The Future of Business: Leveraging Autonomous AI Agents
Heads up!
This summary and transcript were automatically generated using AI with the Free YouTube Transcript Summary Tool by LunaNotes.
Generate a summary for freeIf you found this summary useful, consider buying us a coffee. It would help us a lot!
Introduction
In the past 18 months, the landscape of autonomous AI agents has undergone a significant transformation. These AI agents have become more capable and versatile, leaving an indelible mark on business operations. As someone who has been actively exploring this realm, I believe that autonomous AI can drastically change the way we work and live. This article will explore the journey of building AI agents, the evolution of their capabilities, and why now is the opportune time for businesses to integrate them into their operations.
The Rise of Autonomous AI Agents
What Are AI Agents?
At their core, AI agents are large language models (LLMs) like ChatGPT, which can perform actions based on reasoning and decision-making. Unlike traditional AI chatbots that are limited to input-output interactions, an AI agent can understand context and execute tasks on behalf of a user. For example, if you receive an email asking you to schedule a meeting, while a chatbot might draft a response for you, an AI agent would take that a step further and actually schedule the meeting. This ability to act and make decisions is what sets autonomous AI agents apart.
The Importance of Data
One of the first lessons learned from building AI agents is that data is king. The effectiveness of an AI agent is directly proportional to the quality and context of the data it processes. Just like onboarding a human employee involves training and providing them with relevant information, AI agents require up-to-date data to perform optimally. Using a vector database, we store and manage this information efficiently, ensuring that our agents are always contextually aware.
Data Collection and Retrieval-Augmented Generation (RAG)
To maximize the efficiency of AI agents, it's crucial to implement robust data collection strategies. RAG refers to the technique of pulling relevant information from a vector database to enhance the agent's decision-making capabilities. To accomplish this, we’ve developed automated systems that keep our data up to date without manual intervention.
Building Effective AI Agents
The Role of Prompt Engineering
Initially, prompt engineering may have seemed like an overrated concept primarily pushed by marketers. However, working closely with AI agents revealed that crafting effective prompts is essential for their success. A well-structured prompt leads to accurate outcomes. An exemplary prompt includes:
- Objective: A clear definition of the agent's role.
- Context: Detailed information regarding the specific task or situation.
- Instructions: Step-by-step directives on achieving the desired outcome.
- Output Requirements: Expectations regarding the form of the output.
- Examples: Providing relevant scenarios boosts the agent's accuracy.
Tools and Workflows
To give an AI agent true agency, it must be equipped with tools to perform its tasks. Using platforms like N8N allows for the construction of complex workflows that enable the agent to navigate various processes seamlessly. These workflows serve as the operational framework for the agent, equipping it with capabilities akin to an employee's functions.
Integration and Architecture
Setting Up the Framework
When designing AI agents, it's essential to view them as human counterparts. This perspective involves understanding essential integrations and permissions necessary for the agents to perform effectively. Every task requires specific platform accesses, making this stage crucial for agent deployment.
A Cohesive Architecture
We've realized that successful AI deployment relies on a well-structured architecture. This framework comprises job functions, workflows, and tasks. For instance, a lead generation setup includes multiple agents, each with a specific workflow—like inbox management, lead qualification, lead enrichment, and nurturing—ensuring that every aspect of lead management is covered efficiently.
The Future of AI in Business
Why Now is the Time for AI Agents
Every technological breakthrough has centered around achieving leverage in business practices. The evolution of AI agents marks the next frontier in this journey, enabling businesses to streamline operations, reduce costs, and enhance profitability. As businesses face mounting pressure to perform, employing AI agents represents a timely and strategic decision.
Potential Benefits
The benefits of integrating autonomous AI agents include:
- Increased Efficiency: Automate routine tasks, allowing human employees to focus on strategic decisions.
- Cost-Effectiveness: Deploy AI agents at a fraction of the cost of hiring new staff.
- Enhanced Data Management: Utilize efficient data collection and management strategies.
- Scalability: Easily expand the capabilities of AI agents to meet the growing demands of the business.
- Leverage: Use AI agents to replace traditional roles, thus optimizing workforce deployment.
Conclusion
The integration of autonomous AI agents into business processes stands to revolutionize the way we operate. The lessons learned over the past 18 months underline the potential of these agents to serve not only as tools but as intelligent partners in driving productivity and enhancing decision-making. The future of business is undoubtedly intertwined with AI, making it imperative for organizations to consider this technology in their strategic planning.
okay so today we're going to cover the last 18 months of us building autonomous AI agents um it's been quite a ride uh
AI agents themselves have gotten so much better uh over the course of the last 18 months and we know because we've been U
playing with them messing with them in the space for a while um you know kneed deep really trying to figure out how how
to build them first and foremost and where exactly they fit into a business setting right uh if you don't know me or
or anything about us I'm not a coder I'm not an AI engineer um I obviously I don't know how to code but I do think
that AI is going to completely change the way we operate and build businesses um and then obviously for you know our
personal lives and whatnot AI will have a role there but um when it comes to business I I fully fully believe that AI
can be a massive point of Leverage um and really help us ultimately just make more money and build our businesses in a
in a way where you know we're decreasing margins we're decreasing the cost to gain leverage right um it costs a lot of
money to hire an additional employee but that additional employee gives you a lot of Leverage right you can delegate a lot
of tasks to or start new initiatives because you have this new employee um but it costs a lot of money to get that
leverage right so um we think AI has is that is that opportunity to give get a similar leverage that an employee would
um at a fraction of the cost like way less than what it would cost to get a human on board so this is just us over
the course of the 18 months stuff that we've worked on um and ultimately uh the learnings that we took from building a
agents and why we think now is the right time to start embedding agents into your business in general and so we've
obviously you know we've worked on chat Bots that was a chatbot um that we worked on a couple of different things
for a fractional cro client um more of like the chatbot esque thing we built our own applications AI powered SAS
applications obviously various AI agents and then just tons of agents and and automations in general so we've been
in the space um attacking it from many different angles trying to figure out where the future is going how do we how
do we build something that's not going to become obsolete uh in six months how do we get involved in the AI space and
our involvement isn't just going to be obsolete you know when open AI comes out with a new model um and we think that AI
agents are that answer and so what are AI agents right put very simply they're just llms I think like chat gbt that can
take actions right so an llm a large language model is what's powering chat gbt chat GPT is just kind of that the
basically do is just do input output right like let's say you received an email um from somebody and you want chat
gbt to write write you an email response to send back to them right so you would copy the email that they sent you put
into chat GPT say hey write me a response chat GPT will write a response and then you copy CH PT's response and
then put it back into uh you know Gmail and then hit send right so CHT is writing the email but it's not actually
sending the email for you an agent would actually write the email and send the email on your behalf right so the the
big unlock for AI in general is not just it having the brain of a human but being able to do things uh based on like its
reasoning and its decision-making uh you know and and its role inside of your organization or whatever and so that's
that's a huge unlock and I think a lot of people are aware of this and are trying to bring this into reality and so
what you're seeing are tons of AI agent Frameworks just being basically being launched like every single week and TMS
of no code platforms that allow you to build you know AI agents right um and so some of the popular
ones are crew AI this one's getting a ton of traction tons of people are are obsessed with it there's a lot of
downsides um one of the biggest downsides is that you have to know how to code right this is like a coding
framework that you use um but clearly tons of people are using it they have tons of uh what they call crws which is
just multiple different AIS kind of working together um yeah I mean crei is a popular one we we tried it obviously
but you have to know how to code so there's there's a huge learning curve autogen same thing you got to know how
to code um this one was created by Microsoft it's pretty popular it actually it actually works pretty well
the improvements that they're making on this are uh incredible and they make it a lot more intuitive for kind of like
your average person to hop in there and start messing around with agents but still um there's a lot of fragility it
feels like in the system things seem to be brittle and break uh when we use autogen so you know I don't use it much
at this point um and then zapier so zapier actually released what they're calling zapier Central which is like a
centralized platform to build essentially AI agents they're calling them assistants and I I honestly think
that zapier is um kind of like going in the right direction but on the on the wrong path like the way that they have
created their platform for you to build your own agents is not super intuitive and I don't think that they fully
understand and how a how like someone is is thinking about this how your average person is thinking about this you should
be thinking about agents as um like almost like low skill virtual assistant but like human at the end of the day you
should be thinking about your agent like that rather than like an automation that you kind of like embed an AI decision
maker into that automation somehow you know um that that's just to say like zapir is a place where you can build
agents I just don't like the approach that they taking to doing it um and then you have platforms like
voice flow and stack Ai and other um chatbot traditionally like chatbot Builders right like voice flow is
probably one of the most popular platforms for just building chat Bots right um and what they've started to add
is hey start building AI agents right don't just make it a chatbot let's add actions to it give it tools like give it
the ability to do things so what you're seeing is even the the chatbot building platforms are now leaning into this AI
agent space uh all the momentum is is moving towards this all of these um these companies and these startups who
are like okay we're going to build like this this cool AI SAS it's going to be a platform where anybody can build their
own AI application onto it all of them are realizing that agents are the next step and so they're obviously adding
these features into their tools um and then you have some other platforms like relevance AI is a really popular one
that's gaining a lot of traction I think it's some the um I think these guys are out of U Australia but they started out
as I think Vector Ai and so they were just like a vector database provider or they embedded data for Vector databases
something like that something in the AI space um but more to do with databases and then they they spun off and did uh
relevance AI I think it was a I think it was a hard pivot and I like their platform a lot because it's very
intuitive and it makes sense um it it's easy to understand how to build an agent and uh you know what the different
components are for the agent but um some downsides are it it's it breaks a lot um there are some things that are kind of
hard to understand like the flow and how you're supposed to build the flow exactly I mean these are new new
startups and things are changing all the time but um relevance is one of my favorites I just don't think it's the
best and they're they're very similar to uh mind Studio mind Studio was one of the first ones that we were messing with
um because it's a very simple like uh no code way of building very powerful AI applications and now they're really
pushing the the AI agent or AI assistant side of things so um or AI automations you know so my studio is a good one but
what we have landed on ultimately is using inate in and in8 in is very similar to zapier or make some of those
automation platforms it's a little bit more technical than those but it's it's been the best platform that we' used so
far because of the AI features that it has and so if this thing will load I'll just show you some of the
features that you can tap into it's very visual too like like the way that you can build the agents
visually just makes way more intuitive sense than some of these other platforms like for this one this is like a
personal agent that I use you have like obviously like the agent here like this is the component this is the agent right
The Entity and then we can give the agent multiple different things we can give give it all the tools that it needs
to do its job we can give it the database so it has a knowledge base of information so it can do its job
accurately right and then we give it some couple other things like I like to give it Wikipedia calculator just in
case I have oneoff questions where it needs those um but I could give it like a Google search so let let's look over
here if I go to Advanced AI you can see all of the features that they have for AI they have tons of templates for you
um but AI agent you can get an open AI um module basic llm chain question and answer chain summarization chain these
are all uh nodes for Lane chain in particular which is a really good um probably the main um open- Source like
agent Builder you know it's a it's a coding thing so there there's nothing we need to get into there but um all of
these are backed by Lang chain text classifier and other AI nodes so you can get in here and there's other uh things
that you can use for AI like vector stores that's a that's a popular one we use pine cone I'm going to get into that
later um but needless to say we use nadn because it is the most it is the easiest to understand intuitively about how it's
all working together and it's no code and it's super cheap and it allows you to host all of this on your own server
so the data security side of things are um very very strong with inen so that's a long a long uh look into um what
agents are but but clearly the entire industry is moving towards agents clearly people are super super excited
about it us included um and clearly the the potential is there the potential is there for this to be um generally role
changing so here are a couple things that we've learned the number one thing is data is king um this is almost like
like a I don't want to say a Trope or anything but it it's it's almost like been over said at this point where it's
like data is the new gold data is the new oil you know data is like this this Omni important thing and uh it's almost
gotten annoying at this point but it's so so true when you start messing with the agents it's very clear that they're
only as good as the data you give them and you need to ensure that the data that it has is fully up to date on the
context of what it's going on in your business so let's take this from kind of like an esoteric conceptual thing to
literally ai ai agents in your business doing work for you right if you were to hire a human the human gets onboarded
they get trained up uh they Shadow a little bit and then they go live right there's a multi-step process to getting
the data about your business and about what that person is supposed to do getting that data into their head and so
they can go and they have the agency to go and do exactly what they need to do right what you hired them for the same
thing is for agents we need to give it the data and we need to keep that data up to date so the agent can go it has
the contextual awareness to go and do the job that you want it to do right and so to do the data to do the data to
build the database and give the data to the agent we create a vector database and we use pine cone for that so let me
Vector databases you obviously saw an nadn they had a couple different options but they are the one that has so far
been the easiest to get up uh get to get up running right away um works like 100% of the time it it really doesn't break
and if it does break it's usually something that's my fault um and yeah it's just it's super cheap like it's
it's free basically up at up to like a point of uh usage but we haven't even reached that point yet and we've been
messing with these for months so pine cone has been our favorite we're not loyal to it obviously if there's if
there's a vector database provider that's um you know better or comes out that's better then we obviously switch
but we really like using pine cone for now super base also they're super popular one they offer um you know
vector embeddings vector stor but there it's a little bit more complicated to set up inside of a super base because
they're not designed specifically for Vector stores like pine cone is right and so when I go back to
n8n you can see pine cone Vector store right we're using pine cone to save all of the information
collection and rag so rag means retrieval augmented generation it's basically just um the function that
allows the agent to pull the information from the vector database right um data collection the reason why this is so
important is because you can build an agent you can uh build a vector database you can shove that Vector database uh
business you can extract that information put it all in the database cool a month later the database is out
of date the agent is um doing it's not contextually aware about what's going on in the business at the moment and so you
need to keep things up to date to pretty much the the minute right up to the minute about like this email was sent
that email gets saved right into the database calendar event was created the details of the calendar event get saved
right to the database new project new task um a new hire a new lead enter the system you know anything that's going on
inside of your business any information that's streaming in or out any interactions that are happening they
should be collected and saved into the database and so we build automations in order to do that right so database data
collection two extremely extremely extremely foundational components to building really really really highly
effective AI agents this is something that we knew intuitively at the beginning we knew conceptually at the
beginning that this would be important um but we didn't understand quite how important and at the time how to do it
right now we figured out how to do it and do it in a way where it's scalable sustainable keeps the database up to
date keeps the agents uh working effectively it keeps our hands off we don't have to get in and upload uh new
data to the database like every night like no all of this is automated and it works exactly how you expect it to work
right which is automated prompt engineering so this one this one's funny at the beginning of all
of this AI hype I wasn't super um high on prompt engineering I I kind of found it grifty in some ways like I was seeing
tons of ads of like get my pack of a thousand prompts you know for marketing and it's like yeah like that's kind of
cool but it just seemed kind of grifty like it was like okay you can't just write a prompt like just tell exactly
what you want right um over time working with these agents became clear that without a structured way of writing the
prompts the agents are not going to perform the way you want them to they could have all the tools um everything
could be set up properly but if the prompt The Prompt is not strong enough the agent's going to flounder and it
might work the the worst case scenario so you could have the agent like not work at all and then it's clear
I got to fix the prpt it's just never working it works zero% of the time I need to fix the promp or you can have
the agent work 100% of the time and you're like okay this is great the pr works okay awesome
um but then the worst case scenario is it works like maybe 60 to 70% of the time so you're testing it testing it
testing it it's working it's working it's working and then you run into like an edge case or a scenario that you
hadn't thought of and it breaks and you realize it's because the prompt didn't either account for that scenario or um
when you're setting up the instructions for how to solve the problem uh it it it trips over itself in some way right and
it's hard to explain that without being in there and and like having like a clear troubleshooting case of that
happening but um needless to say clearly clearly outlining your prompt is is um one of the most important pieces to
here and each of our prompts at least on the the system prompt level right we have an objective which is just a clear
overall objective for the agent so an example can be for an agent that we have managing our inbox Your Role is to
manage my inbox you must accurately categorize every email that I receive so it's super broad we're not telling it
how to do its job we're not telling it kind of any any real details about it right we're just telling it this is this
is your role and then we give it context this is where we get into some of the details about okay so I'm I'm managing
this inbox so what does that mean exactly like whose inbox am I managing so we give it like okay you're managing
you know the founders um inbox where you're you're managing like our CEO's inbox they receive a ton of emails daily
um it's it's a huge mix of work rated stuff personal correspondence newsletters spam Etc uh these are the
categories that we want you to organize them into these are the priority levels that you need to attach once you
categorize them and then here are the tools that you can use if you need to do something so if there's an email that
you need to reply to you can use this tool um to categorize the emails use this tool if you need to notify the user
that it's an important email use this tool if you need to access the CRM for some reason use this tool if you need to
if this was a lead that came in and uh they responding interested to one of our campaigns you need to kick this to the
lead qualification agent and it'll take it from there right so giving it context and then we give it clear instructions
give it detailed instructions on how to do its job so if you were to hire a person you would give that person
detailed instructions on how to do its job right it's like okay you're you're uh an SDR and doing cold calls this is
how you cold call this is the script this is how you save the information into the CRM this is how you book a
meeting literally this is the information that you put in the description of the uh the event that
you're creating right like if you if you ever worked for a really high functioning team you know that there are
very strict processes and instructions and Sops in place for how to do your job instructions and so we do the same with
the agents you also have output requirements um like once the agent agent has done the work what should it
output does it need to Output a Json package or does it need to Output anything at all right like for our inbox
agent our inbox agent doesn't actually have to Output anything whenever it needs to reply to an email or or do
anything like that it's sending it to a tool so the Tool's doing that so the the inbox agent is just simply categorizing
could have all of the other components in place and without a bunch of really really good examples uh it's likely that
your agent just isn't going to work um over the course of a long period of time after like a 100 tests it's probably
going to fail probably 90% of the time if you have the perfect prompt and no examples with examples it's like 99.9%
of the time it's not GNA wait it's probably going to fail no it's probably going to succeed 90% of the time without
examples but with examples it's going to succeed 99.9% of the time um yeah I think we got that straight but examples
are pretty much uh yeah they're the key to getting the results you want with examples the agents have trouble doing
exactly what you need to need them to do so in our example about the inbox uh manager what we want to show is here's a
new email that was received what do you do with that email right first step is you need to categorize it you need to
categorize it as work you need to categorize it as like high priority like in this one it's like we were're excited
to present a new project a new project proposal um we think this will be good for both of us here's the proposal let
us know what you think so it's important that's obviously an important email right categorizes work categorizes high
priority that I need to see it and so the tools that it used to do that were the categorize tool um the CRM tool to
extract any information that I might need and then like notifying me right the notifi user tool so we give it other
examples for other scenarios that might happen we don't need to do scenario but having plentiful plenty of uh examples
makes it work so much so much so much better okay so prompt engineering tools tools are what turns
it into an agent they're what gives it agency right um it doesn't become an agent until you give it tools because it
literally can't do anything you could tell it your your your job is to uh manage my inbox and reply to emails and
it's like okay cool I'll do that but if it can't actually reply to email by using a tool then um you know it's not
an agent doesn't have agencies so reason why we like to use inad is because another reason is because you
can build custom workflows and use those as tools and give them to the agent so uh you know a common like tool that
we use is a workflow that allows an agent to pretty much do whatever it needs to do inside of an email inbox
know reply to a thread whatever you whatever happens inside of an inbox like add a label whatever um we basically
created one kind of like workflow that can account for any kind of scenario that might happen in order for it to do
um the action that that it needs to do right so to put this simply let's say that um you know your agent has the
whenever it needs to send an email get get uh get an email get many emails uh delete an email Market email is read
because we can build workflows like that pretty complex tools or tools that uh you know that are for a massive scope or
for like one platform and uh use that workflow as a tool I don't know if I'm saying that clear but let me just kind
has email actions tool caler actions tool and update database tool technically this database right here is
like actually let's try this let's see see if here I'm just going to test it right
now let's see if this is going to work I'll pull it up on my phone let's say so what I'm just going
it automatically started working it's going to use my email actions because I said send an email to Andrew so it's
going to go and Trigger the email actions to send him an email and then what it should do is uh
actions okay so what it thought when I when I sent the message what it thought I sent here let me just go into here so
you guys can see it instead of me yapping away okay so I sent this I said send an email to Andrew Lewis and
schedule a meeting with him for tomorrow at two so what it's thinking based on what I said was um send him an email to
ask if he can meet at tomorrow at two um and that's we're going to get into this stuff a little bit later but there is
like a slight if you're if you're trying to interact directly with an agent there's like a slight like language um
just like a a little slight learning curve with how to talk to it so it understands exactly what you're talk
talking about um but ultimately you know you don't you don't really need to change how you talk so let's look at the
logs so what happened is the agent used the database here the pine cone database it used our window buff buffer memory so
this is just the memory from past conversations we hadn't we hadn't talked in this session so nothing pulled up
here it called the Open aai chat model which basically just tells it the various tools that we have so some of
this stuff is hardcoded into um you know the the agent itself so we don't actually go in here and like edit any of
this um and then it called the email actions tool and then the email actions tool went through and then it said
sent and it used the memory accordingly so probably a little bit hard to understand but you can see that like the
human said this and then the AI said this looks like they made a little bit of update here on this uh this
output so yeah those are tools um the agent basically sees the message that you're sending it decides what tools
you see this email ACS here I just want you to visual ize what exactly that is right like email actions tool okay
that's cool let's look okay so you can see it's like a workflow inside of naden so we can build our entire literally our
entire AI agent team inside of naden which is just awesome so this is that email actions tool I was talking about
it's pretty much like any action that you need to take inside of an email uh or inside of an inbox this entire like
workflow can handle that for you so that's what it did it just called this tool that we had built already here okay
that's all I wanted to show you Integrations integration pretty straightforward especially if you've um
uh worked with any automation platform um but the way to think about it is a little bit different than just
automation platform um you need to think about it in terms of at least from our experience right
doing this for 18 months from our experience the best way to think about it is think of the agent as a human
being like truly just think of it as as a human and think of it as like what does it need access to in in order to do
its job and not necessarily like um what is like the the perfect like automation workflow like this information gets
saved in this slot here and then it gets uh manipulated in this way and then we take that and we add it to there like
it's some kind of hard automation don't think about hard automations just think about like access to the platform and so
sometimes like with Google and Microsoft in order to use use their apis to get access to like inbox and calendar and
whatnot you need their apis right their apis are broken down into very particular Scopes same thing with if if
you were to create a slackbot right like there's a a scope that's like okay if you access this API the only thing
you're allowed to do is read a message right can't send a message you can't um reply to a message you can't can't
even delete the message all you can do is read right right and then there's a scope that says write and then there's
one that says delete and then there's one and so there's like dozens of Scopes within just like one platform dozens of
just actions within one platform right and what you need to do with your agent is is give them all of the actions
essentially right all of the actions that they might need usually it's um they're not going to need all of them
but usually we air on the side of safety by just giving them the ability to do any action that possible within your
business or within your platform right within your Tech stack um another way to think about Integrations is okay this
agent if they were a new employee which platforms would I sign them up for um in order for them to complete their their
job right like I have an existing Tech stack I have a CRM I obviously need like this new Str strr to have access to the
CRM right so same thing um and then I would also think about Integrations in terms of data right what data does the
agent actually need in order to complete its tasks what what what is happening on one of my Platforms in my tech stack
that triggers that should trigger the agent to go and do something else right so Integrations are obviously
straightforward I think this is probably um something that people are pretty used to working with especially if you're
doing automations um but there is a slight change in how you think about it when you're building agents as opposed
to just hard automations and that's something that uh we had to learn architecture this was the biggest
learning so far um and I think we unlocked it just in how we in the framework and how we like build our
agents and how we think about them um it was a huge unlock when we finally kind of like crystallized okay this is how
you do it okay and so like I'm I'm going to tell you now this is how you do it basically the way to think about it
workflows right so like an SD has to uh they have to Prospect and save that information Prospect and reach out so
they got to find people online um do a little bit bit of research on the people and then draft like a really good email
workflow that's following up you're following up with engaged leads people who are interested people that you might
have had call with in the past right following up that's one workflow finding finding uh you know content that they
might find relevant how do I craft like a really good followup message right that's one workflow another one is um
how do I like let's book a call right they're interested in booking a call let's qualify the lead and book the call
um that's another workflow so that's a couple different workflows within just one job function
of SDR right and then within those workflows are various tasks and when we're thinking about just hard
automations not agents just hard automations you're usually just automating a task right and then
occasionally if you're really good you can you could automate entire workflows by just automating a bunch of different
at automation are actually really good at just like automating um almost everything the issue is with hard
automation obviously is any edge cases any reasoning that might need to happen any decision making that might need to
happen anything where it's like it's not like clearly this or clearly that um it's not binary um anything like that
automations break so agents are really good at at um you know doing various different tasks that are all going to
add to a specific workflow using kind of like reasoning reason like and thinking behind it so how does this look
so let's actually let's look I have this I have this set up what am I doing trying to come up with it off the top of
entrepreneurs agency owners find themselves creating tons of content or maybe they're doing like ads or cold
email and they're getting some leads like into their ecosystem they're getting people on their email list um
fairly consistently but the issue is you know they're focus is on servicing their current clients trying to do sales calls
trying to create content and everything in between their focus isn't on going through their lead list um trying to
enrich those leads like do research on them send follow-up messages like there most people most Founders most business
owners are just creating content or doing cold email or doing ads and then hoping that the people who engage with
that stuff are just like booking a call right there and then they hop on a call and then you close them and then you got
a new client the reality is there's so many steps in between like them engaging with your
stuff at the beginning to them hopping on the call that you're missing out on so much revenue there's so much revenue
that could be extracted from your lead list that you're just ignoring um because you just don't have the
professional so let's talk about Legion in particular how it would look architecturally building
agents so in order to to do the workflow flow that I mentioned earlier for that one s strr you're probably going to need
four different agents at least that's how we've set it up and each agent tackles one workflow so it starts off
with the inbox management agent this one is obviously managing our entire inbox and so if any leads come in that are
responding interested in interested um and maybe they have some questions or they want to set up a call it's going to
kick that to the lead qualification agent right and so obviously this inbox agent has various different tools that
we give it so it can do its job but like let's say a league came in and it was interested and we kicked it to the
League qualification agent let's get to that one down here and so with the lead qualification
agent needs access to in order for it to do its job which its job is just to um respond to uh queries from the leads and
schedule appointments schedule appointments with qualified leads qualify them by ask asking the correct
questions you know uh so it needs access to the CRM obviously it needs to it needs email actions um and it needs
calendar it needs to see calendar availability and to book the calendar right so lead qualification that can
also be triggered uh via the DMS or something right anybody responds interested interested lead qualification
agent gets called lead enrichment so let's say uh you posted a lead magnet you require uh name and email in order
to access that lead magnet now that that lead is added to your CRM immediately the lead enrichment
agent would go and search Google find their LinkedIn scrape what information it can find their website scrape what
information it can uh summarize that information maybe uh assign some kind of quality value to the lead like unsure or
very qualified or not qualified at all or whatever um and then save that information into the CRM as well we can
make that lead enrichment agent um better and better I think over time lead nurturing agent so leads in
the system um maybe it's it's triggered to schedule uh to to go off like every morning it goes off it uh reads the CRM
the lead list it finds anybody who hasn't been responded to in maybe like seven days or hasn't received a touch
point in seven days and then it goes and obviously reads reads the information about the lead drafts a really good
email maybe it goes to your Google Drive and finds like a piece of content or a case study that might be relevant that
might res res with the lead and it sends that along right and nurtures them and so if we look at all of these
together we can see that we don't need an agent group chat we don't need like a centralized uh Commander this thing is
that all of these other Frameworks offer like crew aai or something like you don't need it you don't need a group
chat with your agents right you just need your agent to uh accomplish like one part of its job one workflow and to
be triggered off of like one or two events right so inbox management gets triggered off of a new email lead
nurturing is uh scheduled to go off and then lead enrichment just works every time a new lead is added to the system
there's no reason to connect all of these I mean the lead qualification one gets connected here but lead
qualification should technically also get connected to uh lead nurturing although there's no need because Lee
nurturing is not Fielding inbound emails so yeah that's why we don't have it but anyway you could see why like even me
just saying that you could see why like thinking about the architecture and how these agents are going to work together
how it's uh important that you do it correctly right especially if the goal is to build a team of agents to uh take
architecture um and the last thing is over the last 18 months it's been clear that this is the future um I've
AI is not just going to stay as a chatbot it's not just going to be the voice um in our ear it's not going to be
like our our little um helper that just like gives us information or instructions it's clear that that is not
it right it's not going to stay there what is clear again is that AI is going to start having agency it's going to
start actually doing things we're going to give it um you know arms and legs and fingers and toes and um extremities and
then and then we'll give it tools and we'll teach it how to use those tools and it'll learn more on its own and
it'll actually go out and do things right it'll essentially just be our little workers our Hive of workers that
are supporting us inside of our businesses and so the reason why to bring it back to why we like this for
business is because every piece of technological innovation has been to give us some kind of Leverage in Life or
in business especially in business this is the next major major step in that the next um Frontier if you if uh if you
will um when you're thinking about scaling a business it's not no longer going to be who do I need to hire do we
have enough in the budget to bring on more headcounts to increase head count it's now going to be can we build an AI
agent or can we build a team of AI agents to fulfill this job that we need fulfilled right oh man I really I really
think um I should be getting on podcasts right man do I need to hire somebody to go and reach out to a bunch of podcast
hosts and and book uh interviews or can can I just spin up a team of agents to just go be my podcast um you know team
to go and like schedule me on on various different podcasts right oh man I really wish I could um I really need to hire
person I mean maybe it depends what the role is but realistically you could just have agents like do that stuff for you
um you know oh man it would be really nice if if I just hired more like customer success people if we had more
project managers and it's like you don't necessarily need more project managers or customer success people um just give
each of your clients an agent that has access to just the information about what's going on with their project and
their business and then any common questions or FAQs or anything that's happening concerns documents that they
need whatever the agent can take care of that stuff but anything that's like real about their business or changes in
strategy or building relationships in general your customer success and product managers can do that right um
and all this is doing the entire point is just to give you massive leverage in your business decrease the margins that
it costs or decrease the cost that it takes to scale your business and increase your margins overall and use
that money to hire high value employees that can 100x your business uh as opposed to employees that are kind of
just you're hiring them just to like maintain what you're doing right like man I'm really tired of doing this let
me just delegate this to somebody else instead of that being like a person it's going to now be agents right I think
everybody's moving to this over the next uh over the next few years I'm G to start rambling about this um but if
you're interested in talking about it do book a call with us we're happy to talk um and if you do want to learn learn
more about kind of like more details about how we think about it and build our agents and some examples uh there's
a link in the description that should take you to a YouTube video that kind of dives into uh all of that stuff but