1. Scoping Out AI Agent Projects - SCAI017
Sloup
1 follower
16 Video Views·Sep 15, 2025
Nate Herk - AI Automation Society Plus\10. Client Builds (Real Projects)\1. Working With Clients\1. Scoping Out AI Agent Projects - SCAI017
Hey guys, so I wanted to hop in here and make a quick video about
scoping out AI agent builds. So this video, I guess what I'm going to talk about is less so
if you're trying to automate a specific process. So if you were trying to make an AI automation to
just strictly maybe manage your emails or manage your inbox, it's probably a little bit different
because it's not, it's a little more straightforward compared to when we're really trying to scope out
bringing on a new client and trying to integrate AI automations into their business in a way where
over time, it's just going to build upon and build upon and just get, get better and better. So,
you know, obviously agents are very powerful and a lot of people get stuck at the same point of
like, how can I actually make this practical? How can this actually be useful? Because yeah,
sure you can use AI to, you know, have a cool project, but it needs to obviously drive real
value, save real time, cut costs, make money. Otherwise what's the point? And otherwise you're
probably not going to be able to get someone to pay you X amount of money per month, X amount of
money per project to set something up. So going to be walking through sort of like the, the blueprint
in my head of when I'm integrating these agents and when I'm speaking to clients,
the way that we sort of go about it, because a lot of times they'll, you know, a lot of times
people will come to you with an idea and it's maybe not the optimal way, even though it's like
they want this build, they want this phase first. It makes sometimes more sense to set up
a nice foundation. So anyways, enough rambling. Let me just sort of talk about what's going on
here. So there's really three questions you need to think about, which are, you know,
why are we building these agents? How do they fit into the business process? So it's really also
important to understand the business as a whole, their operations, what they're doing.
And then from there, the most important question really is how can we make them scalable,
modular, and really just like think about the longterm of avoiding a headache, a big headache
down the road when we want to change something or we want to add more agents. So we got to get this
stuff right. So it all falls into place. And when I'm initially scoping out, you know, a build,
I would say probably over half the time, for sure, over 50% of the time spent on a build is usually
planning out whether it's on Miro board, like we're looking at here, or whether it's just in a
proposal, a Word doc. Oh, sorry about that. Planning out the actual build and planning
out different pieces of it. So I'll get into that in more detail, but obviously like AI is a huge
paradigm shift and we know it's the future. So how can we make it ready for the future? So there's
competitive tasks, stuff like that. So when I'm approaching the project, I don't jump straight
into any then I start building out the wire framing, I start building out the components,
the workflow, the mindset I try to have is I try to break down it, break everything down into
as small tasks as possible. So I'll show an example. And I'll use a real example. But
every process has multiple tasks within it. And that's really the best way to break down
some things that you don't get overwhelmed when you're trying to do something a complex build.
But breaking down every single small task, and then it really gets simple as far as, oh, okay,
well, I can go into an end and make a super quick email agent really quick, because I know that's
a super specific task. And then from there, you start to think about you have these blocks,
email, email, send email task, you've got a create calendar task. And then you just have
to connect these blocks. And you want to make them modular in the sense that
you wouldn't have to change this block. Now that we have this function, we can pull this
pretty much off the shelf and use this function wherever else we need to in other processes
without having to change the actual configuration of the workflow.
That way, it's not going to be a super, super custom, time consuming thing every single time,
which helps you out and also helps out. You know, if you're charging a client by the hour,
whatever, you're going to be able to save them some time because you've already got some of the
stuff already prepared. So like what we're looking at right here is probably not the best way to
think about it as far as you know, you've got your input, always, you've always got your output.
And then how do we get from input to output? And this system where you've got these three
different agents kind of working, you know, alongside each other, but not working together.
And then they just can come together at the end for the output is not what we want to do.
We want to have each task broken down so that agents are working with each other. Because like
we said, what happens if the client changes something? What happens if they want to,
you know, add stuff in the future? So this is sort of more of like the way that
you should be thinking about it. And obviously, these are very simple builds are very simple,
little wireframes. But you want to think about the input, you want to think about the output.
And then the key processes within each of them to see what is going on within each task,
and how agents can take an input, and then create something and then hand that to another agent and
then hand it to another agent. And it will basically really easy to drag and drop restructure
the order add something in between each step. So anyways, think of these agents as blocks. So as
you can see, we've got these three agent blocks here. And then you can obviously integrate them
seamlessly within each different systems. So let's break this down as with a real example. So
for a simple project of let's say it's not really a simple project, but a project where you want to
have some sort of content automated. Let's say if you're doing this for client,
there'd be an input, which would be, they would want to submit some sort of topic or,
you know, area, whatever they want to post about, they would put that input there.
And then you could have this first agent, which would be something like a research agent,
it's going to take the input, it will then query the AI model, it'll go out and grab, you know,
it'll use Tavoli or SERP API, or anything like that to search the internet, grab articles on
the topic, it will summarize the topic. And then it can save the summary of the initial research
it's done somewhere. And so that's the input here. And then this agent's output would be that summary,
then this agent would grab that summary, it would, you know, pull that blog, it would convert it to
a post, whatever, for whatever channel we're doing, let's say what's LinkedIn, which is kind
of, you know, the example I'm doing, because a current project I'm working on right now,
let's say it's doing LinkedIn, it would grab that, that initial blog post, it would, you know,
take the relevant products, relevant services, relevant market, relevant customer profile,
whatever it is. And then it would form a LinkedIn post based on all of that input it's getting.
And also, you could easily add in another agent in between these two. And that's the one that's
going to do research on the ICP, the ideal customer profile, and then it would feed in those outputs
to this agent. Then from there, you'd have the agent that maybe actually post this, and then
this agent would take the LinkedIn post, it could do things like, maybe it needs to grab some sort
of images, or maybe it needs to, I don't know, also do some sort of other posting on different
channels, it would take this and this agent's job would be to post. And then finally, you obviously
have the output, which is the finalized LinkedIn post, whether that's going to be automatic,
or whether you just have an output for a client where they can go in there and approve it,
and then send it off. So doing stuff like this, and seeing the process before you get into end
to end, it's going to help you spot those gaps or redundancies before you get in there and start
building things, because that's where the real complexities come up. A lot of times you're
scoping out a project, and it seems so straightforward. And then you get in there,
and then there's so much more complexities that pop up. And that doesn't even include,
this stuff doesn't even include the refining and the iterating process of testing and
reprompting and taking everything out of the vector store and putting it back in a different
way and testing again and asking more questions, which is also, so originally, I think I was
mistaken. I said over half is planning out the build, and then the rest is developing the build.
More realistically, it's like, maybe 40% planning out the build, 25% building the build,
and then what's left, 35% would be testing, getting feedback, iterating, all this kind of
stuff. So maybe that's a better breakdown. But anyways, so we've got this, the actual build,
but then the next thing we need to do is understand, let me zoom out a little bit,
understand where the actual projects and the initial project fit into a business process. So
you've got different operations, I just really, really high level, but obviously there's stuff
like, some people want stuff for their sales, some people want stuff internal facing, they want
project management, they want stuff agents for finance, and all of that eventually, you'll start
off in one place, but then once someone understands, oh my gosh, this agent is saving
me X amount of time per day, now I want this to be implemented in my customer success area of my
business or my client support, or yeah, customer support, wherever. That's when it's going to open
up the eyes and like, okay, well now we're going to start to slowly roll out more and more agents,
more and more frameworks to get implemented into the business as a whole, and that's how it really
starts to scale. So we need to understand those pillars from the beginning so that in the future,
when you want these new projects to be developed and to relate to each other, that they're already
ready to have a cohesive environment. So let's look at an example for, a lot of people have
been commenting on my YouTube videos about SDR agents, which would obviously spend a ton of time
prospecting, researching, summarizing, and then putting that data into some sort of CRM for
the salespeople. So these tasks are obviously tedious, it's a perfect use for AI, and it would
save a lot of time. So we'd have that workflow, and imagine building an agent, or sorry, we have
that job, and then within that job there's multiple different tasks, like researching is one task,
qualifying is one task, summarizing is one task, so you would have different agents obviously
doing all that stuff. And then you need to think about when you move on to another
area of the business, how can you have all of this relate back? So talking about
leveraging AI, obviously it's a tool, it's here to help you save time, reduce your cost,
scale your business, but you need to do it responsibly in the sense that you're breaking
down pillars into smaller workflows, and you're breaking down those workflows into smaller tasks,
identifying where AI can make the most impact, and then being able to build those modular
frameworks so that you can seamlessly plug in stuff later in order to scale it. So yeah.
Now focusing on the long-term sustainability aspect, I've touched on this a little bit, but
you really need to think about the ultimate goal is not just solving one of these
immediate automation tasks, and then it's like we're going our separate ways, it's to create
that model that's going to support future projects, because we're still in the early
adopter phase, and technology is advancing so quick, and it's going to get faster and faster,
so it's being able to build out that framework. So one thing that I always say when I'm talking
to clients is that there's no such thing as a finished product. Every system that you build
as your business builds is going to evolve, new tools, everything's going to need to change,
you're going to want more and more, so that's why we like to make sure it's ready to go from
the beginning of the partnership with a new client. Every agent that you're building,
if you can think about it as a growing ecosystem where instead of starting from scratch,
you're using existing agents, existing frameworks, and I don't think we're at the
point yet where AI can completely replace an employee, unless maybe a customer support,
something like that, but as far as when you start to build up, the idea is we're going to approach
it with the mindset of chipping away at repetitive tasks, saving a few minutes with this project a
day, saving a few minutes for this project today, so like with the example of the SDR we talked
about, and that stuff's obviously going to add up over time, and the most important thing is that
it's opening the eyes of, okay, well, maybe we initially wanted just one project, but now we
see this and we want to expand over into multiple projects, so the three key things here are really
just to build for sustainability. You need to think about being modular, being scalable.
You want to focus on incremental improvements, so each piece of the project is going to want to be,
you have to refine it, you'll have to add new projects, and you want these projects to be able
to relate to each other. Agents being able to talk to each other and understand the different inputs
and their specific roles and then talk to each other when there's failures is just so cool,
and that's something that really just, as it goes on, it makes them smarter, they get more
examples, they see the way that they need to act to correct themselves until they're able to move
on to the next part, and then just aiming for productivity, not perfection. Continuous
improvement, there's no such thing as a finished project. Once again, every agent is just a
stepping stone to something better, and the more you have initially, the easier it is to continue
to add the more on. So hopefully that gave a good grasp at the way, just my mindset of
approaching these projects, and when you're actually scoping out an individual build,
if you guys want more detail on that, it's really just about however your mind can visualize things.
You've got the input, you've got the output, and then how can you get there from A to B,
how can you get there by breaking down things into as small and manageable chunks as possible,
whether that has to be, initially you could have 10 workflows and then you realize, okay, well,
this process happens every time right here without much variation, we might as well combine these two
pieces. So that's where obviously it's very experimental when you have to try different
use cases and then you get a better sense of what works when, what doesn't work when.
But anyways, thanks for listening to the ramble once again. I hope that this was at least
somewhat valuable. Just tried to put my thoughts down on a page and put my thoughts on a mirror
board so you can just sort of see some insights on the way I approach project. But any questions
about this, please feel free to reach out. Love to talk about this kind of stuff with you guys. So
yeah, that's gonna be it for this one. So thanks, guys.
