I just dropped the PDF, I don't say

anything, [music]

and it's like, oh, this is the insertion

order. Okay, it's time to start our

sponsorship scale.

>> How much is your monthly Claude bill?

>> Last month, around 4 and 1/2 thousand.

This month will probably be more. I have

like between six and 10 sessions [music]

at any given time on my computer. I'm

using Logitech MX 4 mouse. One button is

the hands-free toggle for WhisperFlow.

So, I just scroll around the sessions,

give feedback.

>> Any metrics on business outcomes?

>> We've had roles that were on the docket

to be hired, [music] and we've paused

interviewing and hiring for them because

of the operating system. It's really

easy to send the email faster or send

the Slack faster, [music] but are the

highest leverage parts of your job also

AI supported? And the honest answer for

a lot of people, it's not.

>> That's Matt Swolinski, head of growth at

WhisperFlow, and one of the rare

marketers who dove headfirst into AI so

he could build [music] the tools his

team actually needs. In this

conversation, Matt shares his fully

automated system for buying newsletter

ads, how he trains Claude like it's a

junior employee,

>> [music]

>> how his team freed up 20% of their time

by automating admin tasks,

>> [music]

>> and why they've put some hiring on pause

because their current team has gotten so

efficient. We're in the age of AI search

and agents, and a new kind of marketer

is emerging. We call you the marketing

engineer, and this show is for you.

I'm Nick Lafferty from Profound, and

here's my conversation with Matt. All

right, Matt, let's start here. How much

is your monthly Claude bill, and where

do you rank at the company in AI usage?

>> Yeah, so last month, I think I clocked

in around 4 to 4 and 1/2 thousand.

Uh this month will probably be more. Um

I'm the number one person outside of the

engineering team, so definitely

relatively high up there.

>> Amazing. Um okay, so can you give a

quick background of what is WhisperFlow

and what do you own there?

>> WhisperFlow is a voice-to-text solution

that essentially layers in AI to take

what you say

and turn it into text anywhere the

cursor is, which ends up being three to

four times faster than typing. So, we

are the all-encompassing layer to

essentially get everything that's in

your head out there into the world much

faster.

>> Amazing. And then, what do you own at

Whisper right now?

>> Yeah, so as head of growth, I own

every KPI between impression all the way

to the download event. So, all the

website, all of experimentation on the

website, landing pages, all ad channels.

So, quite a bit of moving pieces given

the the scale that we're at.

>> Yeah. Is it mostly self-serve sign-ups

that is your primary KPI?

>> Yeah, so we as a business are optimized

towards B2C,

but obviously that

dovetails into team creation and

enterprise creation. So, we actually

just last quarter

hired our first AE ever from the start

of the business, and a good share of our

revenue already is teams and enterprise,

and that's because we've we've focused

on PLG. That's kind of the the

philosophy behind all of our growth. So,

the primary driver of all of our spend

as well as all of our marketing is to

make it really easy for an individual to

kind of gather that magical aha moment

of what it means to use WhisperFlow and

never be able to go back. And then,

those people will convince their teams

or their orgs to kind of bring it into

the fold.

>> Got it. Okay. Yeah, that that makes a

lot of sense. Okay, so great context on

Whisper and you and what you do. Before

we get into tactics around the marketing

engineer, I kind of want to make the

case for why this work matters. There's

a version of this for someone watching

who

is someone who just likes to build

things. And so, what do you think is

maybe the difference between someone who

just likes to tinker and build and

someone like you who has built more

robust systems at Whisper that is

actually driving really results for

y'all.

>> Yeah, I think the the big difference

there is a lot of people with the power

of AI can build cool micro tools that

are useful to them.

Uh and the difference becomes when you

have to turn that little idea into

something that can actually be

load-bearing, that it's used in

practice, used by various people, and as

scale changes or systems change, that

doesn't break or it has the mechanisms

to alert you if it does, right? And I

think that that's the core difference

here is

everyone's become AI native because of

the existence of the tools. I think very

few people have tacked on the layer of

being an engineer, even if you're not an

engineer, as you're building like an in

our case around marketing engineering,

even as head of growth I've filled that

role here internally at Whisper, where

I'm trying to take my workflow now,

apply it into the team, uh and that's

you know, obviously a a ton of changes

that are needed, but it's the the

repeatable loops of actually building

systems that can, you know, sustain the

the tide of time, uh and not just be a

small little item that gets used once in

a while.

>> Yeah, that that makes a lot of sense and

I do want to get into some of the things

that you've built. The last time we

talked, you mentioned building a

automated waterfall kind of for managing

newsletter, podcast, and YouTube

sponsorships, bunch of creator marketing

work. Can you kind of describe what that

process looks like and then we can kind

of like drill into more details?

>> Yeah, so this was my uh first core use

case for I'd say a deeper system that I

built with Cloud Code, and it was just

like the the biggest thorn in my side

uh when uh overseeing kind of the whole

marketing function at Whisper. So I'll

I'll zoom out a little bit and and and

note that up until December, so the end

of last year, I was the single person on

the entire Whisper Flow team doing

execution in marketing.

Which like the the role definition of

yes, I own the KPIs, but I also own all

the execution. So, all of meta ads, all

of Google ads, and everything that

connects to that,

um I was running the strategy and

execution entirely solo. So, I started

using Cloud Code a bit earlier than most

uh in in October of of 25. So, this is

before the the latest models that made

it truly great, but I already needed the

the support around some automations.

And the first thing I needed to automate

was uh newsletter sponsorships. So,

newsletter sponsorships, and not a lot

of people know this, I wish it wasn't

this way, but they're they're run in a

really antiquated way. So, you interface

with a newsletter over email.

They give you the rates, you negotiate

the rates, then you get a contract, uh

then you have deliverables. Every

newsletter has slightly different

formats, different copy amounts,

different image parameters.

Uh and then you write that copy, you

send it over email, and then you give

them a link, you see how it does, and

then you move on to your next placement.

So, all of that Now, now multiply that

times around 100 [snorts] newsletters

that we run uh placements on any month.

The audience is different, the framing

needs to be different, there's a ton of

copywriting and a ton of data that needs

to be analyzed, and without AI, it's

essentially impossible if you're running

anything else.

Um so, the first thing that I did is is

I built a essentially like a master

skill that invokes other skills along

the way. And all I need to do if it's a

new partner, I just drag in a contract,

and it first looks, you know, does the

CPM make sense? You know, what what can

we glean from the the internet as well

as research around this audience? Are

they a fit? And that research component

is the first thing that I would normally

do. I made that the the beginning part

of of this master skill, and then it

gives me some pushback right whether I'm

negotiating price or trying to

understand maybe this is a newsletter

test, and it's not a core ICP, but maybe

it's worth exploring. I have a thought

partner in in in AI along that. Uh and

then also

um from that moment let's say it goes to

a signature, I drop in the signed

insertion order or signed contract,

and that essentially starts all of the

copywriting, all of the link creation.

So we use dub.co for our affiliate and

link tracking system.

So with the right UTM parameters, with

the right promo code, everything

pre-built into the link, it creates

everything using a certain voice and

tone that we know works in newsletters,

but it's also all fed off of all

previous data of all newsletters that

have run. So experimentation newsletters

is also difficult because

you commit $10,000 to a single

placement, and then you kind of like put

your finger in the air and and hope that

it works.

But you want the more that you run, the

more you've like hopefully learned, and

you can feed this all into a system that

removes all of the boring admin, but

also just makes it way more powerful,

which I've seen no one that is doing a

sponsorship

kind of do this way even if they're an

individual,

but I took it all the way to trying to

fully automate that whole workflow.

>> Wow, I've so many questions. So when you

said you you built a master skill, is

this one are you running this in the

Claude UI or is this in or the Claude

app or is this Claude code, too?

>> Yeah, so I even with co-work or Claude

code in the desktop app, I like doing

everything in VS code, and that's just

because I can spin up many sessions in

the terminal bottom, and then I see all

my files plus the the actual MD or

something that I'm reading right above,

and it's just like a

I've gotten used to it. I probably could

use the Claude desktop app for like a

similar look and feel, but it it gives

me a level of advanced control where I

see where it's pulling context, I see

all of the thinking, I see the file

maybe change in real time that like

removes some back and forth that I still

have to do with some things in the

Claude desktop app. But they are making

it better. Maybe I'll make the switch

soon, but that's yeah, that's cloud code

in the terminal is where all of this

starts. So,

often I don't even have to do anything.

I just drop the PDF. I don't say

anything and it's like, oh, this is an

insertion order. Okay, it's time to

start our sponsorship skill.

First checks if the sponsor exists, so

it doesn't duplicate any actual work. If

it doesn't, it creates, you know, a

folder with context about this partner

and then all of its research and then

all copywriting ever created for is in

one place. So, if someone asked me

what did you run two months ago for

TLDR? How much did it cost and how did

it perform? I could ask the AI. I could

go to a place in the folder cuz I know

how it's organized. So, I think the the

organization and and how you think about

it is also super important from a

systems thinking level.

>> Yeah, okay. Yeah, that's something I've

been thinking about too is how do you

organize all these different projects?

And so, like on my machine, I have a

GitHub folder and then within that

folder is all the, you know, a new

folder for every project I build. And

so, is that kind of how you approach

this too where you have this is one

folder and then within that that's where

all of this kind of operates?

>> Yep. So, the the Whisperflow marketing

OS,

very unique name as it's called, is get

controlled and shared across the whole

marketing team.

Inside of there are essentially

placeholders as well as like a read me

for anyone that's installing it to

essentially it creates like a personal

folder

as well as a code base folder.

So, even though the the whole

like OS folder is get controlled, inside

of it we also pull for our analytics DBT

and our code base

and those are obviously separately get

controlled,

but those are also get ignored from the

the Whisperflow marketing OS. And then

personal context, so like working files,

outputs, things that I don't want synced

live there. And then everything else

that essentially bleeds into shared

context. So, all of the copywriting and

everything that I'm describing here is

in a folder called growth. And then

within that, I have different folders

based on which growth channel it is or

if it's strategy. So, this would be

under sponsorships and then within

sponsorships, we have some JSONs that

look at like with whatever quarter we're

in,

what name of sponsor and day and dollar

amount to placements and then details on

each sponsor within their respective

folders.

>> Got it. And so, and then over time that

system grows too as you work with new

new sponsors, new placements, you run

all this stuff and then I guess most of

it you commit back to the GitHub repo

minus the things you have that are more

personal that you get ignore. Is that

right?

>> Exactly. Right. So, then like if I'm out

or anyone needs context around

sponsorships, they have everything that

I've ever been working on in a shared

place.

So, the the note that I always say,

assume it's shared unless you have to

think twice. Like, am I negotiating

about someone's salary or you know, is

it like sensitive contract details that

like the team shouldn't see. Then those

should be kept personal, but everything

else like even a contract that's signed

with a vendor,

someone is going to ask like, you know,

when we went through like we're going

through the process of series B right

now, you know, from both due diligence

and finance, we're getting, "Please

anything over 100k, please send the

contracts and like organize them this

and this way." Everyone else on the team

other than my org, this took them like

one or two weeks. I just dropped a zip

with everything pre-sorted cuz it was

already organized that way in in the

file system.

>> Wow.

Okay, so for someone listening to this

and they see or this amazing system

you've built, what is your advice for

someone who wants to start building

something like this? Maybe a V1 of their

own kind of influencer waterfall like

sponsorship engagement.

>> I would zoom out even more to to say

Uh, whole thing from zero was built with

Cloud Code for Cloud Code. Uh, and

that's just the the philosophy that I

always take. If you're unsure of

everything that we're talking about

here, everything that you read online,

try and create it yourself with the help

of AI. And if there are things that are

set up a certain way and you don't

understand them, ask. Like, what is it

getting ignored? Why are we doing that?

Like, just ask and and work with the the

LLM because, you know, we live in a time

that we have the biggest possible

learning unlock ever before seen. And

like,

I'm not a developer. I've, you know,

never shipped actual code and I'm

pushing PRs and, uh, you know, building

an operating system simply because I've

taken the time to stress test and learn

this. So, the reason why I wanted to

zoom out and not just say for for

sponsorships is the entire operating

system

I started with a prompt. Uh, I want to

build an operating system for me, uh, to

do my work. We're going to do these

things together. Uh, these are some

workflows. Uh, let's get started on

that. Asked me a bunch of questions.

It tell me more about your role. What

are the things that you need to

automate, etc., etc. And then we built

the the first skeleton together with

Cloud Code of the operating system. That

was my V1. At the end of the day, the

skills that you build and the way that

you use it needs to be meaningful. Like,

you actually have to use it cuz if if

you build skills and they never get

used, then move them into an archive

because evidently there wasn't intention

behind what you built and it should be

repurposed maybe later. So, uh, that's

the main my main frame of reference

here. Like, use Cloud Code to help you

build this because it knows best on how

to organize it. It won't be perfect and

that I think the the the frame of

thinking here needs to be

AI will only give you what context you

put in. Uh, and you have to say, "No,

this doesn't look right. Let's rethink

about this. Let's It's meant to grow and

evolve over time as you work on it." And

that's why like from October of '25 till

now, it's the same operating system that

has just gotten better and better and

better and better every single day via

just me working with it,

micro-optimizing it, and then, you know,

essentially over the last 2 months, I've

turned my personal OS into a team OS,

and that's been a, you know, separate

endeavor.

>> Nice. Yeah, I do want to ask about your

migration from personal to team.

Um before that though, they're built

this now, is there anything you would do

differently compared to when you

started?

>> Yeah, I think

at least from my personal experience, I

probably layered in like the voice and

copy guidelines a little late, but also

that I think the models have gotten

better.

I have to be very explicit to like avoid

slop,

but I would say with the right MD and

like guardrails, the models can be very

good at copy, and they still need that

human touch. I definitely spent a ton of

time editing the early copy, and like

giving it feedback in it. It didn't

persist in the right way, and I didn't

fully understand maybe the best way to

approach what I call the the voice guide

of the do's and don'ts, and how to log

when I give feedback. There's a reason

I'm giving that feedback, and for it to

log that in a certain layer of context,

and that's where the voice guide comes

in. So, when I'm updating copy, I don't

do it directly into the Docker MD that

it's generating it.

I prompt it with Whisper flow, and I

tell it, you know, this doesn't sound

right. This is how I would say it. Let's

try that out, and and then let's see why

does that read better. I know I work

with Cloud code to essentially get

there, and then ask it, you know, from

from where we started to where we ended

up, you know, how can we ensure that we

end up here to begin with? That took me

a little bit to get there, and that's

just like building the muscle of

actually giving context, and like

working with it like a an employee,

where you give it feedback instead of

just like making the edit, and then, you

know, a lot of people do this. They they

give someone feedback, and then they

just make the change themselves, and

then expect the person to do better next

time, but they never actually gave the

feedback. So, it's very similar with

with Claude code.

>> That's really interesting and something

I think this is a bad habit I do of I'll

just make the change in the file, but

you're saying actually talk to Claude

and say no, like I want to make this

change and here's why and tell it your

thinking and treat it like a colleague

basically and then that helps the

context kind of learn and get better

over time.

>> Exactly. Yeah, and and that's like

literally I have like between six and 10

sessions doing something at any given

time on my computer

and

I've gotten to the point where I'm using

like a Logitech MX 4 mouse where one

button is the hands-free

toggle for Whisper flow

and then I press it again to end and

then it pastes wherever I clicked and

then the haptic button on the MX 4 is

enter. So, I just scroll around the

sessions, give feedback and that's like

my primarily way of working with Whisper

flow. The vast majority of my words

dictated are prompts, but they're not

the original prompts. Like sure I can

one-shot things if I do a five or so

minute dictation and I give it the right

context, but the majority of the inputs

into the terminal windows is feedback.

Is is giving the actual context and

understanding of why this is different,

why I'm thinking about it differently

and then it grabs onto that context and

I think building a bit of a memory layer

or understanding like where it should

pull certain things is also what makes

this easier and we can touch on the like

session start and session end a bit

later as well.

>> Yeah, no, I'd love to talk about that.

How do you manage You said you're on six

to seven or eight different sessions.

How do you manage all of those? Does

that get overwhelming at all?

>> Yeah, it it depends on the the duration

of of some tasks. So, they're usually

not related at all and they have a

they have a certain start and end

which which is you know, really

important. Like when I open a a new

terminal, I know what I want it to

achieve.

Uh and there there's low likelihood that

like I I continue on a session for a

couple days. And there's like various

reasons for that that that we can touch

on later as well, but everything covers

either a different channel or a slightly

different headspace of something that I

need to

be looking at or like analyzing meta ads

or coming up with new ideas around

concepts or using the Google Ads CLI to

get like a deep dive into yesterday's

data. And these things, you know, take

time. Like the the models uh I want to

give them time to think. Um so I usually

will push, you know, push forward a

prompt and it will take 10 15 minutes to

to do some next layer of a task that I'm

asking it to do.

Um so because it, you know, usually

those things would take between, let's

say, 5 and 15 minutes, uh I can move on

to the next thing and like check in and

give it some feedback. It'll do its

thing and I kind of cycle around uh the

sessions. And then, you know, some of

them will have actual manual action

items that I need to take. Uh so then,

you know, sessions will go away and then

I'll go and finish the thing, then go

close out the session that I needed to

do with certain set of manual work. And

then I'll see what's left. So like

it sometimes drops down to maybe two or

three, depending on the day, but like

usually my day starts with at least that

many uh of kind of top-of-mind things

that I want to get going.

Uh and then it dwindles as the the day

kind of narrows.

>> Okay. And then, how do you I guess you

touched on earlier kind of this concept

of session start and session end? Can

you kind of talk about that and what

that looks like for you and maybe why

it's important?

>> Yeah, so it's one of the first

uh skills that I built uh that like

endured the test of time. Uh but

essentially uh I'll start with session

end because to me it's more important.

It was built for two reasons. Um so as

as you know, right, with the latest

models, there's a 1 million context

window.

Uh but the way that model token pricing

works is

as you work longer with the session,

context grows and the amount of tokens

every single message sent also grow. So,

you you definitely seen right in the

bottom right

uh {slash} clear to clear 800,000 and

three tokens, right? Uh the next message

that you send and the message after that

is 805,000,

810,000. It just grows every single time

you you message it. And that ends up

being very expensive. So, you don't

normally need that much context.

Normally, like if you work in

checkpoints, so I let's say I'm doing a

let's say a pretty complex task and it

has three distinct pages uh phases. I'll

start a session, let's say it's, you

know,

drafting newsletter copy. Uh and once

it's done, I run session end. And what

session end done uh what it does is it

takes everything that we did in the

session and it logs what we

accomplished,

what things I gave it feedback on and

changed, uh and it logs it in local

memory of something that we recently

completed. But, one of the most

important things that it does is it also

logs what is not complete. Uh what

should be a part like a to-do or piece

of context that will be relevant for my

next session, which will start with

session start. So, session end is like a

all-encompassing

please persist everything that we worked

on and have it be just like a little bit

easier to find and also for me not to do

a lot of manual work to say I completed

this or what was that thing that we did

yesterday? All that is stored in that

way.

So, that's session end. Uh session start

uh essentially starts a session.

Uh it's primarily around like tell me

what you want to work on, uh but it

primarily also pulls from the previous

session ends. Here's the stuff that's

still floating that we didn't complete

and it bubbles it up to the surface. So,

the likelihood that I missed something

that is super high priority because I

didn't have time to finish in some

session or if I close the session and

there's like a million action items

which like is the case like I'm I'm

floating between

100 and 200 rolling open to do's that I

didn't get in time

and the only way I can manage that is

and like a lot of that work happens in

Cloud Code

and session start just bubbles up what

is priority specifically as I start my

day. So those are like the overarching

beginning and end steps of every single

session.

I have like morning and evening rituals

that are separate from that but these

are kind of like session locks.

>> Okay. And [clears throat] then

is it possible to anonymize or share any

of those? I'm so interested in how how

those work.

>> Yeah, I know I'm happy to share like the

the vanilla version. I know it works

because I've gotten every single person

on the team to start using it and when a

skill becomes commonplace you know it's

doing something right. This is the one

I've definitely tinkered with the most

and also gotten other people to like

fully adopt so I'd love to share that.

>> Amazing. Okay, so it sounds like this

has saved you a lot of time and maybe

your team too if they're adopting all of

these skills as well. Can you talk

through any metrics on time or team

bandwidth or just kind of

uh business outcomes that kind of

automating this whole system and

building like the Whisper marketing OS

has saved you?

>> This has been a project that we've I'd

say kicked off in the last 2 months.

Uh in the last 2 weeks we've gotten

about 50 to 60% of the marketing team

were just shy of 20 in like various

roles in the marketing team

and this is this covers like all of

design all the different lanes around

like influencers, product marketing,

country leads. You know, we have country

leads for the UK and India like the our

whole India team. Uh that's the reason

why that that team is is as vast as it

is and we got like 50% of people using

it right now and how I can quantify

value is we've had roles that were on

the docket to be hired and we've paused

interviewing and hiring for them because

we made a couple more people on the team

higher leverage where they could take up

an additional part of work.

Like for example,

we have someone that oversees like all

of our creator partnerships

and his whole scope of work was just

influencers mainly, you know, YouTube

and LinkedIn. But we actually have him

now brought into the system that I built

around newsletters and because it was

something that is self-contained and

pretty easy to pick up once you

understand all the moving pieces, it is

an additional thing that he could pick

up. We were potentially going to hire

for someone to run that system, but

because of the time saving he has on

research and actually working with

creators using the operating system in

his own way that fits his workflow, he's

able to attack on another part because

of savings he's seeing there. The most

I'd say pervasive time savings and like

ROI that we can see on the team that

probably saves, I'd say between 10 and

20% of every person's week is just like

admin. So let's say we we want to launch

a new landing page on the website. Seems

like a simple task, but there are many

stakeholders and people involved.

Someone needs to come up with the brief

and copy.

Someone needs to design it. Someone

needs to develop it. Someone needs to QA

it. We need to make sure conversion

tracking works. We need to then publish

it and do all those things. For someone

to do that set of work and that's like

your project management context

kind of gathering and then sharing

takes a lot of time and that's like any

knowledge worker, anyone that works at a

startup or company spends between 10 and

20% of their week just in that world of

admin. And with Cloud Code you you take

your idea and because there's a skill

built around a certain lane of work, it

asks you questions, fills out a brief,

it puts it into linear, it writes a

message in a shared Slack channel, then

sends messages to individual people with

all of their contacts

and that takes 10 minutes instead of an

hour and a half.

So that you know that that I would say

as an admin level of interacting with

Slack, interacting with linear and

giving context immediately to people and

also making it easier for the person

that also needs to assign a task to

someone.

>> Yeah. Is there a way you log or report

any of that out to leadership or is all

this kind of like

gut feel for lack of a better word?

>> Yeah, right now and I think a lot of

companies are going through this where

like we don't have the mandate to

essentially quantify ROI.

I look at this as the the overseer of

the the marketing OS just to make sure

like in in my point of view if you're

using these systems and we're spending

tokens, there has to be an ROI otherwise

everyone's just doing a lot more busy

work or just like it just feels a lot

more work to work with the AI. And at

least for me that's never the case and I

want everyone that's also onboarding

onto this

feels like a daunting task to begin with

for people that are like less technical

and you know don't know how this works.

But yeah, so I I'm the main person

that's kind of understanding from the

whole organization marketing level

are people more

effective, do they have higher leverage,

are they spending more time on more

important things?

And that's kind of like in one-on-ones

and conversations that I have with the

team.

>> Yeah, it sounds like that

yeah, one of one of the outputs is

higher leverage. Like you said, maybe

you pause hiring for a role because you

know using AI can kind of make everyone

more effective at their job, they can

touch more surfaces, increase their

velocity of output. Is there a way to

tie that to a business KPI for y'all if

it's signups or installs or downloads?

Like have you looked at that too?

>> Yeah, I mean at the end of the day like

since I was the one person doing a lot

of this for a while and the business

case is very clear where like me plus

this for a very long time, I could hold

up essentially the the whole side of the

team. You know that has changed, We we

have a lot more volume and a lot more

growth, and thankfully, there's more

attention and and kind of more focus in

the dedicated lanes.

But, I think I'll reframe maybe the

point of view I have on this

from your question to say that

I think

everyone can be more like more higher

leverage with AI, but in order to do

that, each person needs to be a systems

thinker or, you know, learn to be a

systems thinker cuz it's really easy to

send the email faster or send the Slack

faster, but that's not high leverage.

Sure, you save a little bit of time, and

then you do other things, but like are

the highest leverage parts of your job

also AI supported? And the honest answer

for a lot of people, it's not, right?

And it's like it's too high effort to

take and push AI through the beginning

of a brainstorm or thinking, but once

people can take a step back and analyze

in their role and in their job,

what are the systems that essentially

make the work happen, and how can we, of

course, scale and speed that up, but

does maybe spending more time in

research and understanding the thinking

that goes even before we assign out the

task? AI can also really help with that,

and it can also help out with all the

admin, right? And I think from from the

business case, organizations need to be

teaching this level of systems thinking

because it's where I see some of these

pitfalls in trying to say, "Here's cloud

code and an OS. Go." A lot of people

don't know where to start because

they're not taking a step back and

analyzing their role from a systems

level, and that's where I think a lot of

fallacies, even in AI usage at startups

or enterprises, and like with with

rampant token usage increasing,

I think because there's less of a

narrative on why are you doing this or

could you do it with AI and make you a

little bit more powerful,

like the token usage is is on, I'd say,

some menial things where like if they

took a step back and focused on the more

important higher level things like we've

said, it would make a bigger difference

to the business case.

>> Okay, and so if you're I guess if you're

a marketing leader watching this and

you're like, yes, and nodding your head

like yes, I need my team to be more

systems thinkers. Like what's one step

or how would you like encourage people

to kind of think that way or is that

part of a hiring plan as you kind of

like look to re-composition your team

right now around that?

>> I think everyone in hiring should be

hiring for, you know, what everyone is

calling AI native.

Um but like

anyone you can give anyone a chat window

and they can they can talk to an AI. But

the the thinking that goes into creating

the prompt and giving it feedback and

the the feedback loop around the system

is I think the main thing that

like orgs need to start testing a little

bit more in the hiring process and also

just doing this more as a a learning

exercise in the org

because like however easy it is now to

just speak your mind with Whisper flow

and give a more detailed prompt,

what you say and how much context is

actually needed is like it's still

relatively open-ended. I'd rather test

for how people think and deconstruct

problems.

So like if you know, like newsletter as

we gave as an as an example, the

question would be okay, like you have

this workflow.

What are the bottlenecks and what are

all the inputs and the outputs and can

the person actually map this out? Do

they understand you know, all the moving

pieces? Cuz if you understand all the

moving pieces and then you add in a

little bit of AI to help you, you can

build a really good system, right? It's

it's like the ability to step back and

say here all the cogs in this machine

and this is what makes it hum. A lot of

people are fixated on the well, here's

my piece. That's what I'm focused in on.

I make that a little bit more AI

focused, but the the whole machine isn't

necessarily better, right? And I I

that's that's the frame of reference

that I would take here is just step back

and and understand how the team works

together and what are the biggest

blocking points. And if you can have AI

help with those, then ROI is immediately

clear. Because if there are clear things

that are taking too much time for

people,

and you solve those, then the actual

adoption of AI is a lot easier because

people are actually using it for things

that they immediately feel the value for

instead of being told, "Well, now use it

in your job." Right? It's like it's

really hard to to teach a person that

doesn't know how to think in this way to

actually start that way.

>> Yeah. It almost sounds like interviewing

for marketing roles becomes more about

maybe pulling

tactics or topics from how you would

hire an engineer to like really more

around systems thinking and

deconstructing problems and then

layering on, "Okay, how would you use

AI to to do this or do different parts

of your job or scale parts of your job

that were super manual in the past?"

Like is that something kind of you agree

with?

>> 100%, right? Like zooming out two years,

if I was hiring for people to join my

team, it would be

"Are you the best in the world at Meta?

Are you in the best in the world at

Google?" Right? Channel expertise.

I think now, yes, that's still a filter,

but like I would much rather the person

be like middle-level intermediate to

advanced but not the best in the world,

but be a systems thinker where they can

understand all the moving pieces of

their channel or their responsibility

and be able to layer in AI to 10x

themself or 100% 100x themselves in

their role because if you're

a person that's great now in something

and you don't adopt AI into that

workflow, the likelihood that you are

the best in the world in a month is

extremely unlikely, right? So, I I 100%

agree with that.

>> Okay. And so, I do want to come back to

one one topic and then we can close out

of I know when you started, you you were

the only marketer you kind of built this

system for you and now you're kind of

rolling it out to your team. Maybe if

there's someone listening that is trying

to go through a similar path of one to

two or three people on their team and

take a thing they built and have their

team use it. What is one or two pieces

of advice you would give them and is

there anything you would do differently

now, too?

>> Yeah, so I mean

the whole OS was

it was mine, right? Like the the shift

between

Matt's brain and everything and all the

context that I did,

I had to, you know, it was not designed

for that. There was quite a bit of work

to essentially go through the thinking

of how do I want to, you know, prep for

context, what's personal, what's shared,

how do PRs work, all of this, like there

was a ton of back and forth even in the

the first like one or two people that

were my guinea pigs on the team. Shout

out to Eric and Dan that

uh went through that with me.

So I mean the the number one tip there

is just assume that everything that you

build will be shared at some point.

Because the the individual work that we

do is useful to other people and

adding into a shared pile of context

just makes the whole organization and

the whole team that much more powerful.

So the number one tip and the number one

piece of advice I would give is design

the tools and the systems you build

today to be multiplayer, to to kind of

work with broader context. Because

that's where I see, you know, this whole

industry going

when you look at a lot of new AI

startups and a lot of investment, it's

going into the harness, the the memory

layer and and everything that is

actually allowing ever growing amount of

multiplayer people interacting with this

thing that if you already do some

workflows, assume is this useful for

anyone else? And you know, and this is

like one of the things that I always

connect to as maybe a second piece of

advice. And I made this mistake in the

beginning,

is never hard coding uh or facts into

skills or things that you have live in

the system

uh because when you do that and you

share it, it's stale. It won't update.

Uh but if you have a centralized place

where let's say it has our latest ARR

numbers or the latest composition via

name and Slack ID for the marketing

team, that's housed in some file

somewhere and then the skills are

pulling from that context. So you only

have to update one place. So a new

person joins the team or our ARR

updates, Cloud Code is not losing itself

trying to find what is the latest,

what's stale, what's not. And like this

is usually where it goes wrong and it

hallucinates. It's using hardcoded facts

when it should be It should know that

something is stale and then should run

another skill to pull the latest data,

store it locally so then someone doesn't

need to do that because it sees, "Ah,

Matt already ran this. This is, you

know, most updated information." So like

I push PRs essentially every day

as maybe another little tip. So like

things that you build probably will be

used by someone

and that's like the multiplayer making

sure context is is in one place and not

hardcoding things into skills and and

things that you're working on. Yeah.

>> Okay. Yeah, it's almost like an

engineering concept, right? Of

instead of hardcoding things, you have

variables and then you have a variable

for oh revenue equals this, you know,

sign-ups equals this and then maybe you

have a skill that updates your kind of

like, you know, primary KPI, you know,

markdown file or whatever every day with

revenue, you know, any kind of like

number like that and then all your other

skills then reference that kind of

primary KPI file every day. Is that Is

that an accurate summary?

>> Exactly, right? Cuz then uh the skill

always knows the one source of truth

where those numbers need to go and then

various skills can reference multiple

context files and they are The skills

are always going to be right because

let's say we're doing a investor update

and it needs to have our latest numbers.

In the beginning, when I first was first

starting, you know, I I didn't keep

track of this and I I ran the numbers

and it created, you know, did 30 minutes

it worked on this report.

And every single number in the whole

thing was wrong is because it realized

it had some local numbers. It didn't

start a a hex thread, uh which is our

RBI tool, and uh I had to redo the whole

thing. 30 minutes wasted, tons of tokens

wasted, and just like a ton of

frustrated, you know, time having to

like quickly rush in order to get to

that investor update to make sure that

specifically that everything needs to be

right. There's there's no room for

error.

Uh and uh yeah, like that's a really

easy way to avoid that.

>> Yeah. Okay, so if there's someone

watching this who is like, "Holy cow,

Matt is super advanced and he's built

all these systems." Like, how would

someone start to stair-step their way to

get to the like kind of more advanced

level that you're at? Like, have you

just been doing a process of trial and

error prompting, like just getting

better every day? Is that really like

the solution here?

>> Yeah, I would say that the number one

thing is just just do it. Just uh

give feedback, learn, and start small.

Your system is going to have maybe one

or two workflows and as long as it helps

you, that's already super valuable. It

won't be as, you know, vast and and

complicated as we framed it here. It

should not feel at so daunting that you

can't start. Like I said, I built the

original thing with the thing itself,

right? I told Cloud Code, "I want to

build an operating system. Let's get

started."

Um so, start there and you'll you'll

already be an intermediate at the end of

that session cuz you'll have something

that is unique to you that works for

you, that maybe doesn't follow exactly

these these guidelines, but at the end

of the day, every person in the role is

unique. Uh and that's the beauty of the

system as well, right? Like, it's meant

to mold and adapt to the individual and

that's what makes it powerful and great.

And then uh the second tip I would say

is um X is amazing specifically if you

curate your algorithm.

So, you know, go and find you know

people that are writing about this

stuff. There's a ton of amazing

articles.

I bookmark a ton of stuff on X through

like two or three little time slots I

have in the day and then like before I

go to bed they're fed into read wise and

then I just read through these posts and

I pull out some ideas. I go and test

them the next day and it's all trial and

error, right? Like a lot of what's

written you know everyone that's saying

that they're using their open cloud to

become millionaires is lying to you.

So, take things with a grain of salt but

then you know, take that and maybe try

it on your own. And there's some things

that worked some things that didn't but

it's

from the intention that matters and

that's how like how I fill my day, how I

experiment with everything

and how I got to the place that I'm at

now.

>> Well, I have learned so much from you in

just the time that we have spoken now

and then some of the prior conversations

that we've had and every time I'm like

honestly blown away at the things you've

built and just kind of the systems

architecture and how you think about

about everything. So, thank you so much

for coming on and for sharing some of

your knowledge with us.

Any of the skills you can package up and

send our way, we'll put them in the show

notes or the description here and truly

final final question for you of if

people want to follow you or find

Whisper like where can they find you?

>> Yeah, you can find me Matt Sullinski my

first and last name no space on X,

LinkedIn and Instagram. Probably most

active on LinkedIn. X is right now I

mainly digest will be more active with

with time but that's the best place to

find me.

>> Awesome. All right Matt, thank you so

much.

>> Pleasure is all mine. Thanks guys.

>> That's Matt Sullinski head of growth at

Whisper Flow. We'll drop his LinkedIn

page in the episode description. If

you're listening to this just a heads up

that we put all of these on the official

profound YouTube page. And if you're

watching we're on all the podcast apps,

too. If you want to go deeper on this,

we have a ton of resources for you on

our website, trymarketingengineer.com.

There's a free Marketing Engineer

University, case studies, our manifesto

on why every company on the planet will

hire a marketing engineer this year, and

so much more. I'm Nick Lafferty. Keep

building, and I'll see you right here

for the next episode of the Marketing

Engineer.