Episode 73

full
Published on:

6th Mar 2025

73. Building an AI First Organisation Part 2

In this episode we discuss: Unlocking AI for Non-Technical Leaders. We are joined by Charlie Cowan, Founder of Kowalah and author of "How To Sell Tech" and “The Revenue Operations Playbook”.

Love The Operations Room? Please support us by rating and reviewing it here.

We chat about the following with Charlie Cowan: 

  • Is AI more accessible than we think?
  • What’s the difference between AI agents and workflows—and why does it matter for businesses looking to scale?
  • How can organizations build their own AI-powered tools instead of relying solely on SaaS solutions?
  • What role does leadership play in AI adoption, and why is AI literacy now a must-have skill for future executives?
  • Can a simple “power hour” or habit shift really transform your productivity—and what does AI have to do with it?

References 

  • https://www.linkedin.com/in/charliecowan/
  • www.charliecowan.ai
  • www.kowalah.com

Biography 

Charlie works with CEOs and senior leaders looking to embed AI behaviours across their organisation.

With a 25 year career in SaaS, consulting and enterprise sales, Charlie bridges the gap between AI technology and practical applied use cases that help teams work smarter and faster.

Charlie is the author of two books - How To Sell Tech and The Revenue Operations Playbook.

Charlie is the founder of Kowalah, an AI-powered platform that helps buying teams to run a great buying processes, pick the right vendors and reduce the fear of messing up.

Charlie built Kowalah as a solo-founder using AI development tools.

To learn more about Beth and Brandon or to find out about sponsorship opportunities click here

Summary

26:36 Introduction to AI Adoption for Non-Technical Executives

027:16 Understanding AI Optimism vs. Pessimism

28:08 Defining AI Policies for Effective Implementation

29:52 The Concept of Agentic AI

31:23 Distinguishing Between Agents and Workflows

33:01 Building Effective AI Agents

37:49 Evaluating Future Tech Stacks for AI Integration

43:05 Empowering Leadership in the Age of AI

46:06 Navigating AI Tools for Policy Management

46:23 Crafting Effective Prompts for AI Tools

49:54 Building a Data Repository for Enhanced Insights

51:44 The Power Hour: Maximizing Productivity



This podcast uses the following third-party services for analysis:

Podcorn - https://podcorn.com/privacy
Transcript
Speaker:

Hello everyone, and welcome to

Speaker:

another episode of The Operations

Speaker:

Room, a podcast for CEOs.

Speaker:

I am Brandon Metzinger and I am

Speaker:

joined by my lovely co-host Bethany

Speaker:

Air is having some nice porridge.

Speaker:

Breakfast is what I want.

Speaker:

I'm looking at.

Speaker:

If only it were porridge.

Speaker:

Oh, it's not porridge.

Speaker:

Looks like porridge.

Speaker:

No, this is the chia seed concoction

Speaker:

that, thanks to Zoe, I eat every

Speaker:

day. From chia seeds,

Speaker:

other seeds, kefir,

Speaker:

milk, lots of nuts.

Speaker:

And therefore every

Speaker:

day.

Speaker:

Right. So every trendy ingredient

Speaker:

that I can imagine is on the list.

Speaker:

Pretty much.

Speaker:

So didn't you tell me before that

Speaker:

the taste of your dish

Speaker:

there, the cheesy delight, is not

Speaker:

great or it's rather plain.

Speaker:

Yeah. It's plain.

Speaker:

I have some strawberries in there

Speaker:

that helps the nuts help

Speaker:

it, but it's a bit of a slog.

Speaker:

And also because it's so

Speaker:

chewy, it requires

Speaker:

so much chewing

Speaker:

that it takes me absolutely forever

Speaker:

to eat it.

Speaker:

I don't know how much reading you've

Speaker:

done about ultra processed food.

Speaker:

So one of the things

Speaker:

with ultra processed food is it's

Speaker:

very soft.

Speaker:

And that has made our

Speaker:

jaws stop growing properly.

Speaker:

I've not heard this before.

Speaker:

Is that true?

Speaker:

Yeah. So children

Speaker:

and we were raised on enough ultra

Speaker:

processed food to make the

Speaker:

difference.

Speaker:

All of our jaws are more

Speaker:

recessed, and the reason why we need

Speaker:

our wisdom teeth taken out is.

Speaker:

Apparently your jaw grows

Speaker:

through hard work as

Speaker:

it's forming as a child.

Speaker:

And so it actually comes further

Speaker:

out. And then you have space

Speaker:

for your wisdom teeth.

Speaker:

But with ultra processed food, it's

Speaker:

really soft. So our jaws don't work

Speaker:

anymore. And that's why everybody

Speaker:

needs braces.

Speaker:

Oh my goodness. I did not know this.

Speaker:

Well, I would say this I ate a

Speaker:

tremendous amount of cashews.

Speaker:

And cashews are quite hard, so I

Speaker:

suspect my jaw line is

Speaker:

kind of being exercised.

Speaker:

From the outside, it looks like your

Speaker:

jawline is fine, but I don't know.

Speaker:

Did you take your wisdom teeth out?

Speaker:

Could you have had more of a jaw?

Speaker:

Had you

Speaker:

knocked everyday?

Speaker:

I don't even know.

Speaker:

I literally have no memory or

Speaker:

recollection of having them taken

Speaker:

out, so they must still be on.

Speaker:

Right. And came in in a normal

Speaker:

way. Like they didn't cause any

Speaker:

problems. You didn't need anything.

Speaker:

My superior jawline was

Speaker:

sufficient to house them.

Speaker:

All those cashew nuts and the clean

Speaker:

living of Canada.

Speaker:

So besides

Speaker:

the chia seed.

Speaker:

That's right. The ultra processed

Speaker:

food.

Speaker:

What is happening in Bethany's

Speaker:

world?

Speaker:

Just insanely busy.

Speaker:

So I've been working from home

Speaker:

this week. I managed to leave the

Speaker:

house on Wednesday and

Speaker:

now today it's Friday.

Speaker:

We're doing this at 8 a.m..

Speaker:

I have back to back meetings

Speaker:

like literally back to back in my

Speaker:

calendar. Not a sliver of

Speaker:

space until half six.

Speaker:

So luckily my husband is working

Speaker:

from home today and so he's going to

Speaker:

bring me some lunch around noon

Speaker:

so that I could continue to eat,

Speaker:

although at the rate it takes me to

Speaker:

eat the chia seeds, I might not

Speaker:

actually have finished them by the

Speaker:

time lunch arrives.

Speaker:

Okay, so that's fabulous.

Speaker:

So you have the in-house servant, in

Speaker:

this case being your husband to

Speaker:

deliver your lunch for.

Speaker:

Why not call him that?

Speaker:

That's that's I mean, this is the

Speaker:

give and take in a relationship, so

Speaker:

I don't know either. A quick note, I

Speaker:

just had this small thought this

Speaker:

morning. So I'm working with a

Speaker:

couple individuals right now that

Speaker:

are they do operations.

Speaker:

But really the granular operations

Speaker:

if you want to call it that.

Speaker:

One of the individuals was asking

Speaker:

me for a recommendation for

Speaker:

employment law.

Speaker:

So I just did a quick screenshot

Speaker:

of recommendations coming

Speaker:

out of the Operations Nation slack

Speaker:

channel to her that operations

Speaker:

Nation slack channel is good for

Speaker:

this kind of thing, which is vendor

Speaker:

sourcing recommendations around

Speaker:

products and people and that sort of

Speaker:

thing is fabulous for that.

Speaker:

People are very

Speaker:

proactive. I would say, in terms of

Speaker:

giving thoughts on who to speak to,

Speaker:

who's good to, you know, to work

Speaker:

with these sorts of things.

Speaker:

The other thought was just the

Speaker:

obvious one, which is she's an

Speaker:

operator. She's at that level.

Speaker:

She'd be perfect to be part of the

Speaker:

operations nation, or at least on

Speaker:

the slack channel, is just a good,

Speaker:

useful tool.

Speaker:

When these questions pop up like

Speaker:

this that are quite straightforward,

Speaker:

where people can respond back

Speaker:

and give you the kind of collective

Speaker:

set of good answers.

Speaker:

And it's free.

Speaker:

It's definitely a good community for

Speaker:

like having problems with my pension

Speaker:

provider or

Speaker:

the health insurance quote

Speaker:

has gone up by 40%.

Speaker:

What do I do?

Speaker:

Or is that the same for everyone

Speaker:

else? Like those are the types of

Speaker:

questions you see.

Speaker:

I went to a comedy show on Sunday

Speaker:

night and then a comedy show on

Speaker:

Wednesday night, and I was telling

Speaker:

a friend and he was just like, is

Speaker:

this a new thing?

Speaker:

Like, it's going to comedy shows

Speaker:

you're saying?

Speaker:

And said, in that kind of like

Speaker:

snide, dismissive way,

Speaker:

like, what is up with you?

Speaker:

It's not our new thing.

Speaker:

And I actually booked them ages

Speaker:

ago, maybe July, August time.

Speaker:

And it just happened to arrive now.

Speaker:

And there were two very different

Speaker:

comedy shows. So one of them was in

Speaker:

Hackney.

Speaker:

It was very trendy.

Speaker:

The guy is a Sri Lankan

Speaker:

comic, as you do

Speaker:

a Sri Lankan comic that kind of

Speaker:

looks like Jesus, like he is very

Speaker:

curly hair down to his waist.

Speaker:

Very good looking guy.

Speaker:

And he studied

Speaker:

medicine and then decided not to

Speaker:

become a medic, and

Speaker:

then became a developer and

Speaker:

a comic and is trying to be a comic

Speaker:

more than a developer.

Speaker:

So anyhow, I

Speaker:

have never felt like more out

Speaker:

of place.

Speaker:

Everybody was trendy.

Speaker:

Like, I don't think there were any

Speaker:

other women in the room.

Speaker:

They were just non-binary.

Speaker:

I guess we call them women anymore.

Speaker:

Like, I would have chosen to be

Speaker:

non-binary had I had

Speaker:

the option, because you kind of look

Speaker:

at it, you're like, oh, I could be a

Speaker:

woman, and that looks like a really

Speaker:

bad deal, or I can opt

Speaker:

out

Speaker:

and I think I'm going to opt out.

Speaker:

Yeah.

Speaker:

Fair enough. Yeah.

Speaker:

And then every man had a mustache.

Speaker:

Is that. Is that a thing right now?

Speaker:

It really is a thing.

Speaker:

And here we are like middle

Speaker:

class clap mites

Speaker:

nots with mustaches

Speaker:

and combat boots.

Speaker:

And I was just like.

Speaker:

And not non-binary.

Speaker:

Yeah.

Speaker:

And I just never felt so uncool in

Speaker:

my life.

Speaker:

But it was quite a good show other

Speaker:

than just feeling like such

Speaker:

the square.

Speaker:

And then on Wednesday

Speaker:

night, it was a

Speaker:

Irish comedian who's a couple years

Speaker:

older than me. Woman.

Speaker:

And other than the fact that at

Speaker:

least 90% of the audience were

Speaker:

Irish, everybody was a middle

Speaker:

aged woman and maybe

Speaker:

one and four brought their husbands

Speaker:

along. So I felt

Speaker:

totally in

Speaker:

my element. We all belonged.

Speaker:

We were all square, we

Speaker:

were all 50 ish,

Speaker:

and it was quite good, but totally

Speaker:

different sets.

Speaker:

One was around

Speaker:

like his was around not belonging

Speaker:

and racism

Speaker:

and privilege.

Speaker:

And hers is about

Speaker:

menopause and sex.

Speaker:

And husbands and children.

Speaker:

Congratulations though to you.

Speaker:

Because actually booking kind of

Speaker:

not social but like just outings

Speaker:

of fun and enjoyment.

Speaker:

You know, I feel like my calendar

Speaker:

right now is pretty sparse in that

Speaker:

respect.

Speaker:

So I'm always actively looking

Speaker:

for, I don't know, just interesting

Speaker:

things that would be, I don't know,

Speaker:

fun, enjoyable to do and were doable

Speaker:

just given our schedule with the

Speaker:

kids and all that jazz.

Speaker:

But nothing has really popped up

Speaker:

quite recently. That seems useful

Speaker:

enough or good enough to go to.

Speaker:

All right. So we have got

Speaker:

a great topic for today,

Speaker:

which is a returning topic, as

Speaker:

it were, building an AI for

Speaker:

Organization Part two.

Speaker:

We have an amazing returning guest

Speaker:

for this, which is Charlie Cowan.

Speaker:

We had so much fun with Charlie that

Speaker:

we had a second conversation.

Speaker:

And of course, he is an AI

Speaker:

strategist and AI change

Speaker:

agent within organizations

Speaker:

today. So before we get into our

Speaker:

part two with Charlie, Just wanted

Speaker:

to talk about a couple bits and

Speaker:

pieces here.

Speaker:

One of the elements that he spoke

Speaker:

about was leadership.

Speaker:

Using AI themselves

Speaker:

on a daily basis to really

Speaker:

understand what they're talking

Speaker:

about. So if they're going to encourage

Speaker:

the rest of the organization to

Speaker:

experiment with AI, the starting

Speaker:

point for these things, outside of

Speaker:

speaking about it, is for the

Speaker:

leadership to really understand what

Speaker:

they're actually talking about, or

Speaker:

what value they're actually getting

Speaker:

themselves, to be able to express it

Speaker:

in a more articulate way,

Speaker:

presumably.

Speaker:

What do you make of that? Is that a

Speaker:

hard requirement, do you think?

Speaker:

I don't know if it's a hard

Speaker:

requirement, but like once you start

Speaker:

using it, you can't go back.

Speaker:

So I think it's like the it's more

Speaker:

like in order to understand

Speaker:

what you're talking about, you

Speaker:

should. And also to understand the

Speaker:

what is the constraints, what it's

Speaker:

good at, what it's not.

Speaker:

But it's also just like a really

Speaker:

convenient tool.

Speaker:

So whether or not you're using it

Speaker:

for other people, you should start

Speaker:

to play around with it and use it

Speaker:

for yourself.

Speaker:

It's just so nice.

Speaker:

So we had a new

Speaker:

commission policy.

Speaker:

Somebody had written it.

Speaker:

It was just in legalese,

Speaker:

through and through.

Speaker:

In the old world, I would have had

Speaker:

to rewrite that or have somebody

Speaker:

rewrite it or, you know, give it to

Speaker:

our copywriter to just put it in our

Speaker:

tone of voice or, you know, it'd be

Speaker:

a total waste of time.

Speaker:

And instead, I stuck it in.

Speaker:

Claude said, keep it exactly

Speaker:

the same sections

Speaker:

don't change the content, just

Speaker:

make it sound like our tone of tone,

Speaker:

of voice.

Speaker:

And under a minute later, it was

Speaker:

done.

Speaker:

It was really good.

Speaker:

The only annoying thing is Claude

Speaker:

loses formatting, so I

Speaker:

had to go and reformat it.

Speaker:

But like, it's just so nice

Speaker:

to have those options or

Speaker:

handed a legalese documents

Speaker:

to have to understand.

Speaker:

Couldn't understand one paragraph.

Speaker:

No matter how many times I looked at

Speaker:

it, threw it

Speaker:

into ChatGPT to

Speaker:

just explain to me what the

Speaker:

paragraph is talking about and what

Speaker:

I need to worry about.

Speaker:

And it came out and told me

Speaker:

it's really nice.

Speaker:

It just makes life easier.

Speaker:

I mean, do you use it.

Speaker:

For me creating things I need to

Speaker:

distribute? I always end up passing

Speaker:

it through ChatGPT.

Speaker:

Now to kind of do a once over in

Speaker:

terms of making it better.

Speaker:

And all things being equal, I am not

Speaker:

a fabulous writer.

Speaker:

I can write and I can get my message

Speaker:

across, but every time I put it

Speaker:

through ChatGPT and get my output,

Speaker:

it is dramatically better for

Speaker:

the most part. So I'm like, oh,

Speaker:

great. And I might tweak it here and

Speaker:

there, but definitely use it for

Speaker:

that purpose. And then the reverse,

Speaker:

which is where you just pointed out

Speaker:

any kind of incoming document that

Speaker:

is highly annoying for me to

Speaker:

understand and read, passing it

Speaker:

through to you, but again, to either

Speaker:

simplify it or kind of expressing

Speaker:

it in a different way, where I can

Speaker:

actually pull out the key bits that

Speaker:

I need to quickly.

Speaker:

It's very useful for that.

Speaker:

I'm a little bit sometimes

Speaker:

skeptical, but for things that are

Speaker:

rather important where the detail

Speaker:

matters, I'm a little more worried

Speaker:

sometimes that it's not pulling out

Speaker:

all the bits that I actually need,

Speaker:

to be honest. There's some level of

Speaker:

my going back to the source

Speaker:

document just to check things and

Speaker:

verify that at all.

Speaker:

I'm getting what I need, basically.

Speaker:

But my suspicion is, and

Speaker:

this is separate from ChatGPT and

Speaker:

the generic tools right now.

Speaker:

But there's probably again, probably

Speaker:

AI agents out there that

Speaker:

are verticals that probably do a

Speaker:

much better job of this to ensure

Speaker:

that you're getting exactly what you

Speaker:

need with a higher level of

Speaker:

confidence and accuracy than using

Speaker:

the generic tools.

Speaker:

But that aside, your

Speaker:

use cases or my use cases, and I'm

Speaker:

sure for any C-suite leader or

Speaker:

that kind of incoming document

Speaker:

of complexity, output of actual

Speaker:

things that you need to communicate

Speaker:

seems like a no brainer.

Speaker:

And then the other one I use is

Speaker:

if I have writer's block

Speaker:

or I don't quite know how to get

Speaker:

started, and that blank sheet of

Speaker:

paper a moment, rather than

Speaker:

just the willpower of getting past

Speaker:

the blank sheet of paper that I have

Speaker:

used for the last, whatever,

Speaker:

30, 40 years of my life,

Speaker:

I'll go to ChatGPT or Claude

Speaker:

and just say, this is what I have

Speaker:

to do.

Speaker:

How should I get started?

Speaker:

You know, and just kind of what

Speaker:

would be the typical arguments in

Speaker:

this case? Or what would the AI

Speaker:

structure be?

Speaker:

And even if I don't end up using

Speaker:

that structure. I no longer have a

Speaker:

blank sheet of paper, and that just

Speaker:

really helps get me started.

Speaker:

And I don't need that same level of

Speaker:

willpower I used to have.

Speaker:

Yeah, I think you're exactly right.

Speaker:

So I was having to come up with

Speaker:

kind of a interview process for

Speaker:

sales reps, and also the kinds

Speaker:

of questions that I want to ask for

Speaker:

different elements that I was

Speaker:

looking for in terms of being able

Speaker:

to make judgments on the candidates

Speaker:

and ChatGPT for that kind of generic

Speaker:

interview process and questions

Speaker:

was absolutely fabulous.

Speaker:

And to your point, instead of me

Speaker:

having to I mean, I've interviewed

Speaker:

sales reps so many times before with

Speaker:

so many different question types and

Speaker:

processes.

Speaker:

It's all vaguely in the back of my

Speaker:

head, but it requires me

Speaker:

if I need to do net new, to think

Speaker:

about it again and kind of think

Speaker:

through. Okay, what did I do before?

Speaker:

Do I have some previous documents I

Speaker:

can rip off?

Speaker:

All that takes thinking time

Speaker:

basically, whereas now

Speaker:

I can just ignore all that wholesale

Speaker:

and very rapidly get

Speaker:

the process, get the steps, get the

Speaker:

questions and get the

Speaker:

criteria boxes all set up

Speaker:

basically within whatever, half

Speaker:

an hour and then ship it out to Emma

Speaker:

as part of the process and boom,

Speaker:

rocking and rolling.

Speaker:

So I think for that kind of

Speaker:

activity, for stuff that you kind of

Speaker:

know, but it's annoying to have

Speaker:

to re remember everything.

Speaker:

That's a great use case.

Speaker:

I love that like the annoying re

Speaker:

remembering.

Speaker:

That's why we're like yeah we're so

Speaker:

over it.

Speaker:

So Charlie talked several

Speaker:

times about this, which is every

Speaker:

time he has to do something,

Speaker:

he actually doesn't go to the

Speaker:

generic box, as it were.

Speaker:

He always creates a project for

Speaker:

himself and gives context

Speaker:

or the prompts around the particular

Speaker:

subject matter that he's wanting a

Speaker:

response to in this case.

Speaker:

And so his go to effectively is to

Speaker:

create projects, is what I'm saying.

Speaker:

So my curiosity with you,

Speaker:

do you use projects and

Speaker:

is it useful?

Speaker:

I'm very impatient and a little bit

Speaker:

lazy.

Speaker:

And so I'm using gen

Speaker:

AI to make my life easier

Speaker:

really quickly.

Speaker:

And I find projects kind of annoying

Speaker:

because it slows me down.

Speaker:

And I think if I were trying to do

Speaker:

the next level of work, it would be

Speaker:

worth the input.

Speaker:

But for the most part, I find

Speaker:

that without using projects, the

Speaker:

output is good enough that

Speaker:

I don't find it worthwhile

Speaker:

putting in that next layer.

Speaker:

I'm not like Charlie, who

Speaker:

has created an entire project

Speaker:

without, you know, to tie a product

Speaker:

and code without using

Speaker:

anybody but him.

Speaker:

I just want to know

Speaker:

what would be some good agenda items

Speaker:

for a sales kickoff.

Speaker:

As Charlie described, some of the

Speaker:

projects that he does, and some of

Speaker:

the prompts that he sets up are

Speaker:

super complex.

Speaker:

Like, like, you know, he described

Speaker:

like doing a five page prompt.

Speaker:

I think it was a 20 page prompt.

Speaker:

Yeah. Yeah, that sounds like a lot

Speaker:

of thinking.

Speaker:

Seems like so the subset

Speaker:

of that is he has created

Speaker:

what he called a CEO pod

Speaker:

with pre-built prompts for

Speaker:

interesting topics, and I thought

Speaker:

that seemed very interesting to me

Speaker:

as a senior executive.

Speaker:

So essentially, what he's done is

Speaker:

taking that product trumpet over the

Speaker:

trumpet before It's usually used for

Speaker:

like sales enablement for a buyer

Speaker:

to go in there with all the

Speaker:

documentation around the product,

Speaker:

this and the other is used it

Speaker:

to basically be a place for

Speaker:

anybody that goes to his LinkedIn

Speaker:

feed to go into that little

Speaker:

box world, and in that box world

Speaker:

that is called the CEO pod.

Speaker:

He has three prompts that are

Speaker:

sitting there as defaults.

Speaker:

One prompt is for

Speaker:

all hands prep and

Speaker:

the definition or the.

Speaker:

I guess the objective of that prompt

Speaker:

is to develop your messaging

Speaker:

and talk track to convey your vision

Speaker:

and inspire your employees.

Speaker:

So it's kind of a drop in prompt for

Speaker:

a project, presumably for all

Speaker:

hands that you could use going

Speaker:

forward. So my suspicion

Speaker:

is for people like us where

Speaker:

we don't do spending hours building

Speaker:

prompts of five pages, that

Speaker:

there should be good, healthy,

Speaker:

generic prompts for things like all

Speaker:

hands or this then the other, that

Speaker:

we can simply take it and drop in

Speaker:

and use it, where we actually get

Speaker:

a better answer than what we would

Speaker:

have got generically to the box, but

Speaker:

I don't have to think about

Speaker:

creating that problem myself in this

Speaker:

case.

Speaker:

Yeah. And then the question is

Speaker:

who's going to provide it?

Speaker:

Like is it going to be the rise of

Speaker:

a bunch of apps that are

Speaker:

basically rap arounds of the

Speaker:

labs?

Speaker:

Is it going to be the

Speaker:

labs themselves providing it,

Speaker:

or is it going to be consultants, or

Speaker:

is it going to be like your really

Speaker:

talented 22

Speaker:

year old who comes into the business

Speaker:

is like, you know that going to be

Speaker:

their future, but I don't think it's

Speaker:

gonna be the future for very long,

Speaker:

because ChatGPT

Speaker:

is just going to create a better

Speaker:

user interface and eat up

Speaker:

all of these other businesses.

Speaker:

Another one that I hear is not doing

Speaker:

well is 11 X.

Speaker:

Like the first automated

Speaker:

one of these like SDR offerings,

Speaker:

they got to 10 million super fast.

Speaker:

For what I hear, they're about to

Speaker:

that 50% churn

Speaker:

and wasn't a good product.

Speaker:

Not worth it after people gave it a

Speaker:

go. And then also just loads

Speaker:

and loads of people are now building

Speaker:

their own automated stores.

Speaker:

And so that's another example of

Speaker:

there's going to be a lot of crashing

Speaker:

and burning while figuring out where

Speaker:

the real value is, and that

Speaker:

apps layer is definitely at

Speaker:

risk and probably

Speaker:

like how specialized

Speaker:

can you go?

Speaker:

So when you talk about like ones

Speaker:

for lawyers, one's for

Speaker:

doctors, like, you know, the very

Speaker:

specific horizontals are probably

Speaker:

safe.

Speaker:

But let's say.

Speaker:

I somehow feel it's like the hype

Speaker:

curve where the companies come out

Speaker:

of nowhere. They create first

Speaker:

versions of the product around stars

Speaker:

or whatever people are.

Speaker:

Budgets are excited.

Speaker:

They spend it, they use it.

Speaker:

It's very disappointing for all

Speaker:

sorts of reasons.

Speaker:

Then we go through that phase where

Speaker:

all these companies and products

Speaker:

go to the trough of like

Speaker:

disillusionment and like not getting

Speaker:

the revenues, high churn, this, that

Speaker:

and the other.

Speaker:

But slowly but surely they figure

Speaker:

it out and they figure out how to

Speaker:

build real value and they come back

Speaker:

up the curve eventually.

Speaker:

So it feels like on balance,

Speaker:

I would imagine this will happen in

Speaker:

this sector, but there is

Speaker:

that kind of risk that sits there in

Speaker:

terms of open AI ultimately eating

Speaker:

into these stocks in a way where

Speaker:

somehow it becomes so easy

Speaker:

for a single individual organization

Speaker:

to do it all themselves.

Speaker:

They don't actually need to go to

Speaker:

these companies to buy these

Speaker:

products, to your point.

Speaker:

And it's all happening so fast.

Speaker:

So you have these rises and falls

Speaker:

within months, not

Speaker:

years.

Speaker:

Yeah, exactly.

Speaker:

So what I wanted to ask you was

Speaker:

how to communicate the importance

Speaker:

of experimenting

Speaker:

with ChatGPT and Claude

Speaker:

and know PLM and Gemini

Speaker:

across the organization to get

Speaker:

people that dropped people's minds

Speaker:

around the fact that this is very

Speaker:

important for businesses going

Speaker:

forward. And we need to start

Speaker:

experimenting. We need to start

Speaker:

doing things. And obviously, from an

Speaker:

operator standpoint, there's all

Speaker:

sorts of things you can do to make

Speaker:

that happen.

Speaker:

But purely in terms of just straight

Speaker:

up communication to the company,

Speaker:

what are the kind of like key

Speaker:

messages that are

Speaker:

the most powerful to just get into

Speaker:

people's minds that this matters.

Speaker:

And we need to do something about

Speaker:

it.

Speaker:

So I'm just stuck on your mention

Speaker:

of Gemini, and I'm using Gemini

Speaker:

right now, and I can't really tell

Speaker:

if I'm using the paid version of the

Speaker:

free version, but whatever I'm using

Speaker:

just sucks.

Speaker:

So first of all, I just want to add

Speaker:

that in like it is

Speaker:

massively underwhelming.

Speaker:

It doesn't help me in any way.

Speaker:

It doesn't do any of the things I

Speaker:

would like it to do, and its answers

Speaker:

are stupid.

Speaker:

Slam on Gemini.

Speaker:

Basically, everywhere I

Speaker:

go on the Google suite,

Speaker:

there's a little like, how can I

Speaker:

help you box?

Speaker:

And then I ask it to help me and it

Speaker:

just does not do anything helpful.

Speaker:

And so perfect example

Speaker:

is I took

Speaker:

the commission policy,

Speaker:

ran it through Claude, made it

Speaker:

friendly and pleasant, stuck

Speaker:

it in a Google doc.

Speaker:

Then I asked Google to format

Speaker:

it like the above

Speaker:

policy, and it just rewrote

Speaker:

the policy for me in

Speaker:

a stupid way.

Speaker:

And it didn't reformat anything.

Speaker:

And it only does stuff in the chat

Speaker:

box.

Speaker:

But like, I don't care about the

Speaker:

chat box. I would think that if

Speaker:

you are Google and it's integrated,

Speaker:

it should do things in my document

Speaker:

and not just produce me.

Speaker:

Like all it is, is an integrated

Speaker:

chat box.

Speaker:

That's not the power of Google

Speaker:

having it.

Speaker:

They need to get to that next level.

Speaker:

When you think about canvas within

Speaker:

chat GPT, that is genius.

Speaker:

That works so well in that respect.

Speaker:

And you can imagine from a Google

Speaker:

doc point of view, they should do

Speaker:

exactly the same thing, and it

Speaker:

should be way better given as Google

Speaker:

Docs as opposed to some flimsy

Speaker:

canvas thing that ChatGPT just

Speaker:

created.

Speaker:

So anyhow, that wasn't your

Speaker:

question.

Speaker:

So your question was what are

Speaker:

the reasons why?

Speaker:

What's the message to the company?

Speaker:

You know, we need have this idea

Speaker:

that for employees that

Speaker:

this is the future, this is how we

Speaker:

win. Part of that is to take

Speaker:

A.I., truly embrace it,

Speaker:

Activate ourselves to use it to help

Speaker:

us with what we're doing to to get

Speaker:

to the promised land effectively,

Speaker:

and that this is a winning mindset

Speaker:

and don't see it as cheating.

Speaker:

We should be using it and

Speaker:

it's very sensible.

Speaker:

So we have for

Speaker:

engineering, we've thought through

Speaker:

the workflows and the tools to

Speaker:

be used and have some

Speaker:

guidance is quite specific.

Speaker:

And then for the rest of the

Speaker:

business, what we're doing is

Speaker:

in every single weekly

Speaker:

stand up, somebody does a show

Speaker:

and tell on what they've done

Speaker:

with Jenny that week

Speaker:

and what they've learned and what

Speaker:

worked and what didn't work.

Speaker:

And so it can be really high

Speaker:

level stuff for what marketing's

Speaker:

doing. Or this week's

Speaker:

was somebody in our R&D

Speaker:

team going into

Speaker:

huge amounts of detail with,

Speaker:

I don't know, I think it was

Speaker:

Claude's coding

Speaker:

program and

Speaker:

it's now becoming a genetic.

Speaker:

And so like how it was solving the

Speaker:

problem and working on

Speaker:

multiple versions.

Speaker:

And when it got itself into a death

Speaker:

loop and how they got out of that

Speaker:

death loop. So it was like super

Speaker:

detailed and technical,

Speaker:

but it wasn't a set training

Speaker:

program.

Speaker:

But we are giving people

Speaker:

guidance and

Speaker:

thinking through more than

Speaker:

just a policy on how to use

Speaker:

it and what

Speaker:

to get out of it.

Speaker:

And then the other thing is we do

Speaker:

have budgets to experiment.

Speaker:

So people are looking at windsurf

Speaker:

and cursor and the ChatGPT

Speaker:

and the Claude and

Speaker:

Copilot,

Speaker:

GitHub Copilot and

Speaker:

Amazon Q so we're looking

Speaker:

at like all of the different coding

Speaker:

tools and not

Speaker:

embedding them, but just

Speaker:

experimenting and seeing

Speaker:

which ones are good and which ones

Speaker:

are shit. And just as another FYI,

Speaker:

nobody seems to like copilot very

Speaker:

much. It sounds like copilot

Speaker:

might be a bit like Gemini,

Speaker:

where it's just not there.

Speaker:

So Charlie has services that he

Speaker:

provides.

Speaker:

The primary one is his AI

Speaker:

Inspiration workshop, where he goes

Speaker:

into an organization and

Speaker:

gets them really not just excited by

Speaker:

the possibilities around Cod and

Speaker:

ChatGPT, but they kind

Speaker:

of work hands on.

Speaker:

What kind of use cases might be

Speaker:

interesting for that particular

Speaker:

group? And I think he referenced

Speaker:

working in a organization for this

Speaker:

workshop with the finance team

Speaker:

and with that finance team.

Speaker:

They outline all sorts of possible

Speaker:

ways to use it within finance.

Speaker:

That's part of that workshop where

Speaker:

they got those finance individuals

Speaker:

on their pathway to take

Speaker:

those projects and start building

Speaker:

them out, basically with projects

Speaker:

or custom chat gifts or what have

Speaker:

you, I suppose.

Speaker:

So I was kind of wondering, what

Speaker:

do you make of that activation of

Speaker:

people getting things to happen,

Speaker:

workshops like this?

Speaker:

Yeah, I think it's a really good

Speaker:

idea. And also like

Speaker:

finance might be a bit risk averse

Speaker:

and not wanting to change, but they

Speaker:

have some of the most boring,

Speaker:

tedious, repetitive jobs out there.

Speaker:

If I were the person who had to,

Speaker:

like, handle all

Speaker:

of the expenses

Speaker:

and accounts, I would be crying

Speaker:

out and experimenting to see how

Speaker:

I can make my job less boring and

Speaker:

let the machines do all of the shit

Speaker:

work. And I think that that's kind

Speaker:

of the inspiration is

Speaker:

what are the things

Speaker:

that are mind numbingly boring

Speaker:

that you have to do all the time,

Speaker:

and then it's worth taking the

Speaker:

time to create the prompts and to

Speaker:

create the agent that just does

Speaker:

the shit work for you.

Speaker:

And that's kind of like my

Speaker:

inspiration, you know, it's like all

Speaker:

of those back office roles

Speaker:

that are horrible.

Speaker:

They're the first place to look

Speaker:

because you just have happier people

Speaker:

who aren't just doing

Speaker:

awful stuff, and they

Speaker:

don't have to do awful stuff the

Speaker:

whole time. But like months end,

Speaker:

nobody looks forward to month end in

Speaker:

finance.

Speaker:

Yeah, it's a lot of pressure.

Speaker:

A lot of pressure and a lot of

Speaker:

boring shit.

Speaker:

So I will report on this at

Speaker:

some point in the future, but I'm

Speaker:

very keen in an organization I'm

Speaker:

currently working with to take

Speaker:

all the policies and to

Speaker:

stick them into notebook alarm

Speaker:

specifically. So this company uses

Speaker:

the Google Suite.

Speaker:

So notebook alum, you can share

Speaker:

your notebooks across your

Speaker:

organization seamlessly.

Speaker:

And notebook alum seems very well

Speaker:

suited for this task which is

Speaker:

taking documents inserting them,

Speaker:

not using the extended world of

Speaker:

alums, whereby gets confused between

Speaker:

your documentation and the greater

Speaker:

world of documentation around those

Speaker:

policies, and

Speaker:

also with that podcast

Speaker:

interface and different ways to

Speaker:

access the information.

Speaker:

And it seems so much more obvious.

Speaker:

I wouldn't say fun, but just like

Speaker:

different ways to like allow

Speaker:

the employee to get information

Speaker:

in a fast, efficient, timely basis.

Speaker:

So I think that as an actual kind of

Speaker:

low hanging fruit as a CEO,

Speaker:

that seems like just a fun

Speaker:

thing to start with.

Speaker:

Yeah, I agree.

Speaker:

Let me know how it goes.

Speaker:

So let's park it here and let's get

Speaker:

on to our conversation.

Speaker:

Part two with Mr. Charlie Callum.

Speaker:

The more that I learn about AI,

Speaker:

the more I realize you don't need

Speaker:

to know anything about AI to

Speaker:

embed AI and benefit

Speaker:

from AI, and you

Speaker:

need to know about neural networks

Speaker:

to benefit from AI.

Speaker:

In the same way, you need to know

Speaker:

about nuclear fission to benefit

Speaker:

from electricity.

Speaker:

You don't. You just need to plug in

Speaker:

to the wall and consume the

Speaker:

electricity.

Speaker:

How it gets to your plug

Speaker:

socket is really I have no interest

Speaker:

to you, and we

Speaker:

very much seen this with AI

Speaker:

over the last, you know, sort of 10

Speaker:

or 20 years where it's been, oh,

Speaker:

you need to be a data scientist.

Speaker:

You need to have a machine learning

Speaker:

degree to be able to understand

Speaker:

what's going on.

Speaker:

But November 22nd,

Speaker:

with the arrival of ChatGPT,

Speaker:

that is the plug socket.

Speaker:

The plug socket is the chat input

Speaker:

window.

Speaker:

That is all you need to know what

Speaker:

happens behind it.

Speaker:

Neural networks.

Speaker:

Reinforcement learning.

Speaker:

You don't need to know that to

Speaker:

benefit from it.

Speaker:

We are all

Speaker:

CEOs of organizations

Speaker:

that don't want to be left behind.

Speaker:

We understand the importance of AI.

Speaker:

How do we get started?

Speaker:

There's a few things that I would

Speaker:

think about as a sort

Speaker:

of setting the scenes as an

Speaker:

executive, as a CEO, as part of that

Speaker:

leadership team.

Speaker:

And that is firstly, just

Speaker:

understanding what is our

Speaker:

approach here.

Speaker:

You can be an AI

Speaker:

pessimist, which is where

Speaker:

you're like, oh, you know, I'm not

Speaker:

really sure about this thing.

Speaker:

And if we are going to use it, I'm

Speaker:

going to be thinking about how do we

Speaker:

strip out costs from

Speaker:

our business.

Speaker:

How do we become more

Speaker:

efficient and do things with less

Speaker:

people?

Speaker:

I think of this if you're a company

Speaker:

of a thousand people, well,

Speaker:

we can get to the same destination

Speaker:

now with only 500.

Speaker:

So that's the pessimistic view.

Speaker:

The alternative is to be an

Speaker:

AI optimist and go right, we're

Speaker:

going to lean into this and

Speaker:

we're a company of what do I say, a

Speaker:

thousand people and we're now over

Speaker:

500 people. We're not going to act

Speaker:

like a company of 5000 people.

Speaker:

We're going to be able to go further

Speaker:

with the same amount of resource.

Speaker:

And that kind of, are we leaning

Speaker:

into this thing or are we leaning

Speaker:

out of it is a real

Speaker:

precursor to everything that you're

Speaker:

going to do afterwards.

Speaker:

So having defined right, we're going

Speaker:

to lean into this and we're going to

Speaker:

figure out how do we act like a much

Speaker:

bigger company by using

Speaker:

these tools.

Speaker:

Next thing you probably want to do

Speaker:

is very quickly codify

Speaker:

that into some form of

Speaker:

AI policy.

Speaker:

Now I'm very nervous about

Speaker:

mentioning AI policies

Speaker:

because it's like, oh my goodness.

Speaker:

Like, this is just like, you know,

Speaker:

who's going to read it?

Speaker:

What's the point.

Speaker:

Where CEOs, we love a good policy.

Speaker:

You don't need to say that policies

Speaker:

are a problem here.

Speaker:

One of the common things that we see

Speaker:

is that if you ask companies,

Speaker:

you know, what are you doing with

Speaker:

AI?

Speaker:

You know, what benefits are you

Speaker:

seeing? You often see quite

Speaker:

muted responses.

Speaker:

We're not really seeing it.

Speaker:

We've got some pieces.

Speaker:

There's nothing really that's out

Speaker:

there in production.

Speaker:

If you go and speak to the people

Speaker:

one by one, are you using

Speaker:

ChatGPT every day?

Speaker:

Every day? I'm using it all the

Speaker:

time. You know, it's where I go to

Speaker:

ask any task that I've got.

Speaker:

So where is this discrepancy

Speaker:

coming from?

Speaker:

People are saying they're using it

Speaker:

all the time, and yet companies

Speaker:

are saying we're not really using it

Speaker:

after this because people are not

Speaker:

telling their manager, their

Speaker:

leaders, their colleagues when

Speaker:

they're using it. It's kind of under

Speaker:

the radar a little bit.

Speaker:

You know, oh, I'm just going to ask

Speaker:

ChatGPT or Claude.

Speaker:

It might be seen as cheating.

Speaker:

And often this is because certainly

Speaker:

when I'm speaking to teams, I'm

Speaker:

not sure what the rule is.

Speaker:

I'm not sure whether I'm allowed

Speaker:

to use it, so I'm definitely doing

Speaker:

it because I know that it gets me

Speaker:

where I want to go faster, but

Speaker:

I'm not being very public about it.

Speaker:

So one of the things that you can do

Speaker:

with creating a short,

Speaker:

pragmatic, positive

Speaker:

AI policy is

Speaker:

we are leading into AI.

Speaker:

We encourage your use.

Speaker:

We encourage you to experiment with

Speaker:

new tools, but be

Speaker:

sensible. Here are some guardrails

Speaker:

that you might want to follow so

Speaker:

that you know we're not sharing

Speaker:

confidential data or so on,

Speaker:

but having that policy where we're

Speaker:

setting what's right and what's

Speaker:

wrong, but reinforcing to everyone

Speaker:

we are leaning into this and we want

Speaker:

you to lean into it is

Speaker:

it's a really good approach.

Speaker:

And I can't remember if I asked this

Speaker:

question before, but it's something

Speaker:

I really want to know about.

Speaker:

And it's probably like the next

Speaker:

layer of depth rather than like,

Speaker:

should you have a AI policy?

Speaker:

But being in tech, everybody says

Speaker:

that 2025 is the year of

Speaker:

a genetic.

Speaker:

It's all about a genetic AI.

Speaker:

And yet we mostly just talk about

Speaker:

the chat bot and using ChatGPT

Speaker:

to do a bit of thinking for us,

Speaker:

rather than to actually start to

Speaker:

automate our lives and be

Speaker:

proper agents.

Speaker:

I really want to understand about a

Speaker:

genetic, because it's

Speaker:

like the year of people creating

Speaker:

agents to sell to other people,

Speaker:

but if you're in a business and

Speaker:

don't want to buy 500 million

Speaker:

different vertical agents,

Speaker:

how do you start to actually

Speaker:

automate with agents and are there

Speaker:

any good platforms to do that?

Speaker:

Agent says the password.

Speaker:

So first let's define what

Speaker:

an agent is.

Speaker:

And as everyone's talking about it,

Speaker:

suddenly it becomes just like this

Speaker:

fake. No one's quite sure.

Speaker:

For an agent to have agency,

Speaker:

it has to have the ability to choose

Speaker:

what tools, if any,

Speaker:

it is going to use to complete

Speaker:

the task.

Speaker:

The opposite of that is what we

Speaker:

might think of a workflow

Speaker:

or an automation.

Speaker:

So if people are building things

Speaker:

with it's happier if they're

Speaker:

building things would make if they

Speaker:

built integrations with Boomi

Speaker:

or Informatica.

Speaker:

If those things are predetermined,

Speaker:

even if it is an alarm call

Speaker:

and then another alarm call

Speaker:

that is still more of a workflow or

Speaker:

an automation because whatever

Speaker:

happens has been predetermined.

Speaker:

The true agent is

Speaker:

that I've got this alarm

Speaker:

tool and it is able to

Speaker:

decide now I'm going to

Speaker:

go and check a stock price.

Speaker:

Now I'm going to create a document.

Speaker:

Now I'm going to go and ask, you

Speaker:

know, another alarm for for

Speaker:

something else.

Speaker:

And I'm going to decide because

Speaker:

I've got the agency about

Speaker:

what to use and when.

Speaker:

And so for any company that is

Speaker:

thinking of building agents

Speaker:

or is being sold an agent

Speaker:

by someone else, that's the

Speaker:

important thing to dig in.

Speaker:

Have we really got an agent which

Speaker:

is able to make decisions, or

Speaker:

we just got a workflow?

Speaker:

And why that's important is that

Speaker:

if we're saying 2025 is the year

Speaker:

of the agents.

Speaker:

What does that mean?

Speaker:

We start to think about true

Speaker:

digital workers,

Speaker:

digital employees that

Speaker:

can make decisions.

Speaker:

So if we I don't

Speaker:

know if someone working in HR as

Speaker:

an HR administrator,

Speaker:

I'm able to as a digital

Speaker:

agent, I'm receiving a

Speaker:

request or a job posting

Speaker:

request from a manager.

Speaker:

I'm then able to go and

Speaker:

use another tool to go and write up

Speaker:

the job description.

Speaker:

Then able to go and decide if

Speaker:

we're going to have that budget and

Speaker:

when we're going to post it, I'm

Speaker:

then going to choose which

Speaker:

job boards we're going to post it

Speaker:

on. Then maybe if and I'm going to

Speaker:

choose which candidate I'm

Speaker:

going to shortlist and get invited

Speaker:

in for an invite, that would

Speaker:

be one or a series of agents

Speaker:

that are making decisions as

Speaker:

opposed to a true workflow.

Speaker:

But that's when you're starting to

Speaker:

really replace humans.

Speaker:

So that's a great definition

Speaker:

of agents.

Speaker:

This year is supposed to be the year

Speaker:

of a genetic AI.

Speaker:

By the way, genetic is a word that's

Speaker:

been basically made up.

Speaker:

That means agent, IC

Speaker:

agent. Like now everybody

Speaker:

in here, like I think I saw a

Speaker:

genetic for the first time about

Speaker:

nine months ago. And I was like,

Speaker:

what is this word? Is it a typo?

Speaker:

And now, like my entire life is

Speaker:

about a genetic.

Speaker:

It sounds great, I want some

Speaker:

if this is the year of a genetic, I

Speaker:

said, this is a year that all

Speaker:

software companies create agents

Speaker:

that other people can buy.

Speaker:

Or is this the year that I can

Speaker:

suddenly have a million agents in

Speaker:

my business by

Speaker:

using ChatGPT?

Speaker:

I'll answer that by talking

Speaker:

to maybe some of the steps that

Speaker:

a CEO or a company might

Speaker:

want to go through when

Speaker:

embarking on trying to make that

Speaker:

decision, both on anthropic

Speaker:

docs, website documentation,

Speaker:

and also on eyes.

Speaker:

They talk about passing the the

Speaker:

intern test, which is

Speaker:

you should think of I as

Speaker:

a brand new, eager,

Speaker:

but poorly informed intern.

Speaker:

And if you provide that intern

Speaker:

with a badly defined

Speaker:

instructions, you will get

Speaker:

what you deserve in return.

Speaker:

The original request

Speaker:

is quite poorly defined,

Speaker:

and that's often because

Speaker:

the person that's doing the job

Speaker:

today has been doing it for a while.

Speaker:

They kind of know it.

Speaker:

There's lots of sort of assumptions

Speaker:

or well, I'm sure you know what that

Speaker:

means.

Speaker:

And this is often why

Speaker:

you see in companies that when

Speaker:

someone asks, you know, how do I

Speaker:

do this? Or, you know, need and you

Speaker:

can speak to Gary.

Speaker:

He's been here for 20 years and he

Speaker:

knows how that works.

Speaker:

You can't take what's in Gary's

Speaker:

head and just give that to

Speaker:

the LM, because there's so much

Speaker:

of Gary's experience and knowledge

Speaker:

that's wrapped up in there.

Speaker:

So I find myself writing

Speaker:

a lot of long, prompt documents

Speaker:

that are maybe anything from 10 to

Speaker:

20 pages long, which is first

Speaker:

we do this, then we do that.

Speaker:

Then we do that.

Speaker:

This is what that means

Speaker:

to pass the internal test.

Speaker:

If I gave that Google doc to

Speaker:

an intern, could they do

Speaker:

what I'm asked them to do and I

Speaker:

would get the output?

Speaker:

And with anthropic and OpenAI,

Speaker:

when they say pass the intern test.

Speaker:

If the answer to that is no, that

Speaker:

if I gave that document to an

Speaker:

intern, would I get the response

Speaker:

I wanted?

Speaker:

Then how is an El Alam going

Speaker:

to do that?

Speaker:

It's not.

Speaker:

So I'd say a large proportion

Speaker:

of the time that I spend is in

Speaker:

getting the ask correct.

Speaker:

Once you've got the ask correct,

Speaker:

then creating whether as a

Speaker:

custom GPT or whether it is

Speaker:

building out some form of agent that

Speaker:

is able to access different function

Speaker:

calls, that becomes relatively

Speaker:

easy, because what you've said to

Speaker:

the intern is, let's do that HR

Speaker:

hiring example, right.

Speaker:

The first thing that you're going to

Speaker:

do is you're going to be on the

Speaker:

lookout for job requests

Speaker:

that come from hiring managers.

Speaker:

The second thing that you're going

Speaker:

to do is you're going to take all

Speaker:

the details of that job request, and

Speaker:

then you're going to start drafting

Speaker:

up a more detailed job description.

Speaker:

This is what our job descriptions

Speaker:

should look like, and all of the

Speaker:

details of what should be in each

Speaker:

section.

Speaker:

And then you're going through all of

Speaker:

that. We use some examples

Speaker:

of how that H.R.

Speaker:

Hiring person would have

Speaker:

the choice of choosing which job

Speaker:

boards to post on.

Speaker:

Well, how do we make that decision

Speaker:

and how do we access those job

Speaker:

boards.

Speaker:

And that's really the instructions

Speaker:

that you're going to end up providing

Speaker:

to the AI.

Speaker:

Now, if you don't need to access any

Speaker:

tools because the human wouldn't

Speaker:

access any tools, then you're

Speaker:

thinking more of just a simpler

Speaker:

implementation.

Speaker:

If you are looking at accessing

Speaker:

tools or you need to access your HR

Speaker:

system or your payroll, then

Speaker:

you're thinking more of, I need to

Speaker:

create some kind of agent that can

Speaker:

choose when to use these tools.

Speaker:

And how do you create the agent?

Speaker:

That's my question.

Speaker:

My recommendation is a tool

Speaker:

called the vessel

Speaker:

AI SDK.

Speaker:

So vessel VR CEO

Speaker:

and vessel is a web

Speaker:

hosting and deployment company.

Speaker:

So anyone that's an AI product

Speaker:

management product development role

Speaker:

may have come across V0.

Speaker:

So that's a v not dev.

Speaker:

That also comes from vessel.

Speaker:

And that is a AI

Speaker:

generative UI.

Speaker:

So you can design your applications.

Speaker:

But there's vessel AI SDK

Speaker:

is a really neat way

Speaker:

of plugging in to open AI

Speaker:

and to anthropic and to Gemini

Speaker:

and to any of your preferred

Speaker:

LMS, and it

Speaker:

gives you the ability to create

Speaker:

these calls out, these tool calls,

Speaker:

which is what makes an agent.

Speaker:

Well, because I guess if I step

Speaker:

back to what I'm thinking about

Speaker:

and maybe it can help is in

Speaker:

the past, and particularly

Speaker:

as my experiences as Crow, which

Speaker:

I complain about bitterly.

Speaker:

Not the crow part, but the number

Speaker:

of tools I had to buy.

Speaker:

Because you had this tech stack, the

Speaker:

revenue tech stack and

Speaker:

everything costs like 20

Speaker:

K other than Salesforce, which

Speaker:

of course costs shitloads more.

Speaker:

But it'd be like, okay, now it's 20

Speaker:

K for this and 20 K for

Speaker:

this and 20 K for that, and you add

Speaker:

it all in, and then suddenly

Speaker:

you had a tech stack

Speaker:

of, I don't know, 20 different

Speaker:

things and half 1 million pounds.

Speaker:

What I don't want to do

Speaker:

when I'm looking at my

Speaker:

IT stack for the future,

Speaker:

is to have that 20,

Speaker:

30, 40 pieces

Speaker:

of tooling that are critical for the

Speaker:

business, and some of them are big

Speaker:

and some of them little, and they're

Speaker:

all whacking some AI on,

Speaker:

so they all have a chat bot.

Speaker:

Nobody seems to have a particularly

Speaker:

compelling vision.

Speaker:

I know that we are not all that

Speaker:

impressed by GitHub copilot,

Speaker:

you know, versus Verses like

Speaker:

cursor and some of the others.

Speaker:

It's I don't want to blindly

Speaker:

buy all of these tools

Speaker:

that say they have something, that

Speaker:

maybe have an agent, and

Speaker:

I still just have to pay shitloads.

Speaker:

What I would like to understand

Speaker:

is how to think about

Speaker:

what the future

Speaker:

tech stack should be for

Speaker:

a scale up. So a couple hundred

Speaker:

people to a thousand people.

Speaker:

Technology's moving really quickly.

Speaker:

This is the year of the agent.

Speaker:

What should I be buying today

Speaker:

or what should I be evaluating?

Speaker:

Because my ideal and my vision,

Speaker:

and I'm guessing that other people

Speaker:

would have this vision, is that

Speaker:

everybody has

Speaker:

their own assistant

Speaker:

who can do all of this shit work,

Speaker:

and it's really easy.

Speaker:

And you just say to your assistant,

Speaker:

can you look through all of these

Speaker:

CVS for me and stick

Speaker:

them and highlight the best

Speaker:

ones, or contact them or,

Speaker:

you know, and be able to write The

Speaker:

foreign intern prompt that's 20

Speaker:

pages long.

Speaker:

But once I do that, I never

Speaker:

have to do that work again.

Speaker:

And I want everybody in my company

Speaker:

to be able to do that, not just the

Speaker:

individual who understands how to

Speaker:

use VSL or

Speaker:

API's.

Speaker:

How far away are we from

Speaker:

that and

Speaker:

what are the steps to get there

Speaker:

this year?

Speaker:

It is such a

Speaker:

fast moving space

Speaker:

that it is just impossible

Speaker:

to say, you know, use

Speaker:

this tool or use that tool

Speaker:

because so much is coming out,

Speaker:

whether that's models or whether

Speaker:

it is tools that are set over the

Speaker:

top of these models that

Speaker:

help to create these workflows

Speaker:

is speaking to a CEO.

Speaker:

This would be my advice

Speaker:

is that there is never

Speaker:

been a better opportunity right now

Speaker:

to internalize

Speaker:

some of these skills.

Speaker:

Previously you might have said, you

Speaker:

know, build versus buy while the

Speaker:

building is just too complicated.

Speaker:

We're not going to go and set up

Speaker:

some massive infrastructure

Speaker:

and start investing in people to

Speaker:

run that. So we're going to buy

Speaker:

we'll buy a CRM, we'll buy

Speaker:

an air platform, whatever that might

Speaker:

be. But I think we're seeing now

Speaker:

that pioneering companies,

Speaker:

and that doesn't have to be just

Speaker:

like some funky startup.

Speaker:

But actually, I mean, you saw this

Speaker:

with Klarna recently

Speaker:

and saying, you know what with some

Speaker:

of these tools, whether that is

Speaker:

universal, whether that is V0,

Speaker:

whether it is a

Speaker:

new development platform

Speaker:

that people are going to, you know

Speaker:

what, I can take non

Speaker:

development people.

Speaker:

So a product manager or

Speaker:

someone in an operations team and

Speaker:

I can start building out proof

Speaker:

of concepts and building internal

Speaker:

products without

Speaker:

having to go and get a SAS

Speaker:

subscription from someone else.

Speaker:

And suddenly I'm relying on

Speaker:

their roadmap rather than

Speaker:

our own.

Speaker:

So I think there's going to be,

Speaker:

you know, some companies that just

Speaker:

don't get involved in this at all

Speaker:

and just stay. Let's leave AI

Speaker:

aside.

Speaker:

Then there are going to be some that

Speaker:

go. We want to buy stuff.

Speaker:

So let's rely on someone

Speaker:

else's a platform

Speaker:

for stars

Speaker:

or for marketing or whatever.

Speaker:

But I think there's going to be an

Speaker:

increasing number of companies said,

Speaker:

you know what?

Speaker:

We have got insight

Speaker:

into what are our challenges.

Speaker:

And we think it's quite unique to

Speaker:

us. And we're just going to

Speaker:

have a go at building something

Speaker:

internally using these

Speaker:

publicly available tools.

Speaker:

It's about using as

Speaker:

many tools as you can come across,

Speaker:

as they keep getting developed to

Speaker:

improve your learning and the way

Speaker:

of working.

Speaker:

This all feeds into that first

Speaker:

bit, which is about, you know, AI

Speaker:

positive or a negative

Speaker:

AI going to lean into AI

Speaker:

or lean out of it.

Speaker:

And again, checkers is important.

Speaker:

So having done that

Speaker:

then the primary

Speaker:

driver of a good change program,

Speaker:

and everyone will know this

Speaker:

regardless of what it is you're

Speaker:

trying to change, is having

Speaker:

empowered and inspired

Speaker:

senior leaders that are at

Speaker:

the top of that program.

Speaker:

So I would then be looking at

Speaker:

who's the rest of your executive

Speaker:

leadership team. So if you are the

Speaker:

CEO, I'd be looking left at the

Speaker:

CFO. I'd be looking at that chief

Speaker:

Revenue Officer.

Speaker:

I've been looking at the Chief

Speaker:

People officer and a level

Speaker:

down into those VP's

Speaker:

and figuring out quickly,

Speaker:

how do we get those people

Speaker:

comfortable with using

Speaker:

AI for their own tasks?

Speaker:

It is so important to have that

Speaker:

team really understand

Speaker:

how I can help them personally,

Speaker:

because that immediately filters

Speaker:

down to the rest of the

Speaker:

team. So you've got to get that team

Speaker:

on board so that could be just

Speaker:

running.

Speaker:

So inspirational workshops for those

Speaker:

leaders.

Speaker:

Maybe have a breakout session at

Speaker:

your next leadership offsite.

Speaker:

And it's not about encouraging

Speaker:

them about how their teams should be

Speaker:

using AI.

Speaker:

It's about how do you

Speaker:

use AI?

Speaker:

If you're a c o,

Speaker:

you have a set of

Speaker:

direct reports that you need to

Speaker:

manage and inspire.

Speaker:

You have got colleagues

Speaker:

who run finance or on

Speaker:

HR, or who run legal.

Speaker:

And how do you better understand

Speaker:

their view of the world as a

Speaker:

CEO? Or you may be just

Speaker:

going through a merger or

Speaker:

acquisition, and you may be having

Speaker:

to figure out, how do I

Speaker:

align this whole new set of people

Speaker:

and processes and data into our

Speaker:

organization?

Speaker:

So it's thinking about how does

Speaker:

ChatGPT, how does Claude, how

Speaker:

does Gemini, how does that solve

Speaker:

your personal, daily

Speaker:

and monthly working processes?

Speaker:

And if you can figure that out

Speaker:

immediately, it gives that

Speaker:

permission to everyone below

Speaker:

to explore and experiment as well.

Speaker:

I tried to just have one

Speaker:

subscription, but I now end up

Speaker:

having to have two.

Speaker:

So I have both Claude

Speaker:

and Betty because I

Speaker:

just find Claud writes things

Speaker:

better. But ChatGPT has a wider

Speaker:

range of capabilities and

Speaker:

functionality and also is attached

Speaker:

to the internet, and I use it when

Speaker:

I'm stuck thinking,

Speaker:

but and kind of in those use cases

Speaker:

you talked about and like the team

Speaker:

use it for ideas for deal

Speaker:

reviews or how to write a better

Speaker:

proposal.

Speaker:

But what I really want to use it for

Speaker:

is all of the hard things

Speaker:

that are boring.

Speaker:

And I hate like org charts,

Speaker:

but I ended up spending about 2.5

Speaker:

hours trying to get it to write me

Speaker:

an org chart, and I could not.

Speaker:

And I was asking it to help me

Speaker:

figure out how to tell it, to write

Speaker:

me an org chart, and I could not.

Speaker:

Is this what it's worth, going for

Speaker:

some sort of AI training course?

Speaker:

Or is it just really bad at making

Speaker:

org charts no matter what?

Speaker:

Because for whatever reason, they're

Speaker:

incredibly difficult.

Speaker:

So there's some basic tips that you

Speaker:

can follow. So whether you're using

Speaker:

code called ChatGPT.

Speaker:

Gemini.

Speaker:

Any of these tools.

Speaker:

Firstly, in both Claude and ChatGPT,

Speaker:

you can set up a project and

Speaker:

set a project as a wrapper

Speaker:

around a task or a set of

Speaker:

chats.

Speaker:

So as a CEO, if

Speaker:

you're doing anything repeatedly.

Speaker:

So that might be a set of chats

Speaker:

about one of your direct reports.

Speaker:

It might be a set of chats about an

Speaker:

acquisition you're going through.

Speaker:

It might be a set of chats about a

Speaker:

policy or process that you're

Speaker:

crafting.

Speaker:

Create a project.

Speaker:

You then upload background

Speaker:

information so that could be

Speaker:

documents. It could be

Speaker:

just your instructions about how

Speaker:

you want ChatGPT to

Speaker:

respond to you.

Speaker:

And then you get into how do you

Speaker:

craft a prompt.

Speaker:

And I'll give you some specifics

Speaker:

here. So my typical

Speaker:

prompt, especially something when

Speaker:

I'm reusing it, is anything from

Speaker:

6 to 8, nine, ten

Speaker:

pages long ago.

Speaker:

That's that's longer the little chat

Speaker:

window there.

Speaker:

But because I'm, I'm using that

Speaker:

prompt again and again, it's

Speaker:

worthwhile.

Speaker:

So I break it up with

Speaker:

XHTML tags.

Speaker:

Now you don't need to be a developer

Speaker:

to know this. I'll just explain it

Speaker:

super simply. If you don't want to

Speaker:

add XML tags.

Speaker:

So a left arrow kind of opens

Speaker:

it. And then you write the word

Speaker:

objective and then a right arrow

Speaker:

to close it. So that's your opening

Speaker:

tag objective.

Speaker:

And then to close that XML

Speaker:

tag you do exactly the same, except

Speaker:

there's a backslash up at

Speaker:

the start after the first opening

Speaker:

bracket.

Speaker:

And I'll put some of the notes about

Speaker:

this prompt and guidelines so that

Speaker:

you can have it in the show notes.

Speaker:

Now when you're writing your prompt

Speaker:

you're basically structuring.

Speaker:

So I say objective.

Speaker:

This is the first part.

Speaker:

The objective of this prompt is

Speaker:

to develop a job description

Speaker:

or a merger plan, whatever it might

Speaker:

be. Then the next tag might be

Speaker:

open the tag instructions.

Speaker:

Right. This is how I want you to do

Speaker:

it. The user is going to provide you

Speaker:

with this information.

Speaker:

I mean, I'm the user, but I'm going

Speaker:

to provide you with this information.

Speaker:

I then want you to ask me some

Speaker:

questions. And here are the

Speaker:

questions which I put in another

Speaker:

question tag.

Speaker:

Then you get to the two most

Speaker:

important parts of a really

Speaker:

great prompt.

Speaker:

And they're often the hardest bits

Speaker:

to put it in.

Speaker:

And this is what takes their long

Speaker:

time and makes it a ten page

Speaker:

document.

Speaker:

So examples.

Speaker:

So think about the intern.

Speaker:

If I ask you to build

Speaker:

me an org chart, but I don't

Speaker:

show you what a good org

Speaker:

chart looks like.

Speaker:

2 or 3 of them, you're just

Speaker:

running around in the darkness.

Speaker:

If I don't show you a bad

Speaker:

org chart, you don't know what bad

Speaker:

looks like.

Speaker:

So whether it is,

Speaker:

you know, a job description,

Speaker:

whether it's a spreadsheet table,

Speaker:

whether it is a project timeline,

Speaker:

whatever it is that you want.

Speaker:

A good prompt should always have at

Speaker:

least 2 or 3 examples,

Speaker:

a good one, and then a bad one.

Speaker:

So that can take up quite a bit.

Speaker:

The final tag that I always

Speaker:

have in there is exceptions.

Speaker:

So what should the

Speaker:

AI do if

Speaker:

you don't have certain information

Speaker:

that you've asked for?

Speaker:

So for example, if I'm drafting

Speaker:

a job description and I've asked

Speaker:

the AI and the instructions to

Speaker:

ask me for the salary range,

Speaker:

let's say I haven't got the salary

Speaker:

range yet. Well, what should the AI

Speaker:

do in that situation?

Speaker:

You might say if the user does not

Speaker:

have a salary range, then

Speaker:

know that this is to be verified

Speaker:

later and at the bottom of

Speaker:

your response, put some next

Speaker:

actions or follow ups or whatever.

Speaker:

So this sounds quite

Speaker:

detailed, but if you're

Speaker:

doing one to ones with your direct

Speaker:

reports, you do that regularly

Speaker:

and you're going to do it a lot.

Speaker:

So it's worth spending an hour

Speaker:

writing the prompt

Speaker:

so that you've got that in there.

Speaker:

So there's just a bit of guidance.

Speaker:

If you put the the right

Speaker:

in front to really

Speaker:

pass the intern test.

Speaker:

If you gave that to an intern,

Speaker:

they'd give you a good response.

Speaker:

Then I think you'll start to find

Speaker:

you get better.

Speaker:

Whether that's an org chart

Speaker:

has on that, because maybe you're

Speaker:

trying to get a visual, which it

Speaker:

probably wouldn't be very good at.

Speaker:

But in terms of the hierarchical

Speaker:

structure would probably be

Speaker:

pretty handy at doing that.

Speaker:

We in our company

Speaker:

kick off this year, we had everybody

Speaker:

working groups to identify use

Speaker:

cases along the customer journey

Speaker:

where applying.

Speaker:

I would be really cool.

Speaker:

And every single

Speaker:

team, they came up with loads of

Speaker:

different ideas, but every single

Speaker:

team came up with one idea that was

Speaker:

the same, which is basically

Speaker:

putting all

Speaker:

of our internal data

Speaker:

in a data repository and putting

Speaker:

ChatGPT on top, or a GPT

Speaker:

on top. So like all of the

Speaker:

historic slack

Speaker:

information, all of our Google

Speaker:

drives.

Speaker:

Then you have Salesforce

Speaker:

data, etc., then be able

Speaker:

to from that repository, ask

Speaker:

like, give me a summary of

Speaker:

what the customer relationship has

Speaker:

been like to date.

Speaker:

Give me a summary about like

Speaker:

usage of that customer, all of those

Speaker:

types of things.

Speaker:

So that we go think about the

Speaker:

build versus buy.

Speaker:

I'd be taking a look at glean.

Speaker:

I'm not sure if you come across Glen

Speaker:

GLE and.com.

Speaker:

And it's exactly that use case.

Speaker:

So founded by some ex Googlers.

Speaker:

How do we use AI

Speaker:

to search across our

Speaker:

existing data sources.

Speaker:

And at first glance you say,

Speaker:

well, you know, the challenge

Speaker:

is quite easy because we just go

Speaker:

and, you know, plug it in via APIs

Speaker:

to all these different things, but

Speaker:

you need to preserve the

Speaker:

data privacy rights of

Speaker:

the source material.

Speaker:

So if I go search customer

Speaker:

X, how much did they spend last

Speaker:

year. Well, I should only be able

Speaker:

to get that answer if

Speaker:

I would be able to get that

Speaker:

information in the actual source

Speaker:

system, because I've got access to

Speaker:

that data or whatever.

Speaker:

Anyway, so I had a chat with

Speaker:

one of their team just before

Speaker:

Christmas and it was a super

Speaker:

interesting use case.

Speaker:

But that's exactly what Glenn is

Speaker:

targeted, that.

Speaker:

When we first kind of reconnected

Speaker:

and started talking and you talked

Speaker:

about your Power Hour,

Speaker:

and I think I have told everybody I

Speaker:

know about it, and I thought, why

Speaker:

not share it with the rest of

Speaker:

our listeners?

Speaker:

What's your power hour?

Speaker:

So my Power Hour started

Speaker:

during Covid,

Speaker:

and like many of us,

Speaker:

I was a commuter before

Speaker:

Covid, jumping on a train.

Speaker:

God knows what hour into

Speaker:

London.

Speaker:

Covid came along and suddenly

Speaker:

I had all this time in the morning

Speaker:

and my wife would take the kids to

Speaker:

school and, you know, I'd

Speaker:

be up and about, you know, having

Speaker:

taking the dogs for a walk.

Speaker:

7:00.

Speaker:

And then the kids are off.

Speaker:

And then I had to sort of hour from

Speaker:

eight till nine when nothing

Speaker:

was really happening.

Speaker:

And so I decided that I was going

Speaker:

to start to write a book

Speaker:

during that hour.

Speaker:

And so I would just write, you

Speaker:

know, a chapter in an hour, having

Speaker:

thought about what was going to

Speaker:

write in the shower.

Speaker:

And suddenly I found it was like

Speaker:

super productive.

Speaker:

Like by the time I got to 9:00, I

Speaker:

had done a ton of work and probably

Speaker:

my most creative work of the

Speaker:

day. I was lucky enough to finish

Speaker:

published that book, and I just kept

Speaker:

that power hour going.

Speaker:

And it's the hour that I work for

Speaker:

myself, whether that's on

Speaker:

sort of personal writing, whether

Speaker:

it's recording a YouTube video.

Speaker:

And I'm very humbled

Speaker:

that I get lots of lovely comments

Speaker:

from people like, I've got no idea

Speaker:

how you create so much stuff.

Speaker:

You know, there's always a new

Speaker:

e-book that you've published, or

Speaker:

you've recorded a video for YouTube,

Speaker:

or you've done a podcast, or you've

Speaker:

published a book and you know, how

Speaker:

are you doing all this stuff?

Speaker:

And you've got four children and two

Speaker:

dogs and you've got, you know, your

Speaker:

work and everything, and it's that

Speaker:

Power hour has been

Speaker:

super helpful.

Speaker:

And once you get into the

Speaker:

habit of it, then

Speaker:

it's very difficult to break.

Speaker:

It doesn't actually seem like work

Speaker:

anymore. It's just a routine I

Speaker:

find myself.

Speaker:

Weekends included at

Speaker:

my desk from eight till nine.

Speaker:

Just crank through some great work.

Speaker:

The book that inspired that was

Speaker:

Atomic Habits, which came out

Speaker:

pre-COVID. I'm sure many people have

Speaker:

read it was an Amazon bestseller,

Speaker:

but it was all about, you know, if

Speaker:

you create the framework

:

00, I will sit down

:

at my desk, create the input,

:

and then the output arrives

:

magically.

:

Lovely. So if you like what you

:

hear, please leave us a comment or

:

subscribe and we will wrap on

:

this episode of The Operations Room.

:

Thank you for joining us, Charlie.

Show artwork for The Operations Room: A Podcast for COO’s

About the Podcast

The Operations Room: A Podcast for COO’s
We are the COO coaches to help you successfully scale in this new world where efficiency is as important as growth. Remember when valuations were 3-10x ARR and money wasn’t free? We do. Each week we share our experiences and bring in scale up experts and operational leaders to help you navigate both the burning operational issues and the larger existential challenges. Beth Ayers is the former COO of Peak AI, NewVoiceMedia and Codilty and has helped raise over $200m from top funds - Softbank, Bessemer, TCV, MCC, Notion and Oxx. Brandon Mensinga is the former COO of Signal AI and Trint.

About your host

Profile picture for Brandon Mensinga

Brandon Mensinga