mate u told me more about application of genai than 99% of videos on youtube. You explained WHAT is an agent and WHY should we use them. The rest are just trying to explain a tool or an SDK or a transformer but if don't know what is the end goal of all this, its quite irrelevant. Thanks for the video, much needed.
This was extremely valuable. It exudes intelligence, and I will definitely watch other videos from you. My takeaway is that there is a lot of work to be done by non-coders around thinking about what their workflows are and how they are enabled by different tools or subtasks and how to lay these tools / tasks down on a piece of paper and connect them. The one area that still prevents me from embarking on that journey is the lack of familiarity with a lot of the tools you mention in your video and the lack of time in my day to both do my job as well as take time to automate it. However, videos like this one show that it's possible and worth it in the long run. As always, play long term games, get long term results
I'm in the market for AI agents and did about 3 hrs of research yesterday. I tried a few free trials for a simple task.... I was so exasperated at the clunky architecture they pushed me into that also *didn't work* or only worked somewhat. N8n reminds me of an in-house data pipeline tool used at a past Big Tech co job which was adopted by hundreds of employees, it made things so easy. Hoping n8n is too. Thanks for the presentation. Awesome work.
Mr. Devin I would hug you if I could, for this video. I would love to see a series of videos based on this overview, especially an in-depth episode on the whole Architecture section. Cheers!
Why do you see CSM being replaced though? I don't envision enterprise level clients that expect a white glove experience from their vendors being okay with working with an AI agent.
Amazing video. As a software engineer of over 10yrs, I know your life would’ve been way easier if you were an engineer. I think I will explore a lot of things around this over a weekend. And hopefully documents my experience like you did.
With AI virtual agents, businesses can offer more dynamic and responsive customer interactions. These agents leverage advanced algorithms to understand and address customer needs effectively, leading to higher satisfaction and improved service
00:03 18 months of building autonomous AI agents 02:10 The future of AI agents and their role in AI space 06:25 AI building platforms are focusing on AI agents 08:28 Inen platform is intuitive and feature-rich for building AI agents. 12:43 Using Pinecone for Vector database storage 14:51 Database and prompt engineering are foundational to building highly effective AI agents. 18:30 Providing context and detailed instructions to autonomous AI agents is crucial for their performance. 20:15 Examples are crucial for AI agent success 24:11 Demonstrating AI agent's email and calendar actions. 26:12 Overview of Email Actions Tool and Integrations in Building AI Agents 30:06 Building agents requires a change in integration thinking 31:58 Agents excel in complex workflows with reasoning and decision-making. 35:40 Automated lead qualification and enrichment process 37:34 AI agents evolving beyond chatbots and voice assistants 41:20 Leverage AI agents for business growth
Love this. You should start a group where you teach this at price. Question: Can you provide guidance on a framework for prompting for an assistant/agent? You have a quite comprehensive prompt there.
Hoping to come out with courses soon. I have a custom Prompt GPT that uses our framework. Here's the link: chatgpt.com/g/g-Cehuz3C7p-custom-ai-studio-prompt-generator
Thanks for sharing this material! I’m also working on AI automation in Python-mine is custom, but what you’ve shown here is quite similar, especially with the user-friendly visual dashboard. It reminds me of how email platforms evolved about 10 years ago. The great news? Just 100k views on this video means there’s still room for us to learn and build, staying ahead of the curve! 😎
Great explanation of agents and their application in the workspace. I am genuinely excited about this next iteration of application development. Sadly, AI is not cheap. You have highlighted that on the technical side, an agentic solution requires marshaling and preparation of numerous applications, interfaces and prompts - much as pre-AI solutions required. This takes time, expertise (even for non-coders) and digital resources, which all cost money to establish. The elephant in the room that NOBODY talks about is the cost of the prompting. Few organisations will have their own copy of a universal LLM like ChatGPT, Groq etc., so they will have an API contract with an LLM provider or broker instead. Iterative use of an API using a previous response as embellishment to the previous prompt (APIs don't have memory at the LLM source) makes round trip, lights-out enquiries quite expensive. (Just my $5-per-million-tokens-worth...)
@@customaistudio I'm sure the agentic semantic will be the dominant model to build business applications, but I'm not so sure there will be a dramatic price drop. Those LLM companies have got to claw back those invested billions somehow, and it won't be via big balloon payments for massive contracts. What to do then? Check out the Economist magazine (Aug 8th edition). In the Culture section, there is a review of one of the latest books on AI, "Co-Intelligence: Living and Working with AI" by Ethan Mollick; (btw, over 100 AI titles have been published in the US in the last 12 months). The book covers all things a layman needs to understand to interact successfully with a LLM (via Chat GPT, say) rather than considering the agentic approach, but the review ends with a final comment about an unexpected emergence in AI: "...Amid AI hype in business, where companies say a lot but seem to do little, “Co-Intelligence” usefully notes that innovation is hard for organisations but easy for individuals. Hence, do not look for how AI will change business from chief executives’ statements but from ordinary worker-bees who quietly incorporate it into their everyday tasks. The revolution will be noticed only in hindsight." Agentic solutions may be less ad hoc than the glorified chat sessions covered in the book, but their footprints will be modest and their impact profound (so long as we can really, really trust the LLM responses!).
A Docker container on a cloud service provider like AWS will let most businesses be able to run their custom LLMs in their own agent system, albeit for most meaning they have to rely on open source weights.
Very nice video and very helpful. Thank you, Davin. Clean and direct explanation. I would like to see a video that covers more platforms, frameworks, tools, databases, and data structures available and ready to implement for building agentic and autonomous systems, as well as your peer review on those. Great content!
Thank you! Our main focus is showing how SMB or agency owners can start to leverage AI Agents for their business. Reviewing platforms and other tools isn't really on the roadmap. But if we get more requests for it, then I'd happily talk about it!
Ngl i like this no code view. U up for a voice call? I would love that i come from a perpective of more code the better so to have a discussion would be lovely indeed ngl.
Amazing video, Especially the way you have explained your journey. Even i have been building agents and pushing it to production mostly using crewai. From the frameworks persepctive, they are not goof for production, I have built my own abstraction over the openai assistants and function calling, and it is fast and more reliable than crewai. A lot of work needs to be done by there frameworks to be stable.
Exactly, crewai seems very fragile right now and even fixing bugs or troubleshooting issues is tough. Luckily, N8N offers no-code options for building Agents.
@@customaistudio they do push interesting updates, but the whole thing being written on python makes it "slow", plus i do not know why but normal openai assistant calls are much more reliable in terms of quality + speed
18 months of building autonomous AI agents in just 42 minutes is impressive! SmythOS also focuses on enabling rapid development and collaboration among AI agents. Great work! #AI #SmythOS #Innovation
Outstanding Video. Normally Ive switched off within 5 mins, but I sat through the whole video. Very well explained and great ideas for what to use this for. New Subber and Bell set. Thank you
@@customaistudio The ref to HR staff made me laugh, as my wife is an HR consultant and laughing when I tell her I have an Ai Agent to replace her lol You know exactly what she will reply that the Ai Agent cant do..lol
🎯 Key points for quick navigation: 20:52 *📧 Emails categorized as work or high priority should use tools like the categorize tool and CRM for effective management.* 21:32 *🛠️ Tools are essential for AI agents to function autonomously, enabling actions like managing inboxes and replying to emails.* 22:13 *🔄 Custom workflows in tools like NadN empower agents to handle diverse scenarios within platforms like email inboxes.* 24:20 *📅 Agents can perform tasks like sending emails and scheduling meetings based on predefined workflows and tool integrations.* 28:03 *🔄 Integrations for AI agents involve configuring access scopes and data flows, treating agents like human employees needing specific platform accesses.* 30:34 *🏗️ Building effective AI agents involves structuring workflows under job functions, enabling agents to reason and execute complex tasks beyond simple automations.* Made with HARPA AI
Fantastic insight. I evaluated n8n in the past, but ended up going with Wappler instead. I want to integrate similar agents into it. I would love to see another video where you build the same functionality in Wappler.
It would be great but my approach to AI is to build tools for business owners who don't know how to code. If you're looking for coders in the AI space, I'd check out @vrsen @AIJasonZ @IndyDevDan. They're my favorites.
@@customaistudio Of course but I just mean so you are able to understand the limitations and communicate with your customers who may be technical is all. Just my 2 cents.
awesome video, didn't know n8n added all these AI nodes and features! I've always enjoyed for workflow automation that involves data processing, but found it weak for other use cases due to lack of OOTB integrations.
Wow! That's is exactly what I was looking for as an entrepreneur in this ocean of videos about AI Agents. Thank you very much for sharing your experiences with us! 😊
Came here to say exactly the same too. So much spam tier grift online lately but this cuts through. Another is @dougsilkstone but they do more general AI automation hackery
Awesome overview, Devin! In what cases ARE you using multi-agent architecture where agents are conversing, giving critiques of each others’ output for improved quality?
For output quality, we attach separate AI nodes that will do proof-reading and editing. I don't like to add extra prompt instructions for the Agents on how to write emails or create content. I've found that the output isn't consistent.
25 วันที่ผ่านมา +1
This video is gold, anywhere we can get your prompting structure?
This was amazing. Thanks. Have you thought about fine-tuning an LLM with QA pairs around scenarios/output you're looking for before plugging it into something like n8n to reduce number of tokens required and thus cost, or is it over-engineering for what would be marginal decreases in costs?
Good question. What we want to do is build Agents as fast as possible while still maintaining their effectiveness. At times, we'll do stuff that isn't the most cost efficient just for simplicity and speed sake. So although I haven't thought about your approach, I think you're right in that it could be over-engineering. Always curious about new approaches though, if you've built something similar, feel free to send it over!
When the student is ready, the teacher will appear. I've just finished watching all four videos and am eager to explore more of your free content. It's excellent!
Thanks Devin. That's very helpful. I 've been working on various agents for legal vertical. Whats your take on specializing within certain industries? Or better generalize?
Because no-code/light code solutions put the solution creation in the hands of the people with the primary business knowledge and move light to medium business process automation tasks off the dev backlog.
This is some great stuff man, thank you!! Quick question, when you make ai agents for your clients do you do it in your n8n account or set one up for them? I've been using Zapier and Make but n8n's agentic approach looks spectacular. Again thank you for sharing this :)
Thank you! We have them create an N8N account and we build it in there. We can also easily share the workflows from our account to there's - which allows us to build out templates that we can reuse.
can you please cover what ex google chief Schmidt said about agentic workflow breakthroughs and please please tell us why Agent Zero is such a big deal. thanks.
Yes I will. Agent Zero is cool but still untested in production. It's also not as attainable for an average business owner. I'll touch on those updates and Eric's comments in coming videos for sure.
Really informative video. Thanks! So I would draw a distinction between automations where a human structures the tasks the system performs vs an agent where the system defines/plans the actions it will take and then takes those action, evaluates the result, and may revise the set of actions, iterating until the desired result is achieved. This is not an important distinction until you are trying to assess the capabilities of models. My understanding is that llms can be successful within automation, but tend to fail in agentic systems. Do you think this is a distinction that most people working in this space draw, or is ant automation involving an llm referred to as an 'agent'?
Great point. You can build powerful automations by embedding AI nodes into the automation, but I wouldn't call those Agents. Agents are exactly like you described: they can plan and execute tasks, then continue to revise the plan and execute if the first attempt failed.
You hard core contradicted yourself @31:00. "think of agents as a job function... under each job function there's a bunch of workflows" and then in Miro processed to show how each individual workflow is a single agent with no idea of other agents/workflows. I'm interested to know if you've tried actually giving an agent all the workflows and then saying "decide what the best next action to do with abcdefg is" instead of you hard coding "when client is non-responsive for seven days, spin up AI 'bob' so he can run this exact workflow on this exact customer" - seems very prescriptive when you actually dig into and map the events that need to happen. It's basically only the content that's being generated and the rest are just a fixed trigger, a fixed response type (with a unique message), same with a calendar inv - just send them a calendar app link so they pick the time. I'd love to hear about some of your more complex workflows/agents?
Yes, you're spot on. I must've misspoken about the job function / workflow. We've found that Agents are more reliable with strong guardrails and limited instructions. Over time, the Agents will have better reasoning ability. All of the frameworks today that have multiple Agents design and execute plans together do not work very well. It's just not reliable enough for a business owner. We'll be able to pare down the amount of structured automation as the LLM's get better and Langchain improves. With that said, we do run more complex workflows and I'll be sure to share those here.
Absolutely. But that's the nature of application development. AI APIs are another component in our arsenal of development tools: it's just that the AI responses provide 'unique' results that we don't traditionally receive in other frameworks. It's those responses that have altered our semantics in building application solutions. Otherwise, yes, our workflow responses are still pretty much deterministic.
Hey! Great video. If you’re thinking like a project manager, is it possible to use AI agents? There are so many details and comms that happen in email, task comments, meeting transcripts/chat, etc. Would love your thoughts.
Ok, cool tools, alternatives, python , agents, LLMs, etc, etc. To what end a business will use this? For example, Why would a company like FedEx use agentic AI?
Idk what FedEx's workflows look like but there's no doubt in my mind that autonomous AI Agents would enhance their accuracy and productivity. I lay out a number of examples in this video and others, I'd check those out to get a good idea of the approach we're taking.
the video had a great practical overview of ai agents but i had a question : can you expand on , why you chose n8n over relevance ai ? for what reasons ?
@@customaistudio Thanks for your response, I am evaluating these tools and they both seem very good for agents, I would think that flowise has more langchain nodes. Why do you prefer n8n?
@@customaistudio I just want to be able to, say, make an Udio album of 10 songs on autopilot. So, for example, it would need to click buttons and upload files, etc. Can they all do all that already? I thought we weren't quite there yet?
That could be interesting but I genuinely believe AI Agents are going to kill SaaS entirely. Imagine just having a database (unstrucutred) that houses all data that would typically live in email, drive, CRM, project management software, ATS, etc. and building hundreds of Agents to perform any task a SaaS would do. Agents are the ultimate Software as a Service.
@@customaistudio Fair. Here's my 2 cents. There's a famous quote "People want a quarter inch hole, not a quarter inch drill." In my experience, people don't care what's under the hood. Whether it's a bunch of people in a call center or AI agents. The results are what sell - it's much more difficult to sell "AI Agents" to the masses. It's easier to sell "Get your research done in seconds" or "Get your paper summarized in seconds". This is why SaaS is still a big deal. If you're interested, I'd be happy to jump on a call and evaluate collaboration opportunities? I build SaaS for a living - and I've been learning n8n this past week to further improve my offerings. Neverthless, keep up the good work. Great videos so far. You will reach 100K soon.
"Through SmythOS, AI becomes a reflection of human thought, exposing the paradox where advancing machine intelligence also unveils the mysteries of our own consciousness."
I think of it like driverless cars. They said we’d have them 10 years ago (I didn’t even bother getting a driver’s license because of all the Singularity hype), but we still don’t. I believe the reason is that while the first 90% is relatively easy to solve with AI, the last 10% is extremely difficult. There are often edge cases where the car gets stuck, and you need a human who just ‘gets it.’ I think the same applies to coding. There are always edge cases when using AI to build software, and it would be incredibly frustrating if I couldn’t handle them myself instead of relying on AI. Frankly, AI is currently pretty useless at coding. While it’s improving every year, it still makes so many ridiculous mistakes that I see no reason why coding as a skill won’t be necessary for at least another 10-15 years.
@customaistudio do you really think it's possible to build full stack applications that are robust enough for enterprise grade workloads without knowing how to code? You think you can just send some prompts to agents, let them write code you cannot understand, and let it build infrastructure to support the application without understanding anything about IT architecture? Truly living in a fantasy world. Please let me know 1 real enterprise grade application in the world today that was built without software engineers at all.
Thanks! I've used Make a ton in the past but I don't think they have the AI features we're looking for. The AI nodes we use in N8N are specifically Langchain. Langchain is what allows us to build real Agents. However, you can stitch together AI nodes in Make and configure it to where it would act like an Agent. All you'd need to do is connect scenario's using a webhook (probably, I'm not sure how you do this make exactly). But let's say you have an AI node in Make that field incoming data, then is tasked with doing something with that data. You could build other scenario's and have the AI node trigger those scenario's (using a router) when needed. Hope that help!
You can use Make, activepieces, Latenode or N8n. There isn’t much differences with tools, you just have to understand the fundamentals of how they work - especially HTTP, Webhook, LLMs(learning about custom agents too) and API. With this you can basically build whatever you want.
Can you talk a little about the security framework with n8n? Are there any open vulnerabilities that might get exposed? Each action an agent takes might be compromised without auth protocols no?
i know this guy is legit becuase ... proper exploration of options, n8n and then talking about the grifty prompt people. respect.
Nice, thank you! There's so much grifty behavior in the AI space right now. Business owners just need clear answers and tools that work.
Also the moustache
@@markvandermeulen1612mustache == credibility
Respect. Exactly what made me follow. You can only talk like this from experience.
mate u told me more about application of genai than 99% of videos on youtube. You explained WHAT is an agent and WHY should we use them. The rest are just trying to explain a tool or an SDK or a transformer but if don't know what is the end goal of all this, its quite irrelevant. Thanks for the video, much needed.
Thanks man, there's not a ton of content out there that really show what Agents are capable of right now. I'd like to fill that void.
@@customaistudio thanks for sharing,it's really useful and you explain it very clear
I would love to see a video on how you built your personal assistant with n8n.
This was extremely valuable. It exudes intelligence, and I will definitely watch other videos from you. My takeaway is that there is a lot of work to be done by non-coders around thinking about what their workflows are and how they are enabled by different tools or subtasks and how to lay these tools / tasks down on a piece of paper and connect them. The one area that still prevents me from embarking on that journey is the lack of familiarity with a lot of the tools you mention in your video and the lack of time in my day to both do my job as well as take time to automate it. However, videos like this one show that it's possible and worth it in the long run.
As always, play long term games, get long term results
Thanks!
Bro has the smartest grasp of the whole AI direction. Bro also needs a better lamp
Yes bro does
I'm in the market for AI agents and did about 3 hrs of research yesterday. I tried a few free trials for a simple task.... I was so exasperated at the clunky architecture they pushed me into that also *didn't work* or only worked somewhat. N8n reminds me of an in-house data pipeline tool used at a past Big Tech co job which was adopted by hundreds of employees, it made things so easy. Hoping n8n is too. Thanks for the presentation. Awesome work.
n8n is the best tool I've found for creating Agents. Feel free to join our Discord is you want community help when building.
hands down one of the best videos that explained the AI Agent concept flawlessly. i will defiantly explore n8n and see what I can do with it.
Mr. Devin I would hug you if I could, for this video. I would love to see a series of videos based on this overview, especially an in-depth episode on the whole Architecture section. Cheers!
Damn its been a month and my perspective on the world has changed dramatically , I am all for AI entrepreneurship
The last 2 minutes wrapping up, are just Gold!!!. 41:02 I am agree at 100%. All SDR, BDR, CSM will be all replaced
Absolutely! It's hard to see roles like that sticking around. Even traditional SaaS is vulnerable hah.
Why do you see CSM being replaced though? I don't envision enterprise level clients that expect a white glove experience from their vendors being okay with working with an AI agent.
Amazing video. As a software engineer of over 10yrs, I know your life would’ve been way easier if you were an engineer. I think I will explore a lot of things around this over a weekend. And hopefully documents my experience like you did.
Yea I wish I stuck with it sometimes but luckily we have low code options.
With AI virtual agents, businesses can offer more dynamic and responsive customer interactions. These agents leverage advanced algorithms to understand and address customer needs effectively, leading to higher satisfaction and improved service
This is such an insanely useful video! Fair comparisons, useful info and backed with tips 👏👏
You guys are undoubtedly very busy and you share this incredibly helpful video! Human is still great ❤
00:03 18 months of building autonomous AI agents
02:10 The future of AI agents and their role in AI space
06:25 AI building platforms are focusing on AI agents
08:28 Inen platform is intuitive and feature-rich for building AI agents.
12:43 Using Pinecone for Vector database storage
14:51 Database and prompt engineering are foundational to building highly effective AI agents.
18:30 Providing context and detailed instructions to autonomous AI agents is crucial for their performance.
20:15 Examples are crucial for AI agent success
24:11 Demonstrating AI agent's email and calendar actions.
26:12 Overview of Email Actions Tool and Integrations in Building AI Agents
30:06 Building agents requires a change in integration thinking
31:58 Agents excel in complex workflows with reasoning and decision-making.
35:40 Automated lead qualification and enrichment process
37:34 AI agents evolving beyond chatbots and voice assistants
41:20 Leverage AI agents for business growth
Did you use AI to timestamp all of this? 😅
@@diogokamioka Yes
So refreshing. Someone whos approaching AI from a real world perspective.
loved this walkthrough and introduced me to n8n for the first time. Subscribed, thanks!
Awesome, thank you!
Baller. Definitely over my head at this point, but I’m loving it.
Wow! I predict this channel is going places.
Excited for you to follow along!
Definitely 😊
Love this. You should start a group where you teach this at price. Question: Can you provide guidance on a framework for prompting for an assistant/agent? You have a quite comprehensive prompt there.
Hoping to come out with courses soon. I have a custom Prompt GPT that uses our framework. Here's the link: chatgpt.com/g/g-Cehuz3C7p-custom-ai-studio-prompt-generator
@@customaistudiosuper helpful. I’d pay for an agent creation course today. Happy to discuss offline, if you’re open to it.
this just popped into my recommended and I can't wait to watch this later
Lmk what you think
Thanks God of Algorithm! I found a new Gem 💎 Good job. Subscribed. Definitely 👍
This right here is ABSOLUTE GOLD!
Thanks a lot Devin
Thanks for sharing this material! I’m also working on AI automation in Python-mine is custom, but what you’ve shown here is quite similar, especially with the user-friendly visual dashboard. It reminds me of how email platforms evolved about 10 years ago. The great news? Just 100k views on this video means there’s still room for us to learn and build, staying ahead of the curve! 😎
Tons of room!
Great explanation of agents and their application in the workspace. I am genuinely excited about this next iteration of application development. Sadly, AI is not cheap. You have highlighted that on the technical side, an agentic solution requires marshaling and preparation of numerous applications, interfaces and prompts - much as pre-AI solutions required. This takes time, expertise (even for non-coders) and digital resources, which all cost money to establish. The elephant in the room that NOBODY talks about is the cost of the prompting. Few organisations will have their own copy of a universal LLM like ChatGPT, Groq etc., so they will have an API contract with an LLM provider or broker instead. Iterative use of an API using a previous response as embellishment to the previous prompt (APIs don't have memory at the LLM source) makes round trip, lights-out enquiries quite expensive. (Just my $5-per-million-tokens-worth...)
This is all true. The bet is the costs will decline dramatically over time and these systems will be the dominant force for business ops.
@@customaistudio I'm sure the agentic semantic will be the dominant model to build business applications, but I'm not so sure there will be a dramatic price drop. Those LLM companies have got to claw back those invested billions somehow, and it won't be via big balloon payments for massive contracts. What to do then? Check out the Economist magazine (Aug 8th edition). In the Culture section, there is a review of one of the latest books on AI, "Co-Intelligence: Living and Working with AI" by Ethan Mollick; (btw, over 100 AI titles have been published in the US in the last 12 months). The book covers all things a layman needs to understand to interact successfully with a LLM (via Chat GPT, say) rather than considering the agentic approach, but the review ends with a final comment about an unexpected emergence in AI:
"...Amid AI hype in business, where companies say a lot but seem to do little, “Co-Intelligence” usefully notes that innovation is hard for organisations but easy for individuals. Hence, do not look for how AI will change business from chief executives’ statements but from ordinary worker-bees who quietly incorporate it into their everyday tasks. The revolution will be noticed only in hindsight."
Agentic solutions may be less ad hoc than the glorified chat sessions covered in the book, but their footprints will be modest and their impact profound (so long as we can really, really trust the LLM responses!).
A Docker container on a cloud service provider like AWS will let most businesses be able to run their custom LLMs in their own agent system, albeit for most meaning they have to rely on open source weights.
Very nice video and very helpful. Thank you, Davin. Clean and direct explanation. I would like to see a video that covers more platforms, frameworks, tools, databases, and data structures available and ready to implement for building agentic and autonomous systems, as well as your peer review on those. Great content!
Thank you! Our main focus is showing how SMB or agency owners can start to leverage AI Agents for their business. Reviewing platforms and other tools isn't really on the roadmap. But if we get more requests for it, then I'd happily talk about it!
I just finished my first AI project , i will further learn this Ai Agent , sounds interesting
Ngl i like this no code view. U up for a voice call? I would love that i come from a perpective of more code the better so to have a discussion would be lovely indeed ngl.
Feel free to book a call from the website! customaistudio.io
THANK YOU SO MUCH.. BIG RESPECT FOR ALL THE GENEROUS ADVICE.. GOD BLESS
Amazing video, Especially the way you have explained your journey. Even i have been building agents and pushing it to production mostly using crewai. From the frameworks persepctive, they are not goof for production, I have built my own abstraction over the openai assistants and function calling, and it is fast and more reliable than crewai. A lot of work needs to be done by there frameworks to be stable.
Exactly, crewai seems very fragile right now and even fixing bugs or troubleshooting issues is tough. Luckily, N8N offers no-code options for building Agents.
@@customaistudio they do push interesting updates, but the whole thing being written on python makes it "slow", plus i do not know why but normal openai assistant calls are much more reliable in terms of quality + speed
Im here, new, lost in life. and listening
18 months of building autonomous AI agents in just 42 minutes is impressive! SmythOS also focuses on enabling rapid development and collaboration among AI agents. Great work! #AI #SmythOS #Innovation
Outstanding Video. Normally Ive switched off within 5 mins, but I sat through the whole video. Very well explained and great ideas for what to use this for.
New Subber and Bell set.
Thank you
That's great to hear! Thanks for giving it a watch.
@@customaistudio I can see so many potential ideas for businesses or as a service for businesses.
@@customaistudio The ref to HR staff made me laugh, as my wife is an HR consultant and laughing when I tell her I have an Ai Agent to replace her lol You know exactly what she will reply that the Ai Agent cant do..lol
Hahah yep, exactly
Great content Devin. Maybe we can get AI based human face to these agents (like AI influencers), and deploy them to customer facing jobs/tasks.
That's already being done! Some really cool platforms for that are coming out now
This is super valuable. Thanks for sharing all these knowledges. Can you please share the prompt doc?
@@customaistudioalso requested
Here's the doc: docs.google.com/document/d/1voTacUiSv6nGVXHlNYEbp7u1SeWxYfOmgM3jswG7erk/edit
Use this doc: docs.google.com/document/d/1voTacUiSv6nGVXHlNYEbp7u1SeWxYfOmgM3jswG7erk/edit
The one I sent earlier is our internal one
🎯 Key points for quick navigation:
20:52 *📧 Emails categorized as work or high priority should use tools like the categorize tool and CRM for effective management.*
21:32 *🛠️ Tools are essential for AI agents to function autonomously, enabling actions like managing inboxes and replying to emails.*
22:13 *🔄 Custom workflows in tools like NadN empower agents to handle diverse scenarios within platforms like email inboxes.*
24:20 *📅 Agents can perform tasks like sending emails and scheduling meetings based on predefined workflows and tool integrations.*
28:03 *🔄 Integrations for AI agents involve configuring access scopes and data flows, treating agents like human employees needing specific platform accesses.*
30:34 *🏗️ Building effective AI agents involves structuring workflows under job functions, enabling agents to reason and execute complex tasks beyond simple automations.*
Made with HARPA AI
Love your videos! Keep creating them.
Will do!
Fantastic insight. I evaluated n8n in the past, but ended up going with Wappler instead. I want to integrate similar agents into it. I would love to see another video where you build the same functionality in Wappler.
Great content! Would love some more tutorials how you do things
Thanks! More is definitely coming 🫡
Hey man I would strongly recommend you to learn to code and understand AI/LLMs!
It would be great but my approach to AI is to build tools for business owners who don't know how to code. If you're looking for coders in the AI space, I'd check out @vrsen @AIJasonZ @IndyDevDan. They're my favorites.
@@customaistudio Of course but I just mean so you are able to understand the limitations and communicate with your customers who may be technical is all. Just my 2 cents.
I agree man, thanks for the recommendation. Are there any resources I should dig into?
awesome video, didn't know n8n added all these AI nodes and features! I've always enjoyed for workflow automation that involves data processing, but found it weak for other use cases due to lack of OOTB integrations.
The AI features are gold!
Same with thinking prompt engineering is a grift, but its essential.
Especially with Agents
Wow! That's is exactly what I was looking for as an entrepreneur in this ocean of videos about AI Agents. Thank you very much for sharing your experiences with us! 😊
Thanks man! That's exactly what we're going for
Came here to say exactly the same too. So much spam tier grift online lately but this cuts through. Another is @dougsilkstone but they do more general AI automation hackery
Greetings from Venezuela !! Thanks to your video I am not sleeping tonight 😂
Thanks for sharing mate, liked and subscribed! Also working on a AI project, but not yet with agents. Planning to integrate later.
Awesome man, thank you! What kind of project are you working on?
Awesome man, thank you! What kind of project are you working on?
Great video! Really enjoyed the content
Thank you!
Awesome overview, Devin! In what cases ARE you using multi-agent architecture where agents are conversing, giving critiques of each others’ output for improved quality?
For output quality, we attach separate AI nodes that will do proof-reading and editing. I don't like to add extra prompt instructions for the Agents on how to write emails or create content. I've found that the output isn't consistent.
This video is gold, anywhere we can get your prompting structure?
Links should be in description of my video about prompts
Say, that's a pretty nice system prompt! And I'm a snob about such.
This was amazing. Thanks. Have you thought about fine-tuning an LLM with QA pairs around scenarios/output you're looking for before plugging it into something like n8n to reduce number of tokens required and thus cost, or is it over-engineering for what would be marginal decreases in costs?
Good question. What we want to do is build Agents as fast as possible while still maintaining their effectiveness. At times, we'll do stuff that isn't the most cost efficient just for simplicity and speed sake. So although I haven't thought about your approach, I think you're right in that it could be over-engineering.
Always curious about new approaches though, if you've built something similar, feel free to send it over!
Thanks for the video ! You didn't mention LangGraph in your video, what do you think of the library ?
Awesome video! Keep it up man :)
When the student is ready, the teacher will appear. I've just finished watching all four videos and am eager to explore more of your free content. It's excellent!
Wow, thank you!
you are correct.
Thanks Devin. That's very helpful. I 've been working on various agents for legal vertical. Whats your take on specializing within certain industries? Or better generalize?
I think either approach works
Great video. Why are we so interested in no code solutions exactly? Just a question
I want to make Agents accessible to business owners who aren't also programmers. The goal is to show that anyone can build Agents.
Because no-code/light code solutions put the solution creation in the hands of the people with the primary business knowledge and move light to medium business process automation tasks off the dev backlog.
@@glennmontague4310 tell me in what situation a no code solution has worked?
super helpful thanks man
please make a discord communityaround all this
Good idea, I'll set up a sever soon!
oh thanks guy. it is a good content.
Thank you, excited for you to follow along!
This is some great stuff man, thank you!! Quick question, when you make ai agents for your clients do you do it in your n8n account or set one up for them? I've been using Zapier and Make but n8n's agentic approach looks spectacular. Again thank you for sharing this :)
Thank you! We have them create an N8N account and we build it in there. We can also easily share the workflows from our account to there's - which allows us to build out templates that we can reuse.
@@customaistudio Thanks brother! 🙏
Great video! Congratulations!
Awsome. Thanks for this
Thanks you so much. What you think about flowise?
I haven't worked with it a ton.
can you please cover what ex google chief Schmidt said about agentic workflow breakthroughs and please please tell us why Agent Zero is such a big deal. thanks.
Yes I will. Agent Zero is cool but still untested in production. It's also not as attainable for an average business owner. I'll touch on those updates and Eric's comments in coming videos for sure.
Great video! Thanks!
Thanks for watching man 🙏
Really informative video. Thanks! So I would draw a distinction between automations where a human structures the tasks the system performs vs an agent where the system defines/plans the actions it will take and then takes those action, evaluates the result, and may revise the set of actions, iterating until the desired result is achieved. This is not an important distinction until you are trying to assess the capabilities of models. My understanding is that llms can be successful within automation, but tend to fail in agentic systems. Do you think this is a distinction that most people working in this space draw, or is ant automation involving an llm referred to as an 'agent'?
Great point. You can build powerful automations by embedding AI nodes into the automation, but I wouldn't call those Agents. Agents are exactly like you described: they can plan and execute tasks, then continue to revise the plan and execute if the first attempt failed.
You hard core contradicted yourself @31:00. "think of agents as a job function... under each job function there's a bunch of workflows" and then in Miro processed to show how each individual workflow is a single agent with no idea of other agents/workflows.
I'm interested to know if you've tried actually giving an agent all the workflows and then saying "decide what the best next action to do with abcdefg is" instead of you hard coding "when client is non-responsive for seven days, spin up AI 'bob' so he can run this exact workflow on this exact customer" - seems very prescriptive when you actually dig into and map the events that need to happen. It's basically only the content that's being generated and the rest are just a fixed trigger, a fixed response type (with a unique message), same with a calendar inv - just send them a calendar app link so they pick the time.
I'd love to hear about some of your more complex workflows/agents?
Yes, you're spot on. I must've misspoken about the job function / workflow. We've found that Agents are more reliable with strong guardrails and limited instructions. Over time, the Agents will have better reasoning ability. All of the frameworks today that have multiple Agents design and execute plans together do not work very well. It's just not reliable enough for a business owner.
We'll be able to pare down the amount of structured automation as the LLM's get better and Langchain improves.
With that said, we do run more complex workflows and I'll be sure to share those here.
Absolutely. But that's the nature of application development. AI APIs are another component in our arsenal of development tools: it's just that the AI responses provide 'unique' results that we don't traditionally receive in other frameworks. It's those responses that have altered our semantics in building application solutions. Otherwise, yes, our workflow responses are still pretty much deterministic.
Hey! Great video. If you’re thinking like a project manager, is it possible to use AI agents? There are so many details and comms that happen in email, task comments, meeting transcripts/chat, etc. Would love your thoughts.
Yes, absolutely. A lot of project management is playing telephone with data. That's a prime use case for AI Agents.
Good stuff. nice presentation.
Thanks for watching
Great video. Thank you!
P.S. The buttons in your funnel are not working.
YT hasn't verified my account yet so the links in the description aren't clickable. Hopefully that changes soon..
SmythOS AI Agent streamlines content creation, helping brands maintain consistent and engaging posts across all social media channels.
Great vid thanks Devin 🤘🤘🍓
Ok, cool tools, alternatives, python , agents, LLMs, etc, etc. To what end a business will use this? For example, Why would a company like FedEx use agentic AI?
Idk what FedEx's workflows look like but there's no doubt in my mind that autonomous AI Agents would enhance their accuracy and productivity. I lay out a number of examples in this video and others, I'd check those out to get a good idea of the approach we're taking.
Devin great video but i couldnt help but think I was watching Cheech Marin
Hahahah
the video had a great practical overview of ai agents but i had a question :
can you expand on , why you chose n8n over relevance ai ? for what reasons ?
I find it more reliable. It's also better for creating tools for your Agents.
hey! great video.
What do you think is better for doing AI Agents, N8N or FLOWISE?
N8N
@@customaistudio Thanks for your response, I am evaluating these tools and they both seem very good for agents, I would think that flowise has more langchain nodes. Why do you prefer n8n?
It's easier to grasp visually and I can build my own tools within N8N.
Great video. Do you have a tutorial to build that entry level SDR agent?
Coming out with one soon
ROCKIN! TY💥🔥🙌
This is great, thanks!
Glad it was helpful!
Hi , it reminds me the phidata workflow,
Am I right ?
😊
Subscribed! What is your opinion of CheatLayer, MultiOn and HyperWrite?
All solid. Most of these tools all do the same thing, it's just a matter of which one you like the most and whether it works with your workflows.
@@customaistudio I just want to be able to, say, make an Udio album of 10 songs on autopilot. So, for example, it would need to click buttons and upload files, etc. Can they all do all that already? I thought we weren't quite there yet?
Maybe, it really depends on the exact workflow for this one.
A great video and love the detail and examples, thank you.
You're very welcome!
Just discovered your channel today. Have you considered creating a front end for these workflows and sell it as a SaaS?
That could be interesting but I genuinely believe AI Agents are going to kill SaaS entirely. Imagine just having a database (unstrucutred) that houses all data that would typically live in email, drive, CRM, project management software, ATS, etc. and building hundreds of Agents to perform any task a SaaS would do.
Agents are the ultimate Software as a Service.
@@customaistudio Fair. Here's my 2 cents. There's a famous quote "People want a quarter inch hole, not a quarter inch drill."
In my experience, people don't care what's under the hood. Whether it's a bunch of people in a call center or AI agents.
The results are what sell - it's much more difficult to sell "AI Agents" to the masses. It's easier to sell "Get your research done in seconds" or "Get your paper summarized in seconds". This is why SaaS is still a big deal.
If you're interested, I'd be happy to jump on a call and evaluate collaboration opportunities? I build SaaS for a living - and I've been learning n8n this past week to further improve my offerings.
Neverthless, keep up the good work. Great videos so far. You will reach 100K soon.
Feel free to book a call from the website!
Thanks!
thanks for watching
N8n great but still not as good as fully custom coded ai agents
Maybe
I just found a goldmine. new sub for you🎉
Awesome! Thank you!
Awesome man ! Can you able to make AI calling agents as well ? It would love to connect with you for the same.
Do let me know !
Hey man, we do calling agents as well. Feel free to book a call with us to discuss
Amazing content
Thank you!
Hey great content !
Btw can you give the prompt template file?
Thank you! Here's the doc with a more fleshed out example: docs.google.com/document/d/1voTacUiSv6nGVXHlNYEbp7u1SeWxYfOmgM3jswG7erk/edit?usp=sharing
Would it be possible to get access to that prompt engineering google doc?
Here you go: docs.google.com/document/d/1voTacUiSv6nGVXHlNYEbp7u1SeWxYfOmgM3jswG7erk/edit
"Through SmythOS, AI becomes a reflection of human thought, exposing the paradox where advancing machine intelligence also unveils the mysteries of our own consciousness."
My bro !! Best vid ive seen on ai agents , if you keep this up we all gonna be making lots of money 😂 subscribing please dont do me wrong 🎉
Don’t you ever feel like learning how to code, so you can make your own Actions?
(Great video by the way, thank you!)
Would be awesome but learning how to code seems kinda pointless now right...?
I think of it like driverless cars. They said we’d have them 10 years ago (I didn’t even bother getting a driver’s license because of all the Singularity hype), but we still don’t. I believe the reason is that while the first 90% is relatively easy to solve with AI, the last 10% is extremely difficult.
There are often edge cases where the car gets stuck, and you need a human who just ‘gets it.’
I think the same applies to coding. There are always edge cases when using AI to build software, and it would be incredibly frustrating if I couldn’t handle them myself instead of relying on AI. Frankly, AI is currently pretty useless at coding. While it’s improving every year, it still makes so many ridiculous mistakes that I see no reason why coding as a skill won’t be necessary for at least another 10-15 years.
@customaistudio do you really think it's possible to build full stack applications that are robust enough for enterprise grade workloads without knowing how to code? You think you can just send some prompts to agents, let them write code you cannot understand, and let it build infrastructure to support the application without understanding anything about IT architecture?
Truly living in a fantasy world. Please let me know 1 real enterprise grade application in the world today that was built without software engineers at all.
very useful
Johnson Linda Miller Brian Moore Mark
Hi! Great video! i'm a Make user. What do you think about it for agents?
Thanks! I've used Make a ton in the past but I don't think they have the AI features we're looking for. The AI nodes we use in N8N are specifically Langchain. Langchain is what allows us to build real Agents. However, you can stitch together AI nodes in Make and configure it to where it would act like an Agent.
All you'd need to do is connect scenario's using a webhook (probably, I'm not sure how you do this make exactly).
But let's say you have an AI node in Make that field incoming data, then is tasked with doing something with that data. You could build other scenario's and have the AI node trigger those scenario's (using a router) when needed.
Hope that help!
You can use Make, activepieces, Latenode or N8n. There isn’t much differences with tools, you just have to understand the fundamentals of how they work - especially HTTP, Webhook, LLMs(learning about custom agents too) and API. With this you can basically build whatever you want.
20:33
exactly
Could you make the template for this agent available?
Templates and tutorials coming out soon!
@@customaistudio I look forward to that memorable day
Can you talk a little about the security framework with n8n? Are there any open vulnerabilities that might get exposed? Each action an agent takes might be compromised without auth protocols no?
n8n let's you host your workflows on your own server, so you have that option.
N8n is free?
You get a free trial if you follow the link in description. After that, it's about $24/mo for the starter plan.