Hi guys, ~ this is Jomar, and I'm going to be talking to you guys about designing

high impact AI workflows for go-to-market systems.

I currently ~ consult for Bowtie Funnel and I created a Bowtie Funnel lab,

and that basically will help ~ myself and other people show,

like, okay, this is what I'm building currently.

And I think it's never been a better time to create labs.

So a little bit about myself. My name is Jomar Ebalida I am,

~ I do go to market revenue operations architect and

I'm a systems consultant for AI.

I also created a book called Dare to Orchestrate.

~ My real career evolution kind of started with events and then it went

to marketing operations, then it went to revenue technology,

then it went to enterprise rev ops,

and then I also had an engineering approach,

which I basically ended up working for as a principal systems engineer

in IT department.

So I was able to bring a lot of my bowtie funnel customer lifecycle experience into

IT. So I got to really understand like system design documents,

maintenance documentation, understand like how does Okta relate

to someone's user profile and kind of like the whole evolution of like,

okay, I understand from the business perspective.

And now I was able to understand from an IT marketing perspective.

And I also have a lot of experience working with startups.

And then also consulting for Marcat,

~ I did Marceto consulting for a third-party agency.

And I also worked at a digital transformation company.

And then I started working for enterprise companies for where

I helped transform from a technical perspective,

like a digital transformation and business perspective,

their ~ tech stack for revenue.

~ okay, so right now there's really three different types of

Agentic

solution like categories, which is like personal assistant,

co pilot, and governance. So personal assistant,

it's getting really popular, especially with open claw,

Hermes. Those are very popular,

personal assistants. And then I know Gemini,

Google's definitely coming up with more of a personal assistant as well.

Copilot is more like hey, you are creating automation with ~

Cloud co-work. And that has a lot of human in the loop criteria,

which is very much more of a like co-pilot,

like you're right next to the pilot kind of perspective.

And then there's governance, which is more of what we're gonna be talking more about

today. And if you look at, you know,

the aviation on industry, they do have control towers orchestration,

and that is kind of where the evolution is going.

I see a lot of influencers selling personal assistant like Hermes and OpenClaw.

To small businesses and large businesses.

And it's okay if you're like a solo ump solopreneur.

However, that kind of breaks when you have multiple individuals,

especially if you have a team and you're trying,

and so that kind of confuses me because I've seen it in the field in the wild.

It does not scale very well. So I always p tell people hey,

be very, very cautious about the personal assistance.

And going into an actual organization,

especially when they're very autonomous,

there's no human in the loop. It's a wild,

wild west and it gets into your system.

It's very scary, actually. So I highly do not recommend that.

For personal assistance, like,

hey, if you're putting it into like shopping for Nordstrom,

if you are doing very like travel a trip,

you know, organize your random budgeting for your friend's vacation.

That's great. I would highly recommend it.

And co-pilots like your regular day-to-day activities.

So today I'm also going to talk about like AI influencers just

as versus AI practitioners. There is a difference.

So be careful when you go on Instagram before TikTok and all of those.

There is a large group of people trying to really commodify

the attention that you're that they're trying to get.

So they're trying to hype up GitHub rebos.

They're trying to

give you little quick splurts without actually showing you the security concerns,

without showing you like the data governance issues.

There's just a lot of influencers trying to get rich really quickly with this

AI hype. If you're looking at like cryptocurrency,

they had NFTs back in the day and they were trying to hype that up.

So just be very, very cautious of who you follow and make sure

you take everything with a gr with a tiny,

tiny grain of salt.

And

then there's a practitioner's reality.

That's really where my reality happens.

If in most of the people that is going to be on this webinar,

this is the reality that you're going to face,

right? There's going to be, I've down like for fun,

and in my lab, in my ~ GitHub lab that I've been creating a bunch of random agents,

I've been downloading. Don't get me wrong,

I've on my free time and especially on the weekends,

I've been downloading GitHub repos from these influencers,

trying to see if it's like, okay,

is there any sense of

correctness in this. It was not correct.

There's a lot of errors and a lot of so you just take everything with

a grain of salt. I would not put that into production if you haven't tested

it in a sandbox using a sample set of your data set.

And then there's a lot of build versus buy.

So ~ I'm just gonna quickly go over a lot of the context back background

of where I'm coming from. So when not to buy,

right? When when if if you're basically gathering a lot of data sets

and you're having to maintain a lot of

random data or if you're having to maintain like workflow operations,

like that's a lot of actual manpower.

When there's a certain type of maintenance manpower needed to function that system,

I would just go ahead and buy that technology.

And so I cause you don't want to do that maintaining.

You just need it part of your AI agentic stack.

Just like there's a revenue technology stack,

there's going to be an agentic AI stack.

And so I'll show you guys later what my stack is.

for like small businesses or for people that you might want to,

you know, bring on. Of course,

enterprise is going to have a little bit more rigorous technology stack

for their agentic building. ~ so when to build,

I rec I recommend if it's like more deterministic approaches.

And so we're gonna cover a lot of those things like what am I building?

What type of tools am I like just using?

And right now I feel like there's a really interesting

Pricing

that people are not really pricing for users anymore.

They're really pricing for like outcomes and pricing for tokens usage.

So that's very interesting. So there's another thing that's come

to happen is self-hosted versus cloud.

So when self-house happened, is ~ at first people were trying to go from,

you know, self-hosted like technology and then everyone was trying to go to cloud.

Now it's gonna be the opposite.

Like I feel like now it's kind of going into a more hybrid approach where

Yes, you are also on the cloud,

but you are host self-hosting a lot of your LLM providers.

You're also self-hosting a lot of like if you have free LLM models,

you're also self-hosting those.

And so my thing is I'm mapping agents to the bowtie fun funnel and the AI maturity.

And so what that means is that I basically found

out that I did not.

like how Hermes or the other type of agents were basically using

~ like roles as an example.

They're using roles and ways to segment their agents.

What I found out after like trying that approach is that it actually closed

out contacts needed for the rest of the customer life cycle to function correctly.

And so

That was like a very drastic part that I found to be an issue.

And that's why I moved a lot of my build into more

of the customer lifecycle. And for me,

I use the bowtie funnel. That's why the lab is called bowtie funnel.

And I break it down by AI AARR maturity.

So I don't just go ahead and ask the client,

hey, what do you guys need?

Cause I I get a lot of I see a lot of influencers like just go ask

the company what they need a genetic solution for or workflows that need

to be built out. That's great.

If you if you have like a f like 300 million ARR and up and you are,

you know, you're sufficing that business,

you are already making, you know,

over 300 million or 500 million,

that's amazing. ~ then you could go ahead and select,

you know, what is the best, you know

type

of agentic flow that you should be building.

But for me, I really look at the ARR stage.

If you are C to 1 million, I would focus on like personal agent.

That's when you can use a Hermes or you can use a basic co-pilot.

~ But if you have like one to 10 million ARR,

you're looking at a little bit more sophisticated type of agentic AI solutions.

You're looking at, you know, how are you enriching these leads?

Are you creating battle cards for these leads?

Your lead to speed type of workflows need to be a little bit more built

out and sharpened. And then if you're looking at 10 million ARR,

you're looking at more like you know,

continuous like pipeline monitoring.

You're looking at ~ seeing if there's an upsell,

cross-sell opportunity. You're really looking at that whole bowtie funnel.

So, bowtie funnel, if you guys don't know,

it's it's basically a breakdown of

The different types of because before it used to be just be a regular funnel,

right? It would end in handshake.

For this one, it's basically awareness,

education, selection, mutual commit,

onboarding, retention, and expansion.

And so it was really popularized by winning by design.

But before that, they had ARG too,

which also has more of a bowtie funnel shape as well.

And then, so it's there's a lot of iterations to this,

but for me, how I plan my agents is I map them accordingly to the bowtie funnel.

Which is a customer lifecycle.

And I also merge the ARR maturity stage that they're currently at.

And that has a higher EBITDA score.

If you guys don't know what EBITDA score is,

is basically what type of operations will the company be more successful for.

~ and that's really important for the business as well.

And then so there's a lot of also roles that are coming out within,

you know.

This

type of realm because you know,

I know Clay made GTM engineer as like very popular,

but however, you know, now it's starting to evolve a little bit more.

And so the GTM engineer is popular,

but what I'm seeing a lot of people starting to figure out now is that when

you are a GTM engineer or you're a growth engineer,

or you're a forward deployment engineer,

or a Rev workflow engineer, they're all very similar words actually.

And their primary objective is to

you know, make money at the end of the day for the whole company and

to make sure their AI is operationalized and it's efficient for AI.

And so for me, what I am have seeing a lot too is that a lot of people don't

do discovery really well. And I think that's where peop consultants do really,

really well for companies is like if they are being if they understand

the business and they're doing discovery,

they understand like, hey, this is the customer life cycle.

This is the different type of agents that go inside the customer lifecycle.

~ and so I broke this down into the growth engineer,

growth engineer. And here's the primary objective requires for that.

I really, really gear towards that framework because the bowtie funnel lifecycle

is the macro framework. And then the agentic workflows is going to

be a micro framework to build things out.

And that's where you have context engineer,

harness engineering.

And so it's these roles are all very much similar.

I what I'm seeing right now is more companies are starting to like hire more forward

deployment engineers that will work with a go-to-market engineer.

And the forward deployment engineers are kind of spread

out throughout different departments.

And they're kind of very much juniors from my perspective that I've seen

in the wild. when I say the wild,

just like clients I've talked to or executives I talk to.

So yep, this is like the

Different type of GTM roles that are out there right now and that I'm seeing.

There could be more. I don't know,

but these are the ones I've seen.

~ so today I I've really focused on neurosymbolic architecture.

And if you guys don't know what neurosymbolic architecture for AI,

~ back in 2025 in December,

they actually met up and they said that,

you know, neurosymbolic agentic pattern building is actually.

the

gold standard for enterprise. And for a while,

most companies haven't don't really understand how to build those.

I know there's a few companies,

I know Orby and some other companies that really like situated their entire company

to have neurosymbolic agentic patterns as their pattern of choice.

And for a while there wasn't technology that can,

you know, that was sophisticated enough to build these types.

Of neurosymbolic patterns, but now there is.

And so ~ what I'm going to be breaking down is like different types

of agentic solutions that you guys can build.

And the reason why neurosymbolic is so popular for enterprise is that it doesn't,

it has a very low hallucinations.

It has auditable workflows, it has real world outcomes.

And and basically you could answer like,

what is it doing? What is it gonna do later on?

And so back in like

Before October of 2025, I was using NAN and it was like spaghetti slop on wheels.

It was just like I was getting a migraine looking at all the little arrows

and in like lines. So I also switched to TypeScript solutioning.

So we could talk about that later.

But for neurosymbolic, it's really broken down from the first part,

which is neural, and the second part,

which is deterministic, symbolic part,

and the last part, which is neural.

So it's an 80-20 split, but what you do is you

Put the first part as neural consuming the data and then being able to be like,

okay, what is, you know, how do I organize this?

And then the deterministic part,

which is 80% of most of the workflow solution,

those are going to be like these are the step-by-step solution.

And then the last mile is like,

hey, I'm gonna either ask for a human in the loop for that as a,

hey, does this meet my criteria of output?

And then yeah, put that in a business ledger at the very end and then

That's gonna be part of your source of truth for that one workflow.

And I believe for me, how I've been ~ great creating these is I've

had a business ledger for each AI workflow.

So what that means is that every time a workflow runs,

I will be taking that information of like,

hey, this is what the neural happened,

this is what the symbolic part happened,

this is what this neural part happened too.

And making sure that I have that business ledger,

because that business ledger is gonna be very important for most companies,

~ like business intelligence in the few in the future.

Like you could swap out these LLM models very,

very easy. And certain LLM models are actually going to perform better

for certain tasks than others.

However, that business ledger of that bit and also like that skill

set for that workflow is going to be very important because when

a company purchases you or you go buy a different technology.

Or an or you know, you buy a different company if you're a cup you're

a bigger company and you buy a smaller company.

I think in the future they're gonna be like,

hey, what type of AI workflows have you done?

What are the runs? And can I see the business ledger of that AI workflow

so I can see how effective it is?

Like what is the confidence rate of this workflow?

I think that's gonna be in the future.

And I know that cryptocurrencies has that business l has a certain type

of ledger for the wallet, and I can see that truly happening with AI as well.

And

so there's like five components to an AI architecture for execution.

There's the LM neural part, the concept construction,

the tools and actions, the control and cognition loop,

and then the safety alignment and evaluation.

And then there's a cole core exome,

which is more of like the conservative part,

making sure that you know there's safety gates in place.

~ is this looking at hallucinations before even sense of the LLM kind of thing.

And there's certain types of like GitHub repos that are free that

are actually pretty good at doing that.

~ but there is a lot of companies like Langchain and other companies that

for enterprise that do do that as well.

So there are it is very like I went in this rabbit hole and I'm coming

out of the rabbit hole thinking,

wow, there's a lot of exciting things and it you can get really,

really niche. But I

The reason why I focus so much in neurosymbolic execution is that it just

has a very high confidence rate in a high in a low hallucination rate.

And it's you could probably sh you could like ship this to a production

~ instance and be confident, like,

hey, like I did that, I put my name on that.

Cause for me, like if I'm gonna put my name on something,

I would at least, you know, want to make sure I provide the best type

of workflow or a gentic workflow.

for that client or for the company that I'm I would be working for.

And so right now there's also an issue from my perspective when

it comes to deploying a Gen Tik solution.

A lot of people still use NAN and that's great for one off instances.

If you're the only GTM engineer,

use it. Or if if you are like

very new and you want to use the cloud,

the cloud ~ co-manage core management solution,

whatever their workflow in like tool is,

you could use that too. There's just so many different tools that

you could effectively use. I personally use trigger.dev just because I switch from

a TypeScript and I let the AI solution basically help me write that.

And I would just like it for me,

it has a higher sense of reliability.

And also the reason why is because once say there's a node that's missing,

like that doesn't trigger, it would restart.

So NAN and some other platforms,

it would restart the whole trigger flow and it would use up my tokens.

So what happens if I do trigger.dev is that it would pause,

stop, and then it would basically use it again.

And if I went to an enterprise company and they didn't allow trigger.dev,

I would just, you know, self-host it because there is self-hostable.

Most of the solutions I select are going to be either,

you know, cloud-based or self-toast.

Obviously, self-host, you have to be cautious that you do,

you will have to maintain that.

You will have to host that. And it will,

you know, eventually break one day,

and you will have to create a maintenance documentation and make sure that that

platform does not break if you self-host.

There's always a chance that self-hosting is something will break eventually.

~ if you're lucky, probably not during the stint of your time there.

But

make sure it's properly documented.

~ so the reason why I also switched from GitHub 2 to more of an

NAN solo also is because just like the auditing,

and I could work with multiple types of individuals.

Like I can mo work with a GTM engineer,

I can work with multiple GT GTM engineer and a forward deployment engineer from like

if a different department decided to give me an individual,

then I could basically get that type of

support from another different person and I would segment them

off in different GitHub repos.

So it's basically getting that development heavy iterative instance

from multiple players and making sure it goes to a sandbox,

it goes to an environment, it goes to a gateway,

and then it goes into the different types of tools.

So for me, I branch all of my GitHub repos and then I do empirical reviews

for all of the sandboxes.

And then I go to a staging environment that has the most up-to-date information

~ from the production. And then I put everything right now in Superbase staging.

That's for most small companies.

You can use ~ Databricks has a lot of good tools.

You can use Snowflake, you can use really any SQL database vector.

There's I just use this one because it was self-hosted and it was,

you know, free ninety nine. Who doesn't like that?

And a lot of people use Slack.

as staging, I know the new the old CEO of Twitter came up with the new type

of Slack competitor, but for now I'm using Slack because it is widely used within

all of the different types of enterprise or mid-size or small companies because

you know Salesforce acquired them.

Probably one of the best of acquisition,

in my opinion, because of the chat feature.

And now a lot of people are using that as a human in the loop.

So once it goes to the staging environment,

it also goes into production.

And so when there's a production for trigger.dev,

I do have a production hook. So even in my even in that like type script trigger,

I have a sandbox, I have a staging,

and then I have a production environment when it comes to setting up a trigger.

So when that goes, it goes into Zod,

which is a drizzle schema, and that looks at,

you know, like it basically audits certain things as well and to make sure it's

safe. And then I go to cloakpipe.

So it's a P I I. ~

Tool, which basically cloaks your personal data.

And for me, that's really important as you are creating AI workflows across

the customer lifecycle. So if you go from awareness,

that doesn't have that much PII data because that's public information.

But as you go into like, I don't know,

selection and you go into mutual commits,

onboarding, retention, those start to have higher PII data because you're

Taking information, personal information from your customer.

And so that tends to have, you know,

PII data that you might want to shield from LLMs,

especially because you don't know what they're gonna do with that data.

And I would also be cautious about ~ a lot of small businesses that are using

a lot of agentic AIs from Cloud or ~ or Chat GPT.

They're giving an army of AI agents to that to small businesses,

and they don't have

people

that they can hire that are AR architects.

And they're just giving away their personal data and workflows to these

big LLM providers. You know, if a big company like that gives

you something for free, you're actually the product or your data is the product

of ~ of theirs. So they're basically taking that from you.

And then ~ so it goes to Vercell as a gateway.

So Vercel, a lot of people would think is just like a hosting platform.

I use it as a gateway to actually push a lot of the

Trigger.dev or a lot of information back into like a way where I can send

it out to other technology like Slack.

And then I also put an LLM switchboard so in the middle of my stack.

So I can like trigger, I could use different types of LLM providers,

especially if you know, I think now a lot of companies are going

to be very stringent on price on like price of LLMs for different workflows.

Cause I know a lot of big consulting firms.

They're basically using a lot of tokens using Fable Five to like make PowerPoint

presentations from PDFs, right?

Like a lot of people are just max tokenizing them.

So in the future, probably in a few months,

I think, because everything's just happening super fast.

~ everything is going to be more stringent.

There's gonna be a closer eye on like what LM tools you're using for each workflow.

And is it price conducive to that particular workflow?

And then I use Langthuse, which is one of my favorite tools.

And it basically logs trace all of the LLM.

So you have that black box is open and you see what's inside the LLM provider.

And then I put Superbase Productions,

and that is where I put all like the logs of each of the AI activities.

So I use Langviews to basically open the box,

and then I put Superbase to put all the things back inside a different table.

So it's all organized and clean.

And then I use Slack as human in the loop to basically get ideas from different

types of, you know.

verification. If it's a yeah, if it's hey,

I approve of this. No, I do not approve of this.

And why don't you approve of it?

Right.

I'm gonna just show you guys in my lab so you guys know.

You guys can also access this.

This is all free. I decided to just give everything for free.

I don't really have a particular need for like gatekeeping anything.

if anything, sharing is caring.

~ so for neurosymbolic, I basically started like here is like

an actual way of how to use the neurosymbolic agentic pattern.

Building and I what I did is I ~ separated all the first mile neural,

middle is deterministic, like I said,

last one's neural. And if you look,

a lot of the revops for forecasts and wind loss churn signal,

these are these parts are going to be a little bit more heavy in the neural part.

So that's why they have a 1080-10 split,

80%, 20% ~ of neural,

but they're split against each of the very end.

And so if this is the middle part,

and this one talks about like the CRM loss,

attribution, what not to do, why not,

and native parent call-outs, when win loss.

And it also show shows you the switchboard that I made,

and that shows you which type of agent would probably be best

for that particular type. And I actually just researched a lot of different types

of models, what they would be good at,

and then specifically

put

it inside an LOM and then decide and then let them decide which one based

on the different details I put into it.

So I mean you could test it yourself.

That's the best part about this lab.

You could test everything yourself.

and so you there's also guarded ones,

which is a little bit less than ~ the more symbolic,

less. And as you get more symbolic,

your percentage of confidence rate for each of these workflows actually increase.

So that's a great part. ~ and then yep.

So there's more deterministic ones.

These are very much more straightforward.

You don't even need an AI. Some p some people are like,

I need an LLM for everything. No,

you don't. there's things that you don't really need an LLM for.

You can just like straight shoot it and you could use,

and if you are to use it, you could have it run like at the very end.

Cause like call brief. I have it like there's none LLM at the very b

at the very beginning, but there's an LLM at the very end,

which is like going to try.

like transcript the API, write a CRM A AI API.

And that she uses like the sh the most like not the smartest model for

I'm gonna not gonna say that because each model is good at the it's certain things,

but a very cost effective LOM model.

And so these I actually just, you know,

put this all together for you guys.

~ I've been putting these up together.

Like I've just been shipping as many things as I can ~ to help,

but I just haven't really like

talked it out to anyone just because I've been having so so much

fun building honestly. And I also did the marketing side as well.

for neurosymbolic, I only had paid ads because a lot

of the actual AI workflows,

you don't, you don't really need like that much neural,

not not that much ~ LLM for the beginning part,

but you might need it for the very end.

So an examples like SEO, I made this one and

This one's also in my GitHub repo.

You guys can have it. It's all free.

Everything's free. ~ so yeah, that one basically looks at your daily,

weekly audit. It ~ we it basically scrapes from a third-party platform.

It might be for free. That's an SEO.

We can actually go ahead and look at it.

To do to. All right.

There's so many, actually. It really depends on what your use case is.

So this one I use ~

Data for SEO. It's an on-page audit and I scrape that.

And then I most of these I started realizing I should really build

a system design document ~ for each one,

like a topography. I hope I built it here.

No, I didn't, but I will edit all of these to have a topography.

Here's ~ a lead enrich and source score route,

which basically leads enriched ICP scoring,

read loud.

And

for this one, I basically I don't even use a spaghetti AI anymore.

I rarely use NAN. And if I did use an NAN,

there was a moment in time before October that where everyone was just dumping

L ~ NAN workflows for everything.

I would just sometimes look at that and I created a skill that converted that

NAN skill into an actual trigger.dev so I didn't have to like

overcomplicate it and then I would basically use the neurosymbolic pattern

to then layer that on top so then I'm not starting from scratch and then

you know use my taste and judgment.

Taste and judgment is actually super important.

Like how like how is this going to scale right?

How is this does this even look good?

~ so hopefully you guys are able to you know look at that reference

of neurosymbolic and see how

different splits of the LLM pattern,

what LLM I'm using, potential tools that you could potentially use for each

of these agents. So for me, like an example is this is actually

so lead enrichment and routing pipeline.

If anyone's interviewing for any job,

this is one of the like assignments they give you.

I'm not even joking. Like this,

I've like when I was I was curious,

so I started like talking to people and talking to recruiters

and talking to companies.

And a lot of people use the lead in enriched score route.

Like probably out of the 35 interviews that I've done,

probably 80% wanted me to rebuild this one.

And it was getting like so so annoying in terms of like,

okay, that's like a very easy one.

~ and so this is one, this is the one I've been using,

and it has a

100% rate of saying you pass this technical challenge.

So if you want to win, or if you want to pass one of those technical challenges,

this is the one I used. ~ so usually they do like 30 mock-up leads,

they give you a mock-up lead and they're like,

hey, can you enrich this? So most of the companies that I've interviewed for,

they actually want you to, they just give you a list and can you,

you know, make this you use API,

use your LLM skill.

So most of the enrichment is really not that much LLM at all,

if anything, probably towards the very end for routing,

you might need. But these type of tools,

~ let's see it for this one. I basically got,

I they give you the fields, they give you like columns,

they give you proprietary category field by.

And so what I did was basically,

you know, got all the fields. I use three technologies,

I believe. Like you can.

Basically just stack all the data enrichment tools that you want

and have trigger.dev basically enrich until you can't say anything else.

And then if you can't find these enrichment,

go search the web, crawl the web,

and tell me if you find all this information.

And then that's basically how I do the enrichment part.

And then the routing rules for the very end,

it goes into like, ~ how are the routing rules for this?

Is it going to be

Is the routing rules going to be for existing customers?

Is it gonna be for industries?

Is it gonna be by ARR stage of that company,

like company size? Or is there a particular partner that would be interested?

And then making sure you segment,

you know, the actual customer versus net new.

So you're not like sending a net new to an existing customer war

or a existing customer to a net new business development rep.

So this is a this is when you could download.

And even if you see an agent on GitHub,

I would highly, highly, highly recommend to just make sure you take a sample

set of your suit ~ of like your current data set from like either

CRM or whatever tools that you're currently trying to ~ create an agent

for and make sure you you could put spin up a free superbase ~ like workflow.

Like here's one. you can just spin these suckles up and then just simulate,

like have an identical twin of your ~

of your stack and that will basically help you create like an environment.

So just make identical twins of each of your the development stages in your API.

And you can look into like their help documentation for that.

And then you can make an identical twin of the environment that you're

in and then plug in your agent so you have a higher threshold of seeing

if that works or not. ~ so this is the build enrichment routing pipeline.

On

After you build something too,

you realize, I could have done that better.

And one of the things I've done better for lead enrichment is

I actually created like an AI staging of clin of data.

So once new data comes into my platform,

like a HubSpot, it goes into a super base,

and that basically is the staging,

like the waiting room at a receptionist desk.

And then I I dump like all the enrichment ~ providers I have.

And then once it gets enriched,

I push it into a Slack channel,

like through for sell through ~ like trigger.dot through for sell through Slack.

And then I ask the human in the loop,

is this okay? And once that is okay,

then that staging ~ basically pushes into the CRM.

So you have like a safe space for your agentic solution for your

AI to basically work without you being nervous that am I hurting anything?

~ like am I hurting my entire instance?

So like I I the reason I did that is because I actually when I

was buying technology for big law,

a lot of their data enrichment had a staging environment.

So I just borrowed that element and then applied it to my AI solution.

So for this lead enrichment and routing,

this one I use Bitscale and I use NREV to basically enhance

out enrich my LLM.

And so that basically helps, you know,

if NREV doesn't have it, does Bitscale have it?

And if BitScale doesn't have it,

you know, does Clay have it? And then if not Clay,

then can you browse the web? So sometimes what I do is I use because

I'm very cost sensitive to a lot of like the tools,

especially for my clients. ~ what I do is I put that most expensive enrichment

at the very end. So if I don't need it,

I don't use it. And so

That really, really helps. And then this one also scores.

I totally forgot I put a scoring rubric.

And so the scoring rubric is a skill that I also put,

and it's symbolic. And you can change that as well.

What I'm seeing also in the industry is a lot of people are having

a very difficult time creating skills that can work in a co-pilot and governance.

So you want that consistency.

Because

when you score a certain lead,

you make sure that that scoring is consistent within all your revenue technology

stack. And it's also like those type of skill is also consistent

in your governance of how those trigger.dev or N,

how those workflows are working with the LLM,

and also like how individual co-pilot people are using like Cloud Code

or Cloud Cowork. That skill is consistent.

So what I'm seeing right now is there's a lack of consistency.

happening with different skills in different ~ like

so ps and so i would highly recommend to put all of that in github

and then have that as your central of tooth so you can push it within

all these other platforms so you have a consistent scoring rubric

you have a consistent way of you know explaining your product

to new customers you have a consistent way

Of explaining your retention syllab or your retention like battle cards

to individuals. So from the scoring perspective,

I highly cr recommend making a skill or like a very defined ~ way.

And it has to be consistent and making sure if you are scoring

in other platforms that that is also accounted for.

So I've seen that a lot and I've heard a lot of people that

are in revenue enablement, sales enablement have a very difficult time because

everyone's messaging is so like

Wild, wild west, because it's not consistent and there's no center

of excellence for messaging, scoring,

or like you know, branding. So that is that is what I consider also

to be very problematic. And probably you can I've been using GitHub as that.

Probably I can show you some other day.

But ~ and then there's routes.

So it riches, it scores, it routes to the correct indiv place.

And so yeah, that does that one.

And then you get download that as well.

I also did a few like both and SEO agents.

There's more. I'm probably going to be posting just like a bunch

of agents throughout the month.

and I also made a neurosymbolic agent,

which is more of just like it looks at your agent and then was like,

how can you make this more neurosymbolic?

And then there's a competitor agent.

So there's a lot of different agents in here that you can play around with.

But like I said,

No matter if it's me, if it's someone else that you like,

I would never take someone like just take it right away and send it to a client.

Really bad practice. Make sure you test that environment.

Make sure you have an identical twin of their environment or set it up

so you can test that and then make sure you do change management accordingly.

So what does change management look like for AI?

It's a little different. So what I always recommend is like if you have like

a Slack, like I've been testing this one for

Fishes and giggles. ~ this one,

~ so like here's a daily writing one just for fun.

I wanted to like test it out. Each one will have channels.

I highly recommend doing channels for each of your different types

of workflows and making sure that the team can actually see you interact with

the Slack channel. Like if you approve and deny,

get like that, lets them that shows them like how they should be interacting.

with your agent. So for governance,

I use Slack as the center of like truth for communication perspective

for the human in the loop. But I make sure that the first thing is like,

you know, you want to expose them to like how to interact with your agent.

And so the first thing is like,

hey, how do I show them how this actually works?

And so I would open the Slack channel for the first like day or a first week

and show them how I would interact with it.

Like

This is when a new lead comes.

This is what it would look like.

Right. And then ~ basically taking their regular day and their new AI day.

Cause there's like a new AI way of doing things where they don't have to

go open a million tabs, go and click and clack at like thr s ~ Hubspot.

Then they have to go to ~ what do you call it?

Salesforce, then they have to go to ~ sales loft,

then they have to go to like Zoom Info,

then they have to go to like everything.

They can just go on their Slack.

open up, hey, like talk to a chat.

Or, you know, it for me, I've already created a cron job that would automatically

send them, you know, different types of leads that they have,

different types of accounts they need to be looking out for.

And so that those are different types of ways you could do change management from

the get-go. It's just exposing your Slack channel and showing them like,

hey, this is how you interact with your AI for governance.

And I'm not saying let's stop them from in this ~

working on cloud co co-work or op ~ open AI's chat GPT.

Like let them have all these co-pilot solutions tools,

but make sure you control the governance.

That's the that's that split I've been talking telling about.

Like let them have that co-pilot to make their certain jobs efficient,

but to the jobs that really matter,

like if you're talking to a customer,

~ if you're talking to a partner and if your brand is on the line,

I highly recommend to make sure that

is centralized, organized, and that has a s a certain type

of governance applied because everything that leaves your company now,

you know, before, you know, was it you were able to kind of monitor it.

Now with AI, it's like everyone's just dumping AI everywhere.

So ~ or content everywhere. And it's putting your ~ brand and logo name on it.

So yeah. And then the second part would basically be doing workflows.

I mean, I do it, I always include for

change

management to besides opening my Slack channel and showing people always have like

two to three individuals that you would or one real if your company

is really small for that department and team and make sure you work them throughout

the workflow and they get the approval and it's like hey let me run like

10 to 15 work ~ outputs from this workflow in a staging environment then

a staging environment and making sure you start creating that business ledger.

From you know that workflow. So you could at least get a higher confidence rate

score, you know, even higher confidence screen or even if when you push

it to production. Cause then when you push it into a sandbox and

let that like staging environment of one to like two weeks,

let that simmer a little. So he can get really excited and he can hype

up the rest of the team. You always want to pick someone that's like very positive,

very ~ tech forward.

You always wanted to leave if they have a ~ say if they have a stand-up every week,

you want to make sure that that person gets like five minutes.

You talk to their manager, you give them five to ten minutes of that time,

or whatever five minutes. I think some stand-ups are 15 minutes or 30 minutes,

not too much, but like five minutes and give them that time to say,

Hey, this is what I'm doing. This is how it's increasing,

you know, our efficiency rate,

our operational rate.

~ and this is how it's also increasing EBITDA if you're gonna talk to your like

~ CFO or COO. And so that's gonna help motivate the team and it's gonna create

a little bit of momentum going forward.

And then ~ once you do like one for a department or team like agentic flow

and you push it into production,

I would also recommend like a little bit of an office hours and making sure it's

like, hey, the first week, these are how many outputs we've had,

this is how many human in the loops did we have.

Just basically gut check all of them.

~ and s and make sure, like, hey,

are is did you really mean to approve this?

And making sure you check that business ledger.

Because at the end, I think that business ledger for each of these

AI workflows is gonna be a very priced commodity for your company.

Because that no matter if that person leaves,

that business ledger stays. And also if you get bought out,

that business ledger becomes very,

very important from an efficiency perspective.

So yes.

And then after you do those workshops,

I would probably recommend making sure every week after that,

until it gets really good, to look at the drifting that happens with

the LLM because there's lot of drift that happens.

So this is one cycle of an AI workflow.

What I do is I also create a neurosymbolic governance around each

of my AI workflow as well. So I haven't built out that in my GitHub.

I will probably launch that by the end of this week.

Where it basically has one loop that's always,

you know, human in the loop. That's and then there's a second loop,

which is a governance loop that also has neurosymbolic pattern.

So that one checks for more like LLM drift.

Sometimes your LLM likes to drift and ~ from its actual goal.

So that will basically help corral it.

Think of it like those things from bowling where they just go flaw

and like they stop those bumpers.

that will help it guide to the right place.

So that pops up once in a while.

If it's going a little, if the bowling ball is starting to go a little over there,

a little over there, it goes whoshaw.

And it goes straight and then you get a strike.

So ~ yeah, I've you know I explained one of those workflows.

There's so many other ones. ~ another one that I also highly recommend that

I put it in my lab was also like AI agent autonomy,

like how you're organizing your your folder structure is really important too.

I know this one I got, I like looked at Vercell released their folder structure.

~ so just make sure a lot of people are also using folder structures

to decrease on token usage as well.

And making sure, like, hey, is this the right type of folder structure?

making sure that you have all these types of folder structure for each

of your agents makes it look super clean and very like ~ like that your closet looks

cleaner.

in his drawers and folders for everything.

~ so I always recommend that as part that's free on my lab.

I also put like GTM stack on on my lab.

These are all free solutions. I think ~ it's really important to just be part

of a community that's always willing to share.

And I put like certain types of GTM tools that I really like as well

as like the website. And I also created like there's a lot of skills that people

are doing too like agent skills.

I've just been like hoarding AI skills for a while and I realized,

you know what, it's better to just make my own skills.

But it's always good to start with something than nothing.

And just tweak it as you go along the way.

It's an art. This is an arts. So it's you're almost like a creative artist doing

these things. besides the skills,

I also have, like I said, the neuro symbolic ones.

And I will also be posting a lot more.

Like I'm gonna have a link here,

like another column that will basically.

push to the right tool. And what I'm gonna do is like duplicate,

like have a GTM like clone of what instances like Salesforce

is a very popular ~ CRM that you can just duplicate as part of like simulate what

it would look like and have the different types of data fields and

~ API fields flow. ~ Other than that,

I am done with sharing all my cool agents.

I mean you they're all for free.

And

I feel like everyone's also start going to start having GitHubs.

They're the like the Pinterest of GTM workflows now.

Like everyone has a Pinterest board,

like for parties, but you're curating one now for your company.

So there's that. Well,

you guys have a great day.

Thank you so much, ~ Jamar. This has been awesome.

Thank you so much. I mean, I really love this.

I mean, I'm really passionate about neurosymbolic agentic AI pattern building

and go to market systems. It's definitely a mouthful,

but ~ I'm very happy about like at first I was really sad about like,

you know, if I was to send out an NAN and it would break an NAN workflow

and it would break on on my clients,

I get really sad because I'm like,

I've why am I sending terrible work?

Like that's like really like it hurts my moral stuff.

I don't know. It helps my it hurts me.

Saying, like am I am I sending like terrible work?

But now I'm like I'm a lot happier sending work that I know it's a lot more durable.

I absolutely love it. I also love the cheap code if you ever get asked

in an interview or you could probably use that git when you get asked in your

job as well. to do you know, I I w I I wanna

Yeah, I mean I I guess

I wanna point out that also is good for jobs.

Yeah, it's good for jobs. Like I mean a lot I've most of the most of

the workflows I did were like job interview questions,

right? Like, how

That's right.

did you do this? etc. Like a lot of people want to prove now or show

you like how did you do this? They want proof.

And here's the proof.

Love it.