Introduction

So today I'm going to talk a little bit about some of the benefits of automation, how you can use simple AI agents as part of a NAN workflow to get significant efficiencies. And at the middle to back end of my presentation, I will do a little live demo of the tool. So I'm looking forward to showing you guys what NAN can do.

Speaker Background and Motivation

So thank you, Michael, for the extensive background. Just to kind of give you guys a little bit more of a flavor of me, I do have an extensive background in business. I've worked in finance strategy operations and marketing and I have an

MBA as well as an undergrad in business most of my professional experience was in the US but and I've lived across New York San Francisco and Houston and now I am here obviously and looking forward to being more involved in the Lisbon and European ecosystems one of the things I learned over the years is that I love

efficiency I love streamlining operations and making things work faster faster, better. And so that's one of the things that got me into AI and into automations. And that's where we are today.

Cool. So let's get into it.

Why Automation—and Why Now

So why automation and why now? So just quick show of hands, how many of you guys are familiar with automation? Okay.

That's how many of you are familiar with AI? Okay, okay, okay. We got it.

What AI Adds to Traditional Automation

So basically one thing that AI has given, automations have been around for a long time. Automations have been around for years and years and years in various different forms, various different tools.

But what's really different today is that AI give these machines context, right? AI and AI agents and autonomous AI agents

and all of this stuff makes it very easy for machines to understand what we're trying to do and help us to achieve those goals, right? And that's where automation has now become so much more powerful with the influx of AI and AI agents.

Smaller Teams, Bigger Output

And because of this, we can now use small teams to do a lot of different things, right?

You can build these autonomous AI agents that can run off and run your marketing campaigns, run your lead gen, send cold emails, generate your content, all sorts of things.

And so now you can do, one person can do the work of five. I really think there's going to be an influx of founders and entrepreneurs and people that are building things on their own because AI makes it so much easier for us to be able to do these things.

Low-Code/No-Code Tools Make Automation Accessible

And finally, tools nowadays have caught up, right? We have low -code, no -code automation platforms like NAN that we'll talk about in more detail. And the only thing that you need is to think clearly about your process.

process, if you can map out your process and lay it out in a way that's easy for someone else to understand, then you can basically automate that process. How many people have not heard of NAN? So I don't spend too much time here if everyone's already familiar. There's a few people that are new to NAN.

NAN Overview

We won't spend too much time on this, but just a quick overview, NAN is a tool that that can be used to basically drag and drop and create a visual workflow that can integrate a lot of tools and connect a lot of different tools together.

So you can connect Gmail to Google Sheets, to your CRM, to your Slack, to your Microsoft Outlook, if you have both Gmail and Microsoft Outlook, I guess. But basically it allows you to connect a lot of different tools together.

Other Automation Platforms (Make, Zapier, MindStudio, Gumloop)

It's not the only tool that can do this. A few other tools that do this, I've listed here, are Make, Zapier, Mind Studio, and Gum Loop. So these are all other tools that you can use that achieve similar functionality.

But today, for the sake of today's demo and for the sake of the presentation, we'll be using a tool called NAN. All right, cool.

Example Use Case: Email Triage Workflow

So the workflow that I'm going to show you guys in just a few minutes, it's not exactly this, but this is a version of it.

From Incoming Email to Classification

So this is an email triage workflow. flow. So basically you get an incoming email. This is assuming that you're a customer support department that gets emails, right? And so you would parse and classify those emails.

Human-in-the-Loop Review for Safety

You can look up the customer details in your customer database and then draft an AI response, alert the team that you've drafted this response through Slack. And then you can have a human review that message, right? Just to make sure, hey, is this okay before we send it to the client?

I always recommend a human in the loop for for these types of things just in case the ai decides to get a little creative one day and so i think it's always helpful to have that that human layer within the workflow this is not exactly what i'm going to show you today but this is an example of how you

can build an emo triage workflow what we will see is actually something that i built that's a little bit simpler than this but that you can connect to a personal email box to organize your personal no email, for example. So let's get into it.

Live Demo: Automating a Personal Inbox

Do I have any brave volunteers? I only need one or two. Any brave volunteers?

Sure, okay. Would you be willing to send me an email at this email address here? You can say anything you want to, but to make it a little bit more interesting, maybe you can say something like, I'm attending the MindStone AI meetup, will you be there or

something i'm curious how the ai will respond to that but um do you still need this on the screen or do you uh yeah one second okay great so did you send the email yeah go ahead and you can go ahead and send it uh any subject yeah that's fine all right okay so do you guys you guys can see

this right it's it's okay okay so uh the email was sent we're gonna give it just a minute or two for it to uh capture because it takes sometimes it takes a little time for it to capture but um this

Workflow Walkthrough: Trigger, Classification, Logging, and Drafting

is basically the example workflow i'll go ahead and talk you through this while we're waiting for the email to come through so um so basically when a new email is received when this workflow is like it's not live at the moment i just have it in a test environment but when it's live live, you'll see up here it'll be green, there'll be a little green light that says it's published,

and then it'll just automatically run every time an email comes in and do exactly the actions here. I don't have it live because I don't need it live at the moment, but if we have time at the end I'll publish it and you guys can send me emails and see how it responds.

Okay, so basically this trigger here receives the email, and then we extract email fields, and then this is the AI agent that then classifies the email, right? Like what type of email is this, where should it go?

And then it'll log, all emails are going to be logged into this email logger sheet that I have pulled up here, and then it'll map the category and then apply a category to the email.

So the categories are I just put some general kind of like finances for receipts Notifications if you're getting something from an online account something that needs responding to which will probably be this one and then spam is one

as a filter and then and then this this will check if basically if it's an email that needs a response it'll It's an if statement that will then create a draft reply if a reply is needed. So that's basically the workflow workflow.

So okay, that should be plenty of time. So I'm just going to click execute step. We're going to see if it captured the email.

Hello. Okay. It has.

So you can see here, I don't know if you guys can see this clearly, but basically the subject is AI event Lisbon. It says, hello, I'm attending AI mindset meetup in Lisbon. Are you there?

Okay. So the email has come through. So now I'm going to pin this data so that we know that it's this email that we want to process.

I'm just going to click execute workflow and what it's going to do is it's going to run run through, and it's going to read and classify this email. And oh, it's done it pretty quickly.

And then it's gone through all the steps, and now it's logging it to the Gmail. And now the workflow has completed successfully.

Demo Results: Logged Email, Priority, and Draft Reply

So if we go over to the email logger, you'll see, oh, there it is. It's showed up here.

I've gotten an email from Magdalena, and it needs response. The priority is medium.

And it says that Magdalena is inquiring if I'm going to attend the AI MindStone Meetup in Lisbon. And then it says, and then it's drafted a reply for me.

Obviously, this drafting the reply part, you will need to do a little bit of prompt engineering and make sure that it's replying in the way that you want it to.

Clearly, this has required, yes, I will be attending. It did not say I will be speaking, but it did say I will be attending. So this is an example of how you can kind of set up this workflow and how it would work.

Then I'm going to go ahead and open that Outlook message for you, and you can see here, this is the email that we just received right now, and these are some test emails I sent before. It's tagged it as needs response, and if I click in, it's drafted this email and says,

draft, this message hasn't been sent, but it's drafted an email for me. This is how you can see something like this working.

Q&A

I'm going to take a quick pause there and see if anyone has any questions, any clarification

verification needed? Yes.

What “AI” Means Here: LLMs, Agents, and Context

It's using AI. It's using, I don't think it's, it's using an AI agent within the workflow. I don't know if that's officially machine learning or, but it's using an AI on the backend. So it's using Anthropic in this case.

Text mining? It's a large language It's a large language model. Yeah, it's a large language model. It's an LLM on the back end. I'm using Anthropic, but same thing. Same thing, yeah.

Extending the Workflow with Calendar Checks

So you can. You can incorporate your calendar and have it check your calendar and things like that.

In this case, I think it just defaults to being very positive and giving affirmative responses because I asked it a question yesterday just to test it. I said, do you want to hang out? And then it replied, yes, I would love to hang out with you on Friday. So I think it's, in this case, because I didn't give it a lot of instruction in the prompt, it's just defaulting to positive.

But those are things that you can connect your email calendar, for example, or your calendar, for example, and then have it check your calendar to see if you're available at certain times if you wanted to, yeah.

Choosing a Platform: Flexibility and Tradeoffs

Yes? They all have their own unique quirks and what makes one of them better and worse in certain circumstances.

So nowadays all Zapier make, and at least Mind Studio as well and Gumloop, I think all have some version of AI agents that you can incorporate like this into a workflow. So you could, in theory, use any of the tools. It's a matter of which ones you like and gravitate towards, and some of them are a little bit more flexible than others.

In my experience, NAN is one of the more flexible ones in terms of allowing you to do more creative things within the workflow. That being said, each of them have their own kind of unique benefits, And so I would say you could experiment with any of them. Some of them are better at certain things than others. So it depends on your use case.

Credentials and Connections

Do they use your login credentials, and what they're connected to? Yes, yes. So I can show you.

So basically, it is connected. So if you look here, on the very top here, it says credential, and I've named it my credential. So it is reading my actual inbox for that email.

And then the AI agent, oh, oh, oh, I thought I connected Anthropiq, I did connect GPT. Oh, okay. I thought I connected Anthropiq. Okay, so it's GPT. That's why it's defaulting to positive. Okay, now we know. So we learn something new every day.

But yeah, so that's why, and I'm using GPT 4 .0 mini in here, but you can use any model. Normally, I use Anthropic in my workflows, but this one is ChatGPT, which does explain its positivity.

And then the Google Sheets, you'll also need to connect the Google Sheets to connections. So there are some connections that you need to make to make sure that the workflow works as expected.

There's one in the back, and then I'll come back to you, for an actual client use case Yeah, and then we had a question up here.

Scale and Limits: Executions, Cloud Plans, and Self-Hosting

Instead of I think it depends on on your use case. So if it's high in Computational processing like you're throwing like large context windows like you're throwing 30 page PDFs And you know then I think you need something a little bit more powerful than this because I don't think this is great at processing

huge huge volumes of data, but if you're throwing a small amount of data at it, but Multiple times, you know like thousands of iterations or whatever but small chunks of data each time it works

but on the cloud version you'll be limited to the number of executions you have per month so if you're on the cheapest cloud plan

I think the execution limit is 2 ,500 per month which is still a pretty significant number so I don't know what automations you want to run

but I think that should be sufficient for most of your needs but if you do want to get fancier and get bigger plans and stuff

you can self -host with NAN which is one of the benefits of NAN over other tools and then you can run as many executions as you want as long as you don't exceed your own server, more or less.

Templates, Reuse, and Safety When Sharing Workflows

I see there's templates there. Are there workflows and templates to help you out with a lot of things? Yes.

You can find it, like there's resources? Yeah, there are resources. I'm just going to search NAN templates.

So there are 8 ,925 workflow automation templates available on NAN at the moment. and not all of them are free, but a lot of them are free,

and some of them, the creator of the template will say like pay 20 euros or 20 euros a month or whatever to use this, so some of them do have some payment gates, but there are a lot

of workflow templates that are just readily available that you can kind of look up and use.

I don't think there are risks in using the workflows because the way that the nodes are built, if it's a Google Sheet node, it connects next to your credentials, right?

And then it'll pass through your, so when you download one of these templates, it's not gonna have like their credentials in it, right?

Like unless you like hard code an API key into the, somewhere into the box, which you shouldn't do, that's not best practice. But as long as you don't do that, then, you know,

if you connect through the official credentials at the top, those credentials do not get shared. When you share, if I were to download this workflow and give it to you right now,

you would not get my credentials. But could embedding in the workflow, could it be sending something out to it?

I mean, I guess if you don't understand fully the workflow, there could be some things embedded in there that don't act the way that you intend.

But it's visible, so you can just check each node and verify what it's doing, and if you don't want something or you want to change the settings of something, you should be able to make those changes.

Conclusion

All right. Okay.

Well, thank you much. All right.