (generate/improve prompt). My biggest question is what is the optimal organization method/structure for huge system prompts (30k+ signs).

[@theflowgrammer]

Thanks a lot for this, noted and will keep in mind for future content! Will also pass feedback along to the AI squad.

[@danielguesukkang6964]

I love you Max

[@foodfarmersmarkets5249]

Is your workflow anywhere documented for copy & paste?

[@theflowgrammer]

Hey sorry they were not (and did not save them at each step alas). What in there would be useful? The prompts or generally whole thing? Thanks so much for the comment, feedback like this helps me improve!

[@kel78v2]

Structured output node would sometimes break due to special characters, more frequent when code outputs are involved. What's the best way to handle this?

[@Tejesh-r3g]

Is it possible to integrate with semrush, or ahrefs for seo purpose

[@ЖоламанСих]

can i get text of system role?

[@alfatrading]

In this video module Webhook appeared from nothing (was already in canvas since the beginning), and when I tried to create one like yours I found it could workonly if I register it for GET.. Or I would get this error message: "This webhook is not registered for GET requests. Did you mean to make a POST request?"

[@djselcius]

Ah sorry about that! In webhook node you can swap GET/ POST etc. If you want to quickly test it via web browser (e.g: “run” the test URL) then you need to use GET, to simulate a POST request - use postman OR actually an HTTP Request node in another workflow :)

[@GirdharKumar-x4p]

i want to know how to integrate n8n agent with any website like i have a web using DRF and i want to add n8n agent with my web can anyone help to do this

[@ThomasMock-c5n]

Great video! Looking forward to the next in the series!