The general theory was that the more

information you gave an LLM, the better

the [music] output would be. More

information does not equal better

information. You're actually degrading

the memory of the model. And so that's

why they call it context rot.

>> Do you monitor token usage at all or

have alerts if someone's using too many

tokens?

>> We've always been more cautious about

optimizing for efficiency. Let's not try

to token lax.

>> Lucy Hy is the senior content engineer

at Carta. the company building the

infrastructure and connectivity for

private capital. [music] In this

conversation, Lucy shares what context

rot is and [music] how giving more

context to LLMs can actually be a

negative thing. How she took an idea

that she started at Profound's marketing

engineer hackathon and turned [music] it

into a real marketing OS that her entire

team can use. And finally, how token

discipline makes AI sharper [music] and

not just cheaper. I'm Nick Laferdy.

Let's get into it. So your role has

evolved at Carta from more of a content

marketer to content engineer and now

maybe more of a marketing type engineer

at least in spirit. Can you tell me what

that transition has been like for you?

>> Kind of crazy actually. Like it feels

like it's gone at breakneck speed I

think like most people that are working

within the sort of AI space at the

moment. So I joined Carter 3 years and 3

months ago. um started as a content

marketer and yeah was basically writing

everything by hand like spending hours

researching all the topics um working

kind of very closely with our SEO team

to make sure things were optimized and

then trying to localize content as well

from the US for Europe um and it was

really really slow like like maybe like

one article every few weeks or something

and just feeling slightly overwhelmed by

the amount of work there was to do and

not being able to deliver at the pace um

at which I needed to be able to scale

the content operations in Europe. And so

when I um moved across the US team sort

of a couple of years ago, AI started

becoming more of a a thing. There were

kind of the the early days of the

content generation tools like people

were kind of starting to use chat GPT

and then I started to realize that this

was going to become more of a permanent

shift and it was actually something I

was really interested in. So um I had a

super supportive manager and CMO who

kind of backed me to to go and sort of

do the training that I needed to do to

become a bit more technical. And then of

since January this year um I got

promoted into content engineer position

which was really exciting and just like

a great validation of not only kind of

like my sort of professional development

journey but also that Carter was taking

that role really seriously. Um and so

I'm the first content engineer at Carter

which is amazing that they've kind of

created that role. And then yeah, like

literally that was what like 6 months

ago. Um, and then now my role is kind of

almost progressing from like outside of

just pure content remitt to becoming

like marketing agnostic. So I'm working

with like PMM, with events, with DG to

figure out how to like optimize existing

workflows, but more than that kind of

like see what we can do that's like

different and think in sort of new ways

about the way that we that we market to

our customers.

>> That makes sense. And shout out to Carta

for recognizing like the value in you

and your work and then how marketing and

content and marketing engineering is

kind of shifting. It's kind of becoming

more of the future of what your role and

even my my role

>> looks like too. Uh and so as part of

that transition, you're kind of working

more crossunctionally across other

teams. We talked a bit about systems

thinking as well, kind of what that

means. Can you kind of talk about what

systems thinking means to you and how it

applies to your role? Yeah, I think

fundamentally it's

stepping outside of your like dayto-day

and and outside of the silos that we all

too often work in. I mean, Carter is a

big company. It's like 1,800 people and

so like when you've got teams working on

specific things, it's very easy to

forget to make those crossunctional con

connections. Um, so when I started at

Carter, I was on a very small team. I

was part of the Europe business unit.

Um, I was yeah the only content marketer

like now I think they're a team of two

um two to three. So I had to become like

a more of a generalist very quickly. So

I was doing like long form content sort

of like gated assets, social copy

emails, just anything that touched

content. And I think that really helped

me understand like how everything fits

together. And also being able to work

across regions. So we've we've expanded

now. We're in Singapore, we're in Hong

Kong, like we're in the Middle East and

like all over America as well. So, it's

having that like international view um

but also having to get so deep into like

audience messaging and and product

positioning and stuff to be able to

write to the depth that I needed to like

gave me that that breadth and depth. So,

I became kind of T-shaped naturally. And

then when I moved across to the um US

org, I realized how siloed everything

was like because there's there's a team

for events, there's a life cycle team,

there's like a content team and so like

there wasn't one person doing

everything. And so I think because I had

the benefit of being more of a

generalist that helped me yeah take more

of a systems thinking approach like how

does everything fit together and how do

you kind of like make the connections

between the amazing work that individual

teams are doing and ensure that you're

leveraging all of the like the

information the context the the kind of

institutional knowledge that we all have

and helping us like work faster by

making those connections. Yeah, I've

talked to other marketing engineers who

kind of have two approaches to this. One

is you take your old system that you

were doing manually and then you find a

way to use AI to do it at an automated

way. Or other people who think like you

have to tear down the old system and

kind of start from scratch. Like which

of those camps are you in?

>> I think previously I was in the first

camp just because that makes the most

sense, logical sense as humans. It's the

way we've always worked. But I'm

starting to realize that that the second

option is is the way that we need to go.

and our CEO Henry, he's very forward

thinking with AI, which is fantastic and

it's really like inspiring to see

someone at the top who's kind of like um

encouraging everyone to to use AI more

in their daily work. But his his

philosophy on it is that we shouldn't be

using AI like using AI for efficiency is

is kind of table stakes now. We've

actually got to use it for leverage. So

it's instead of looking at your existing

workflows that are linear and kind of

like step by step and then using AI to

kind of like improve some of that, he

says take a step back and think about

the outcome first. Like where do you

want to get to and then almost reverse

engineer that and often using Claude as

a thought partner and helping you

understand like right this is where I

want to get to and like Claude will

often suggest things that you haven't

necessarily thought of. Um, so I think

yeah, like instead of looking at what

you're doing already and trying to do it

faster, better, like more efficiently,

like you say, tear down the whole system

and think, well, what are we not doing

now that we could be doing? So it's like

this kind of like blue sky thinking

almost.

>> That's a lot of how I use Claude, too,

of I one of my most used skill is a

skill that pokes holes in my LinkedIn

posts to find like find arguments like

what am I missing like what what gaps do

I have? What's the obvious as a thought

partner? Like you said,

>> you also told me that you helped with

the roll out of Claude across your whole

company. Can you tell me what that was

like?

>> It was so exciting, but also like

chaotic in the best way. Um, so yeah, it

was we started in March um of this year

and the goal was basically to get 1,800

people across Carter using Claude chat,

co-work, and code within like a 3 to six

week period. Um, and so when I first got

wind of it, I was kind of put into this

early adopters group. Um and it was a

mix of sort of tech leads and business

champions that have been selected from

from across Carter. So I was

representing marketing with um three

other people and we did a bit of a crash

course in like sort of learning to use

the the different interfaces of Claude

and then what we learned from those kind

of like almost like guinea pig sessions

was then put into the broader curriculum

that we then rolled out to,800 people

across Carter. And so like it was very

much like learn on the spot like build

as you're flying the plane. Love that

analogy. Um and it was it was a lot of

fun. Um I think we learned a ton. Um and

like my key takeaway was that these kind

of like trainings have to be adapted to

the level of the audience that you're

speaking to. And so we had some people

that had already started using Claude um

through a CLI rather than like through

the desktop app. So they were like miles

ahead. And so the sessions we had with

them were far more kind of like

practical and like almost like you say

poking holes in things and and really

pushing the boundaries of what we could

do. And then you had some people that

had never touched Claude. And so that

was very much foundational getting them

to understand how skills worked um like

how to prompt effectively. [snorts] Um

and I think the best learning from that

is like how do we then apply that to the

way that we talk to our customers.

Carter operates in a very regulated

industry and and we serve customers from

all the way from like startup founders

to to LPs, investors. Um, and we see

like a broad diversity in how um, our

customers have been like taking to to

AI. Um, I think on the venture capital

side, they're very kind of forward

thinking, very like risk, they're not as

riskaverse. Um, and then on the private

equity side, they're a bit more

traditional. And so, I think like

figuring out like the importance of

tailoring your training to the audience

you're speaking to internally has helped

when we've been rolling out like the

claw plugins and our CLI to customers as

well.

>> Nice. Um, do you monitor token usage at

all or have alerts if someone's using

too many tokens?

>> Yeah, I think we do now. I know a lot of

companies like initially they went down

the route of having these like

leadership like dashboards and like

who's using the most tokens and stuff

whereas I think we've always been more

cautious about like optimizing for like

efficiency. I don't see that it's like

necessarily dampened experimentation.

And I see that it's actually helped us

learn how to aid prompt effectively, but

also be like be really careful with what

we're building and not just kind of try

and apply AI to everything. Like think

critically like okay like do I need to

use AI for this? Like is it going to

give me a better result? Um and then

starting to think about okay well like

if you're using an LLM for the logic of

something can you use something like a

Python script where you've got a more

repeatable process that doesn't need

that logic layer applied. So I think

yeah taking a step back and being like

okay let's not try to token max has

[snorts] been probably more useful for

us understanding how to really work with

AI. Um and I think yeah initially like

we had in the claw desktop app in your

settings you can see like how much

you've used and so we would have like

monthly caps and I think the engineers

kind of had unlimited use obviously but

now they've they've increased it but set

like slack alerts if you're use if

you're kind of spiking. Um, so it's

yeah, it's I it hasn't felt like

monitored. It's more just like

empowering us to make conscious

decisions about the way that we're using

Claude and when we're using it.

>> That makes a lot of sense. And I think

the Slack alert is smart, too. Like you

don't want to encourage people to just

burn a bunch of tokens on things. And

even finding a way of like when do you

use AI and then when do you don't also

seems like a really kind of smart

approach, too, where I see people using

AI for everything. And I don't it feels

very kind of like you have one tool and

it's a hammer and you just like swing it

at every problem. That's not necessarily

the right way to build a nail.

>> Yeah. Exactly. Exactly. Okay. So a few

weeks ago you uh participated in

profounds hackathon. Can you talk about

what that process was like and then a

bit about what you built there too?

>> It was awesome. Um completely

exhausting. Like I think the the next

day I slept for like six eight hours cuz

my brain was just exploded. But yeah, so

I think the the question that we kind of

had to work towards was like take a like

a process um that is inhuman in scale or

scope and use AI to basically attack

that problem. Um and so like the

diversity of what people built that day

was so cool to see. Um I mean like we

were standing at the back when people

were presenting we're going this is so

cool like how does that work? And it was

just great. The energy in the room of

like people kind of going around and

seeing what other people were building

and like people flew in from all over

the world, right? Like it was crazy for

me. The experience was was useful

because of those connections, but also

it showed me like what I can actually

achieve if I kind of time box stuff and

I have like a set amount of time to try

and get from like an idea to MVP to

something that's good enough that I can

present it but not perfect. And I think

that's something that I've struggled

with and a lot of people that when we

were doing like the clawude roll out at

Carter and and various hackathons we've

done internally, people try to almost

like boil the ocean and do too much

because AI gives you the ability to

pretty much do anything. And so it's

like where do you start but where do you

stop? Um and so I kind of got partway

through the hackathon. I remember you

coming over to me and saying like how

you getting on and I just looked at you

like no [laughter]

I can't do it. And at that moment I I

decided to just like pause and be like,

"Okay, I can't do everything today and I

only need to do a few things to get like

a working model." And so I then honed in

on a very specific use case like instead

of kind of Yeah. uploading a ton of

information. I I handpicked and just

built something that I could actually

test against and it worked and that was

all I needed.

>> Cool. Okay. So tell tell us about what

you built.

>> Um so it was essentially like a

marketing brain. The premise of it was

that issues that we've been seeing pop

up at Carter since we started the claw

roll out is that everyone got super

excited and started building skills

plugins like left, right, and center.

And we didn't really have a a sort of

like consolidated way of sharing those

and making sure that they were kind of

all built according to best practices.

And so we've been [snorts] using like a

shared Google drive, but then you have

version control issues where like you

have to kind of download the skill from

the Google Drive, install it into your

own cloud instance, and then if someone

updates it, you've got to update it

again. And also we were seeing like the

context that the skills that Claude uses

to when it runs a skill was almost

hardcoded. So it was living in the sort

of the skill.md file and [snorts] then

that obviously causes version control

issues and everyone's kind of pulling in

their own context like if if context

changes you've got to re-upload that and

so there was no single source of truth

across the skills that we were building

in marketing and so something that I

wanted to do was to see if we could like

build like a sort of shared context

layer um which I termed the marketing

brain and then build like skills and and

scripts that could reference that

context layer and yeah I managed it

which is super exciting.

>> That's brilliant. And so you how much of

that did you get done on the day and

then I guess what have you done with it

since then?

>> So I got like a a very c kind kind of

small subsection working on the day. So

I built I used plan mode in code to kind

of map out the whole thing. Um and then

made a a few tweaks as I went. Um but

that kind of ensured that I was like

following a a consistent track and and

didn't get completely distracted by

trying to do other things. I use

profounds agents to generate like

product um messaging documents. So the

use case I picked was we had to pretend

that we were working for OpenAI and a

lot of the AEO data we had was on like

chat GPT compared to like Claude,

Perplexity, other kind of LLM tools. And

so I said, okay, I'm going to try and

kind of build a competitor takedown

campaign putting chat GPT against Claude

and Perplexity. Um and so that was yeah,

my super narrow use case. Um and so I

pulled some um AEO my AI visibility data

from from profound um connected profound

to my cord instance via API um and then

was basically able to pull in all the

data that was showing up in my

dashboard. And so that was my first kind

of like bit of the marketing brain that

data that was coming in from profound

and then I then used those insights to

build like product messaging documents

for chat GBT. And it was it was amazing.

Like the agent I built was able to yeah

like like search the web, pull in like

all of the sort of competitor intel from

profound. Um and then I gave it a

template and it mapped um the

information it had pulled to the

template that I wanted it presented in.

That obviously then went up to the

marketing brain and I built out a git

repo to store this kind of context layer

um because I wanted it somewhere that

would be able to be updated and shared

with other people and somewhere that was

kind of like safe as well. So we have

like an internal Carter um git repo that

we're going to start rolling out in a

couple of weeks as well. And so those

context files lived in there. Um and

then I as I say used claw code to plan

and build the whole thing. Um and I

once that context layer was was started

to be built out, I built a like a data

filter. Um and we've we've talked about

this quite a lot like the the importance

of kind of avoiding context rot. Um, so

in order to stop Claude pulling all of

the information down um, in one go and

kind of like hitting the context window

immediately, I realized that I needed to

build in like a bit of a buffer so that

it could kind of like filter the

information it was pulling down.

>> Okay. Well, I want to talk about that in

a second. And the git repo, we'll come

back to um, the agent you built for

this. I think we'll be able to put that

in our agent marketplace that that we

just launched. And so if that sounds

interesting to anyone, like that'll be

available. We'll try and link it in the

show notes below. Okay. So, talk to me

about context rot and kind of how you

kind of worked around that problem.

>> Yeah. So, it's something that I wasn't

really aware of until as part of our

cord roll out. Um, we've been doing

these like learning labs like AI

learning labs and one of the ones that

one of our European engineers ran, he

was talking about context raw and I

think everyone came away from that being

like, "Oh my god, we've been doing

everything wrong." Like the when people

started prompting and chat GBT like a

year ago or whatever. Um, I think the

the general theory was that the more

information you gave an LLM, the better

the output would be because it would be

more specific. Since then, I think we've

learned that like more information does

not equal better information. So, it's

like you have to be more specific and

more targeted with what you tell an LLM.

So, the premise of context raw is that

every time you provide information to an

LLM, it will basically load all of that

information plus its own like memory.

And so you you get to a point where you

see the little like wheel in the bottom

right hand corner of a claude session

and [snorts] every time it kind of like

hits its max, it'll almost like stop the

conversation and then try and condense

it and and start again. And so every

time that happens, you're actually

degrading the memory of the model. And

so that's what why they call it context

rot. And so the flip side of that or the

solution to that is context engineering.

And this term engineering obviously

keeps coming up left, right, and center.

But it's the idea that like similar to

prompt engineering where you have to be

yeah very careful and very critical with

the way that you prompt the same applies

to the context. So it's like what am I

giving Claude and like can I give it

information in bits so that it's able to

kind of like chunk the information. Um

so I think that's called progressive

disclosure.

>> Okay. And so is you were able to build

that within the marketing brain that you

you were building at the hackathon.

>> Yes. So that was the purpose of the data

filter. Um, so I knew that like if I had

this kind of context layer living in

Git, I could put so much in there and if

Claude was trying to access all of that

information every time I ran a skill, it

was just going to max out the content

context window straight away. And so I

thought, I've got to build in some sort

of buffer. And so I used I did a bit of

research on Python scripts. Um, and I

found one called like an open- source

Python library called BM25, which is

actually used in elastic search, which

was seemed perfect for what I wanted to

do. So essentially when someone runs a

skill or or sort of starts a

conversation in Claude um Claude will

kind of like analyze the prompt and then

pick out key words from that. It will

send it up through the BM25 script.

Those that Python script will match the

keywords to context that lives in the

marketing brain and then it will pull

down the relevant chunks of information.

So Claude is only using the context that

it needs for that specific use case.

>> That's really smart. Yeah. And we'll

we'll find that and we'll put that in

the show notes too because it's open

source an open source Python script

basically.

>> Yeah. Yeah. And it was super easy to

use. I thought I was going to have to

like do loads of implementation but I

mentioned it when I was using plan mode

include code and then it and I gave it

the um like the the correct links to

kind of go and find the information

online and it it basically found

developer docs and stuff and and

installed it on my Mac. So yeah, it runs

like a dream.

>> It's so good.

Um, okay. So, I I want to come back to

kind of like the broader reason you you

built this and kind of what's happened

since then. So, you had this idea, you

started it at the profound hackathon

>> and then I like what happened after that

like you went back to work and how are

you implementing what you started there

in your day job?

>> Yeah, so um I can't take full credit for

for the idea. It's actually been kind of

the result of a few conversations I've

been having um at work when we started

the the crawled roller. I think we got

to the point where the skill sharing was

becoming an issue, but we were also

realizing that like the context, the

information was just like we needed to

find a way of actually better

controlling um what was going what we

were feeding Claude and making sure that

everyone was kind of speaking from the

same source of truth. And so we had an

amazing offsite in Nashville for the

whole marketing team at the end of May.

and um our like revenue team um and our

like digital marketing team, a few of

them presented on their idea of like the

the context layer um and then how it

would fit into um like a decision or

execution layer and like our data core

and so the the thinking was there um and

then I kind of took that took it brought

it to the hackathon wanted to see if I

could create like a working version of

it and since then I had a couple of

colleagues fly in from SF last week and

we literally locked ourselves elves in a

meeting room for like 2 to 3 days and we

were like let's just hash this out and

it went from us all having slightly

different ideas of how this would work

to a really consolidated what we're

calling the marketing operating systems

the marketing OS for Carter and like a

clear plan to implement it and we

presented that marketing all hands on

Wednesday and it went down really well

so I'm so excited to actually like start

making this happen

>> that's so cool and so is there a

timeline for when you'll work on it and

it'll be done or the first version Yeah,

I think we we want to try and move

quickly with it, but we're also aware

that again there's two approaches.

Either you take it like step by step and

you think, okay, let's let's do it on a

campaign by campaign basis. So, one of

our offerings at Carter is QPS. It's a

sort of like tax efficient um system for

for equity. And we were like, okay, do

we start with that as a very small use

case? Do we get all of the information

from across Carter that relates to that

like QPS and then do we kind of build on

that or do we go the other way and we

try and get all of the teams across

marketing to collect all of their data

and then we upload it all at once and

then we might get to this point where we

kind of don't know what's worked, what

hasn't worked. So I think we're like

still teasing out like how that's like

the way that we're going to approach it.

But we've got meetings booked in like

discovery meetings with specific teams.

So we had the brand team last week.

We've got um DG team today and we're

basically like meeting with them to try

and understand like what types of

information they use in their daily work

but also what types of information they

produce. So like if demand genen run a

campaign, what is the data that they get

from that that could be useful for other

teams? And so yeah, we're kind of

mapping that out. At the same time,

we're doing the skill curation work. um

we've decided on a committee um that's

basically going to look at all of the

skills that people have built across

marketing and decide if they can be made

more crossunctional and then if they

kind of pass that committee stage

they'll get uploaded into the git and so

everyone can access them. So it's yeah

it's [laughter]

trying not to boil the ocean and trying

to do it methodically while also moving

very quickly and like biasing to action

is a real challenge. Um, but I think

like now that we've made inroads and and

like we're communicating regularly with

our CMO Nicole and obviously using a lot

of Henry's philosophy to to make sure

we're on track and and getting feedback

at every stage from different people

across marketing. Um, and then when we

presented it like I think we've got the

excitement going. So like it's yeah, we

just got to keep that ball rolling.

Yeah, I think I think maybe every head

of marketing watching this is probably

fascinated with what you're building

because it seems like the dream like a

true marketing OS with shared context

across different teams and skills, what

everyone's doing, take the outputs and

kind of have a more centralized place

where everything just kind of lives and

uh is is truly centralized uh which I

think is really really amazing. We'll

have to have you back in Yeah, I'm sure

I'm sure it'll work. We'll have to have

you back in a bit to to get out there on

how it's been and the roll out and all

that stuff. is part of that you've kind

of had to learn Git as well. Uh what has

that process been like for you?

>> Yeah, so I I feel like I probably

haven't gone as deep on it as I as I

could have. And

part of the reason for that is that like

I'm trying not to overwhelm my brain

with all of the things that I could

learn. And maybe I'm a little bit late

to the the party with with Git. Like

obviously you've been using it for like

what 10 years or something. Um and I was

never that technical. So, I think the

getting to grits with with Claude and

everything has kind of shown me that

like I need to know enough to like get

by and to teach other people without

trying to go like too deep on everything

because I will just get overwhelmed. So,

we're really lucky that we've got like

an amazing engineering team at Carter

and some of them are tasked with

building out like internal

infrastructure and so they've set up

like they were exploring various options

for for the skill sharing. As I say, we

started with Google Drive. That wasn't

really working. And so now they've

decided Git is the best place for us to

store that, but we were hitting an issue

where we ran out of licenses for for

GitHub. And so they've built like an

internal Git server, which is fantastic

cuz it's kind of secure. You haven't got

people creating their own like personal

Git accounts and sharing company

information. And so I think yeah, as I

say, we'll be rolling that out properly

in a couple of weeks. Um, and I think

they've created like a TLI command um

that you can literally just like yeah

use within within CL code um and give it

the um SSH to your specific git repo

[snorts]

and then yeah you you create this

amazing feedback loop where crawl is

able to pull information from the git

repo but also pass it back up and I

think that's another key aspect of the

marketing OS is that those learnings we

get from every time we run a skill or

run a campaign through claude like it

will ask the user user if they want to

kind of commit that information back up

to the marketing brain or the context

layer. And so it's like a the system

learns from itself which is really

fascinating.

>> Yeah, I think that's the way to go too

is I think when you learn git all you

really need to know is the like what it

does like you can like push and pull and

what a pull request is like once you

know the right terms you can tell claude

and then claude can kind of do

everything else for you. I have it pulls

me

GitHub is smart enough to where it'll

tie your GitHub username and Claude to

it. So

directly for you to the repo which is

really cool.

>> Yeah, it's super exciting and and also

like one [laughter] of the the things we

say most often at Carter and we should

probably create a swearbox for it is

like have you asked Claude? Like if

anyone has questions, technical

questions like that, the first place you

should go is is Claude because it can

obviously scrape the web. there's just a

wealth of information that it can kind

of consolidate and um if you say like

explain this to me like I'm five then

you get a really kind of logical

non-technical answer. Um and so yeah, if

you're not sure how to use Git or how it

works like Claude is the best place to

start.

>> Yeah. Um that was actually my next

question for you is how do you stay

current? How do you learn and stay up to

date on all everything AI? partly like

LinkedIn is really powerful for this and

I think once you start making

connections then that like information

starts being woven through comes up in

your timeline like I get most of my

information from LinkedIn I think and

then I pass it on to my colleagues and

they do the same so it's this really

nice like living breathing channel and

then like inerson events I think since I

moved to New York 2 3 months ago um I've

been like overwhelmed by the amount

that's going on in this space um and I

think like getting to know you has been

really helpful cuz you'll invite me to

events I'll kind of like invite you to

stuff. And then yeah, just just meeting

people that are doing [clears throat]

the same thing but also having the same

struggles. Um, and I think what's so

nice about this community, whether it's

like the SEO, AEO community or marketing

engineering, whatever you want to call

it, is that people are like so willing

to share their knowledge. And like I

don't know if you found the same, like

>> it's yeah, it's just such a it feels

like a safe space. Um, and the stakes

are pretty low just because

I think with SEO it felt often quite

intimidating cuz you'd had people that

have been in the industry for like 10,

20 years and you're like, how do I ever

get to that point as someone who's kind

of like fairly fresh in my career

whereas now like we're all starting from

the same point and everyone's bringing a

different perspective. Um, and it just

feels like a like it feels like the

center of gravity at the moment.

>> It does. Yeah. And there's there's so

few people who are doing this right

right now. so few people who are current

marketing engineers or want to become.

And that that number is definitely

growing, but there's maybe one or two at

every company right now. And so if you

want to learn, you have to look outside

and find other places. And some of the

meetups that you and I have been to have

been really helpful because you learn

that everyone's kind of attacking the

same problem at the same time. Like

we're all stuck on this thing or finding

the right context layer or context raw.

And it's really validating to get to

that point and say, "Oh, like we're all

in the same position right now."

>> Yeah, 100%. and and hearing from like

leading companies that are that are

having the same problems. We were at the

multi-engineer um meetup that was I

think put on by um work OS and NOG um

last week and hearing from people at

RAMP versel um cursor they're having

similar issues in terms of like

skillsharing and and trying to

implement more conformity like while

rolling out AI to to a sort of fairly

big company and then also the way that

they're thinking about solving that and

and the the progression that they've

been on. So I think um one of the knock

um team was saying that they'd started

with like the idea of sessions. So kind

of working in like chat conversations

sort of like isolated tasks and then

moving on to projects. Obviously, a lot

of this is very like clawed terminology.

Um, but thinking, well, how do I work

within like a shared area of context and

kind of like build conversations on top

of each other and then moving from that

to systems and I think seeing that

mapped out on a slide and I looked at my

my colleague from Carter and I was like,

yeah, this is what we're trying to do as

well. It's that idea of the system that

that everyone's working towards.

>> Amazing. Yeah. And I think if you're

interested in attending more local

meetups that profound will have some

exciting updates to to share there

pretty soon. So amazing.

>> Okay. So I guess last question if you're

someone watching this and they're like

wow Lucy how you moved from content

marketer content engineer like I want to

go do something like that too. What

advice would you give those people?

>> There's no better time to be doing

something like this. It's obviously a

really like thriving space and so like

now is a great time to consider like a

career transition whether you kind of

want to move externally or or maybe

bring it to to your manager and say like

this is this is something I'm really

interested in. I think proving the value

of it is often where people get stuck

and so like speaking to to companies

like profound or um Carter ramp where

they've um started like doing this

internally and and trying to figure out

how they got exact buy in. Um, that's

like the crucial thing. And then once

you've kind of been proactive yourself

and pushed it from the bottom up and

then you've got that leadership buy in,

it's so much easier to to kind of like

go full steam ahead.

>> I I totally agree. Someone asked me a

similar question at our zero click

conference in New York last week. And I

think my answer to them was something

around take something you're doing right

now and just use build AI around it. And

>> so take like kind of like the V1

approach that we talked about earlier of

just like take a current process and

just find a way to do it with AI and

then find a way to talk about that when

you're in an interview or when you go

talk to your boss or your manager

afterwards. And then the more you can

show your work, I think the more easier

it is to get buyin for ultimately if

it's a title change, career promotion,

whatever. But like build something and

then talk about it whether it's

internally or on LinkedIn or Twitter. I

think that's generally the secret to

things is like build cool stuff and then

talk about it.

>> Yeah. And I think like the talking about

it is probably more important than the

building. And again coming back to this

idea of like presenting to your audience

like think about who you're trying to

show this to because you've gone really

probably quite deep and technical and

you're really excited about this thing

you've built. But then when you take it

to your manager or an exec, you need to

think like I need to the way that I show

this to them. I need to start with

impact and the why before I go into

loads of detail about how it works cuz

they're probably likely going to be

interested in the first two. And then if

they're like really technical

themselves, they might ask you questions

about how you built it. But I think the

the kind of instincts that when we're

like really excited about something to

be like, "Oh my god, this thing's really

cool." And then they're like, "Okay,

take a step back. Like, why does this

matter to me?" I think that's that's the

key thing that I've learned like a lot

over the past few months like kind of

presenting to to more senior people um

about what I built is like what's the

narrative what's the takeaway for them

um yeah and I think like once you've got

that kind of like relationship with an

exec and they kind of see that you're

someone who's proactive and and wants to

build um you can kind of go from there.

>> Yeah. Like I think your CMO doesn't care

about BM25 but they care about the why.

Like why are you doing this in the first

place? What time savings will you get?

how it will impact the whole org, the

whole like bottom line upfront thing and

how you talk to execs I think is one of

the best skills that marketers can learn

and no one really teaches you this

unless you have a great manager who kind

of coaches you you through it. A lot of

your experience kind of learned that way

too of just trying and figuring things

out with them.

>> Yeah. Yeah. 100%. And like as they say

like if they want more detail they'll

ask for it. And so Nicole, our CMO,

she's she's amazing because she

obviously has to have this like really

kind of like broad like agnostic view of

of what's going on in marketing and and

map that to to the rest of the company,

but she is also able to kind of go quite

deep on the detail when she wants to.

And so when we were presenting our

marketing OS idea to to her and to the

rest of the marketing team, we went

quite high level. And then like she sort

of said to us on the side like I I am

interested in like how you're building

this and the technical details. So I

said to her, do you want me to send you

like a demo of of what I built with

Aquafon? She was like 100%. But it's

like don't start with that. Like let

them almost kind of come to you and be

like, oh no, I am interested in this.

And once they've kind of got the the

highle view, then they can start digging

in, which is really powerful, I think.

>> Do you have any tips for someone who

wants to improve how they talk to

executives? Yeah, I think um aside from

really

reframing your language and and almost

treating them as like a totally separate

audience to how you talk to your team or

how you would talk to like the rest of

marketing or the rest of the company.

Think about like what they care about.

And so it's it is difficult nowadays to

kind of find data to support something,

especially where we've seen like a

massive drop off in like organic traffic

and click-through rate. Um, and so it

can be quite hard to find data to kind

of support the the argument you're

trying to make, especially if you

haven't actually built anything yet.

It's like how do you get data about

something that doesn't exist yet? And so

quite a useful strategy is to think,

well, what is the cost of inaction? Like

what are we losing by not doing this

thing? And then you can get data from

like other companies and and then like

apply that to to your own company. Um

and I think yeah something I heard at a

conference sometime last year was um

taking search data or um kind of like

revenue data from like a comparable

company if you can find even if they're

a public company um and then applying

that to your company size like the

industry that you work in and then

almost like using not dummy data but

kind of like comparable data to show

your exacts okay this is what this

company did this is what they gained

from it like this is what we're losing

by not doing the same thing. Um, that's

quite a useful strategy because I think

that like operating on fear isn't the

best principle, but it's that like yeah,

the cost of inaction, do it or die kind

of thing.

>> Yeah, I've had that same conversation

with companies about writing comparative

articles like us versus our competitor

content. And I've been able to show like

if you don't write this and your

competitor does, guess what will get

picked up or cited in LLMs? It's their

POV, their perspective on how they're

better than you. And so the cost of not

doing that is you let them control the

narrative.

>> Yeah. They get there first.

>> Yeah.

>> So powerful.

>> Great. Well, Lucy, this was brilliant.

Thank you so much for coming on. This

was an honor. Thank you again.

>> Thank you for having me.

>> Of course. That was Lucy Hy, senior

content engineer at Carta. We'll take

everything that she discussed, including

agents, workflows, and put them in the

description for you to download and take

away. We'll also put a link to her

LinkedIn profile so you can give her a

follow and learn from everything she

posts [music] online. You can also go

visit tryp profofound.com where you can

read our manifesto on the marketing

engineer and get resources for how to

become more fluent in AI in your job.

I'm Nick Laughy. Keep building.