Hello everyone, and welcome to another episode of The Operations Room, a podcast for CEOs.

I am Brandon Metzinger and I am joined by my lovely co-host Bethany Air is having some nice porridge.

Breakfast is what I want.

I'm looking at.

If only it were porridge.

Oh, it's not porridge.

Looks like porridge.

No, this is the chia seed concoction that, thanks to Zoe, I eat every day.

From chia seeds, other seeds, kefir, milk, lots of nuts.

And therefore every day.

Right.

So every trendy ingredient that I can imagine is on the list.

Pretty much.

So didn't you tell me before that the taste of your dish there, the cheesy delight, is not great or it's rather plain.

Yeah.

It's plain.

I have some strawberries in there that helps the nuts help it, but it's a bit of a slog.

And also because it's so chewy, it requires so much chewing that it takes me absolutely forever to eat it.

I don't know how much reading you've done about ultra processed food.

So one of the things with ultra processed food is it's very soft.

And that has made our jaws stop growing properly.

I've not heard this before.

Is that true?

Yeah.

So children and we were raised on enough ultra processed food to make the difference.

All of our jaws are more recessed, and the reason why we need our wisdom teeth taken out is.

Apparently your jaw grows through hard work as it's forming as a child.

And so it actually comes further out.

And then you have space for your wisdom teeth.

But with ultra processed food, it's really soft.

So our jaws don't work anymore.

And that's why everybody needs braces.

Oh my goodness.

I did not know this.

Well, I would say this I ate a tremendous amount of cashews.

And cashews are quite hard, so I suspect my jaw line is kind of being exercised.

From the outside, it looks like your jawline is fine, but I don't know.

Did you take your wisdom teeth out?

Could you have had more of a jaw?

Had you knocked everyday?

I don't even know.

I literally have no memory or recollection of having them taken out, so they must still be on.

Right.

And came in in a normal way.

Like they didn't cause any problems.

You didn't need anything.

My superior jawline was sufficient to house them.

All those cashew nuts and the clean living of Canada.

So besides the chia seed.

That's right.

The ultra processed food.

What is happening in Bethany's world?

Just insanely busy.

So I've been working from home this week.

I managed to leave the house on Wednesday and now today it's Friday.

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

I have back to back meetings like literally back to back in my calendar.

Not a sliver of space until half six.

So luckily my husband is working from home today and so he's going to bring me some lunch around noon so that I could continue to eat, although at the rate it takes me to eat the chia seeds, I might not actually have finished them by the time lunch arrives.

Okay, so that's fabulous.

So you have the in-house servant, in this case being your husband to deliver your lunch for.

Why not call him that?

That's that's I mean, this is the give and take in a relationship, so I don't know either.

A quick note, I just had this small thought this morning.

So I'm working with a couple individuals right now that are they do operations.

But really the granular operations if you want to call it that.

One of the individuals was asking me for a recommendation for employment law.

So I just did a quick screenshot of recommendations coming out of the Operations Nation slack channel to her that operations Nation slack channel is good for this kind of thing, which is vendor sourcing recommendations around products and people and that sort of thing is fabulous for that.

People are very proactive.

I would say, in terms of giving thoughts on who to speak to, who's good to, you know, to work with these sorts of things.

The other thought was just the obvious one, which is she's an operator.

She's at that level.

She'd be perfect to be part of the operations nation, or at least on the slack channel, is just a good, useful tool.

When these questions pop up like this that are quite straightforward, where people can respond back and give you the kind of collective set of good answers.

And it's free.

It's definitely a good community for like having problems with my pension provider or the health insurance quote has gone up by 40%.

What do I do?

Or is that the same for everyone else?

Like those are the types of questions you see.

I went to a comedy show on Sunday night and then a comedy show on Wednesday night, and I was telling a friend and he was just like, is this a new thing?

Like, it's going to comedy shows you're saying?

And said, in that kind of like snide, dismissive way, like, what is up with you?

It's not our new thing.

And I actually booked them ages ago, maybe July, August time.

And it just happened to arrive now.

And there were two very different comedy shows.

So one of them was in Hackney.

It was very trendy.

The guy is a Sri Lankan comic, as you do a Sri Lankan comic that kind of looks like Jesus, like he is very curly hair down to his waist.

Very good looking guy.

And he studied medicine and then decided not to become a medic, and then became a developer and a comic and is trying to be a comic more than a developer.

So anyhow, I have never felt like more out of place.

Everybody was trendy.

Like, I don't think there were any other women in the room.

They were just non-binary.

I guess we call them women anymore.

Like, I would have chosen to be non-binary had I had the option, because you kind of look at it, you're like, oh, I could be a woman, and that looks like a really bad deal, or I can opt out and I think I'm going to opt out.

Yeah.

Fair enough.

Yeah.

And then every man had a mustache.

Is that.

Is that a thing right now?

It really is a thing.

And here we are like middle class clap mites nots with mustaches and combat boots.

And I was just like.

And not non-binary.

Yeah.

And I just never felt so uncool in my life.

But it was quite a good show other than just feeling like such the square.

And then on Wednesday night, it was a Irish comedian who's a couple years older than me.

Woman.

And other than the fact that at least 90% of the audience were Irish, everybody was a middle aged woman and maybe one and four brought their husbands along.

So I felt totally in my element.

We all belonged.

We were all square, we were all 50 ish, and it was quite good, but totally different sets.

One was around like his was around not belonging and racism and privilege.

And hers is about menopause and sex.

And husbands and children.

Congratulations though to you.

Because actually booking kind of not social but like just outings of fun and enjoyment.

You know, I feel like my calendar right now is pretty sparse in that respect.

So I'm always actively looking for, I don't know, just interesting things that would be, I don't know, fun, enjoyable to do and were doable just given our schedule with the kids and all that jazz.

But nothing has really popped up quite recently.

That seems useful enough or good enough to go to.

All right.

So we have got a great topic for today, which is a returning topic, as it were, building an AI for Organization Part two.

We have an amazing returning guest for this, which is Charlie Cowan.

We had so much fun with Charlie that we had a second conversation.

And of course, he is an AI strategist and AI change agent within organizations today.

So before we get into our part two with Charlie, Just wanted to talk about a couple bits and pieces here.

One of the elements that he spoke about was leadership.

Using AI themselves on a daily basis to really understand what they're talking about.

So if they're going to encourage the rest of the organization to experiment with AI, the starting point for these things, outside of speaking about it, is for the leadership to really understand what they're actually talking about, or what value they're actually getting themselves, to be able to express it in a more articulate way, presumably.

What do you make of that?

Is that a hard requirement, do you think?

I don't know if it's a hard requirement, but like once you start using it, you can't go back.

So I think it's like the it's more like in order to understand what you're talking about, you should.

And also to understand the what is the constraints, what it's good at, what it's not.

But it's also just like a really convenient tool.

So whether or not you're using it for other people, you should start to play around with it and use it for yourself.

It's just so nice.

So we had a new commission policy.

Somebody had written it.

It was just in legalese, through and through.

In the old world, I would have had to rewrite that or have somebody rewrite it or, you know, give it to our copywriter to just put it in our tone of voice or, you know, it'd be a total waste of time.

And instead, I stuck it in.

Claude said, keep it exactly the same sections don't change the content, just make it sound like our tone of tone, of voice.

And under a minute later, it was done.

It was really good.

The only annoying thing is Claude loses formatting, so I had to go and reformat it.

But like, it's just so nice to have those options or handed a legalese documents to have to understand.

Couldn't understand one paragraph.

No matter how many times I looked at it, threw it into ChatGPT to just explain to me what the paragraph is talking about and what I need to worry about.

And it came out and told me it's really nice.

It just makes life easier.

I mean, do you use it.

For me creating things I need to distribute?

I always end up passing it through ChatGPT.

Now to kind of do a once over in terms of making it better.

And all things being equal, I am not a fabulous writer.

I can write and I can get my message across, but every time I put it through ChatGPT and get my output, it is dramatically better for the most part.

So I'm like, oh, great.

And I might tweak it here and there, but definitely use it for that purpose.

And then the reverse, which is where you just pointed out any kind of incoming document that is highly annoying for me to understand and read, passing it through to you, but again, to either simplify it or kind of expressing it in a different way, where I can actually pull out the key bits that I need to quickly.

It's very useful for that.

I'm a little bit sometimes skeptical, but for things that are rather important where the detail matters, I'm a little more worried sometimes that it's not pulling out all the bits that I actually need, to be honest.

There's some level of my going back to the source document just to check things and verify that at all.

I'm getting what I need, basically.

But my suspicion is, and this is separate from ChatGPT and the generic tools right now.

But there's probably again, probably AI agents out there that are verticals that probably do a much better job of this to ensure that you're getting exactly what you need with a higher level of confidence and accuracy than using the generic tools.

But that aside, your use cases or my use cases, and I'm sure for any C-suite leader or that kind of incoming document of complexity, output of actual things that you need to communicate seems like a no brainer.

And then the other one I use is if I have writer's block or I don't quite know how to get started, and that blank sheet of paper a moment, rather than just the willpower of getting past the blank sheet of paper that I have used for the last, whatever, 30, 40 years of my life, I'll go to ChatGPT or Claude and just say, this is what I have to do.

How should I get started?

You know, and just kind of what would be the typical arguments in this case?

Or what would the AI structure be?

And even if I don't end up using that structure.

I no longer have a blank sheet of paper, and that just really helps get me started.

And I don't need that same level of willpower I used to have.

Yeah, I think you're exactly right.

So I was having to come up with kind of a interview process for sales reps, and also the kinds of questions that I want to ask for different elements that I was looking for in terms of being able to make judgments on the candidates and ChatGPT for that kind of generic interview process and questions was absolutely fabulous.

And to your point, instead of me having to I mean, I've interviewed sales reps so many times before with so many different question types and processes.

It's all vaguely in the back of my head, but it requires me if I need to do net new, to think about it again and kind of think through.

Okay, what did I do before?

Do I have some previous documents I can rip off?

All that takes thinking time basically, whereas now I can just ignore all that wholesale and very rapidly get the process, get the steps, get the questions and get the criteria boxes all set up basically within whatever, half an hour and then ship it out to Emma as part of the process and boom, rocking and rolling.

So I think for that kind of activity, for stuff that you kind of know, but it's annoying to have to re remember everything.

That's a great use case.

I love that like the annoying re remembering.

That's why we're like yeah we're so over it.

So Charlie talked several times about this, which is every time he has to do something, he actually doesn't go to the generic box, as it were.

He always creates a project for himself and gives context or the prompts around the particular subject matter that he's wanting a response to in this case.

And so his go to effectively is to create projects, is what I'm saying.

So my curiosity with you, do you use projects and is it useful?

I'm very impatient and a little bit lazy.

And so I'm using gen AI to make my life easier really quickly.

And I find projects kind of annoying because it slows me down.

And I think if I were trying to do the next level of work, it would be worth the input.

But for the most part, I find that without using projects, the output is good enough that I don't find it worthwhile putting in that next layer.

I'm not like Charlie, who has created an entire project without, you know, to tie a product and code without using anybody but him.

I just want to know what would be some good agenda items for a sales kickoff.

As Charlie described, some of the projects that he does, and some of the prompts that he sets up are super complex.

Like, like, you know, he described like doing a five page prompt.

I think it was a 20 page prompt.

Yeah.

Yeah, that sounds like a lot of thinking.

Seems like so the subset of that is he has created what he called a CEO pod with pre-built prompts for interesting topics, and I thought that seemed very interesting to me as a senior executive.

So essentially, what he's done is taking that product trumpet over the trumpet before It's usually used for like sales enablement for a buyer to go in there with all the documentation around the product, this and the other is used it to basically be a place for anybody that goes to his LinkedIn feed to go into that little box world, and in that box world that is called the CEO pod.

He has three prompts that are sitting there as defaults.

One prompt is for all hands prep and the definition or the.

I guess the objective of that prompt is to develop your messaging and talk track to convey your vision and inspire your employees.

So it's kind of a drop in prompt for a project, presumably for all hands that you could use going forward.

So my suspicion is for people like us where we don't do spending hours building prompts of five pages, that there should be good, healthy, generic prompts for things like all hands or this then the other, that we can simply take it and drop in and use it, where we actually get a better answer than what we would have got generically to the box, but I don't have to think about creating that problem myself in this case.

Yeah.

And then the question is who's going to provide it?

Like is it going to be the rise of a bunch of apps that are basically rap arounds of the labs?

Is it going to be the labs themselves providing it, or is it going to be consultants, or is it going to be like your really talented 22 year old who comes into the business is like, you know that going to be their future, but I don't think it's gonna be the future for very long, because ChatGPT is just going to create a better user interface and eat up all of these other businesses.

Another one that I hear is not doing well is 11 X.

Like the first automated one of these like SDR offerings, they got to 10 million super fast.

For what I hear, they're about to that 50% churn and wasn't a good product.

Not worth it after people gave it a go.

And then also just loads and loads of people are now building their own automated stores.

And so that's another example of there's going to be a lot of crashing and burning while figuring out where the real value is, and that apps layer is definitely at risk and probably like how specialized can you go?

So when you talk about like ones for lawyers, one's for doctors, like, you know, the very specific horizontals are probably safe.

But let's say.

I somehow feel it's like the hype curve where the companies come out of nowhere.

They create first versions of the product around stars or whatever people are.

Budgets are excited.

They spend it, they use it.

It's very disappointing for all sorts of reasons.

Then we go through that phase where all these companies and products go to the trough of like disillusionment and like not getting the revenues, high churn, this, that and the other.

But slowly but surely they figure it out and they figure out how to build real value and they come back up the curve eventually.

So it feels like on balance, I would imagine this will happen in this sector, but there is that kind of risk that sits there in terms of open AI ultimately eating into these stocks in a way where somehow it becomes so easy for a single individual organization to do it all themselves.

They don't actually need to go to these companies to buy these products, to your point.

And it's all happening so fast.

So you have these rises and falls within months, not years.

Yeah, exactly.

So what I wanted to ask you was how to communicate the importance of experimenting with ChatGPT and Claude and know PLM and Gemini across the organization to get people that dropped people's minds around the fact that this is very important for businesses going forward.

And we need to start experimenting.

We need to start doing things.

And obviously, from an operator standpoint, there's all sorts of things you can do to make that happen.

But purely in terms of just straight up communication to the company, what are the kind of like key messages that are the most powerful to just get into people's minds that this matters.

And we need to do something about it.

So I'm just stuck on your mention of Gemini, and I'm using Gemini right now, and I can't really tell if I'm using the paid version of the free version, but whatever I'm using just sucks.

So first of all, I just want to add that in like it is massively underwhelming.

It doesn't help me in any way.

It doesn't do any of the things I would like it to do, and its answers are stupid.

Slam on Gemini.

Basically, everywhere I go on the Google suite, there's a little like, how can I help you box?

And then I ask it to help me and it just does not do anything helpful.

And so perfect example is I took the commission policy, ran it through Claude, made it friendly and pleasant, stuck it in a Google doc.

Then I asked Google to format it like the above policy, and it just rewrote the policy for me in a stupid way.

And it didn't reformat anything.

And it only does stuff in the chat box.

But like, I don't care about the chat box.

I would think that if you are Google and it's integrated, it should do things in my document and not just produce me.

Like all it is, is an integrated chat box.

That's not the power of Google having it.

They need to get to that next level.

When you think about canvas within chat GPT, that is genius.

That works so well in that respect.

And you can imagine from a Google doc point of view, they should do exactly the same thing, and it should be way better given as Google Docs as opposed to some flimsy canvas thing that ChatGPT just created.

So anyhow, that wasn't your question.

So your question was what are the reasons why?

What's the message to the company?

You know, we need have this idea that for employees that this is the future, this is how we win.

Part of that is to take A.I., truly embrace it, Activate ourselves to use it to help us with what we're doing to to get to the promised land effectively, and that this is a winning mindset and don't see it as cheating.

We should be using it and it's very sensible.

So we have for engineering, we've thought through the workflows and the tools to be used and have some guidance is quite specific.

And then for the rest of the business, what we're doing is in every single weekly stand up, somebody does a show and tell on what they've done with Jenny that week and what they've learned and what worked and what didn't work.

And so it can be really high level stuff for what marketing's doing.

Or this week's was somebody in our R&D team going into huge amounts of detail with, I don't know, I think it was Claude's coding program and it's now becoming a genetic.

And so like how it was solving the problem and working on multiple versions.

And when it got itself into a death loop and how they got out of that death loop.

So it was like super detailed and technical, but it wasn't a set training program.

But we are giving people guidance and thinking through more than just a policy on how to use it and what to get out of it.

And then the other thing is we do have budgets to experiment.

So people are looking at windsurf and cursor and the ChatGPT and the Claude and Copilot, GitHub Copilot and Amazon Q so we're looking at like all of the different coding tools and not embedding them, but just experimenting and seeing which ones are good and which ones are shit.

And just as another FYI, nobody seems to like copilot very much.

It sounds like copilot might be a bit like Gemini, where it's just not there.

So Charlie has services that he provides.

The primary one is his AI Inspiration workshop, where he goes into an organization and gets them really not just excited by the possibilities around Cod and ChatGPT, but they kind of work hands on.

What kind of use cases might be interesting for that particular group?

And I think he referenced working in a organization for this workshop with the finance team and with that finance team.

They outline all sorts of possible ways to use it within finance.

That's part of that workshop where they got those finance individuals on their pathway to take those projects and start building them out, basically with projects or custom chat gifts or what have you, I suppose.

So I was kind of wondering, what do you make of that activation of people getting things to happen, workshops like this?

Yeah, I think it's a really good idea.

And also like finance might be a bit risk averse and not wanting to change, but they have some of the most boring, tedious, repetitive jobs out there.

If I were the person who had to, like, handle all of the expenses and accounts, I would be crying out and experimenting to see how I can make my job less boring and let the machines do all of the shit work.

And I think that that's kind of the inspiration is what are the things that are mind numbingly boring that you have to do all the time, and then it's worth taking the time to create the prompts and to create the agent that just does the shit work for you.

And that's kind of like my inspiration, you know, it's like all of those back office roles that are horrible.

They're the first place to look because you just have happier people who aren't just doing awful stuff, and they don't have to do awful stuff the whole time.

But like months end, nobody looks forward to month end in finance.

Yeah, it's a lot of pressure.

A lot of pressure and a lot of boring shit.

So I will report on this at some point in the future, but I'm very keen in an organization I'm currently working with to take all the policies and to stick them into notebook alarm specifically.

So this company uses the Google Suite.

So notebook alum, you can share your notebooks across your organization seamlessly.

And notebook alum seems very well suited for this task which is taking documents inserting them, not using the extended world of alums, whereby gets confused between your documentation and the greater world of documentation around those policies, and also with that podcast interface and different ways to access the information.

And it seems so much more obvious.

I wouldn't say fun, but just like different ways to like allow the employee to get information in a fast, efficient, timely basis.

So I think that as an actual kind of low hanging fruit as a CEO, that seems like just a fun thing to start with.

Yeah, I agree.

Let me know how it goes.

So let's park it here and let's get on to our conversation.

Part two with Mr.

Charlie Callum.

The more that I learn about AI, the more I realize you don't need to know anything about AI to embed AI and benefit from AI, and you need to know about neural networks to benefit from AI.

In the same way, you need to know about nuclear fission to benefit from electricity.

You don't.

You just need to plug in to the wall and consume the electricity.

How it gets to your plug socket is really I have no interest to you, and we very much seen this with AI over the last, you know, sort of 10 or 20 years where it's been, oh, you need to be a data scientist.

You need to have a machine learning degree to be able to understand what's going on.

But November 22nd, with the arrival of ChatGPT, that is the plug socket.

The plug socket is the chat input window.

That is all you need to know what happens behind it.

Neural networks.

Reinforcement learning.

You don't need to know that to benefit from it.

We are all CEOs of organizations that don't want to be left behind.

We understand the importance of AI.

How do we get started?

There's a few things that I would think about as a sort of setting the scenes as an executive, as a CEO, as part of that leadership team.

And that is firstly, just understanding what is our approach here.

You can be an AI pessimist, which is where you're like, oh, you know, I'm not really sure about this thing.

And if we are going to use it, I'm going to be thinking about how do we strip out costs from our business.

How do we become more efficient and do things with less people?

I think of this if you're a company of a thousand people, well, we can get to the same destination now with only 500.

So that's the pessimistic view.

The alternative is to be an AI optimist and go right, we're going to lean into this and we're a company of what do I say, a thousand people and we're now over 500 people.

We're not going to act like a company of 5000 people.

We're going to be able to go further with the same amount of resource.

And that kind of, are we leaning into this thing or are we leaning out of it is a real precursor to everything that you're going to do afterwards.

So having defined right, we're going to lean into this and we're going to figure out how do we act like a much bigger company by using these tools.

Next thing you probably want to do is very quickly codify that into some form of AI policy.

Now I'm very nervous about mentioning AI policies because it's like, oh my goodness.

Like, this is just like, you know, who's going to read it?

What's the point.

Where CEOs, we love a good policy.

You don't need to say that policies are a problem here.

One of the common things that we see is that if you ask companies, you know, what are you doing with AI?

You know, what benefits are you seeing?

You often see quite muted responses.

We're not really seeing it.

We've got some pieces.

There's nothing really that's out there in production.

If you go and speak to the people one by one, are you using ChatGPT every day?

Every day?

I'm using it all the time.

You know, it's where I go to ask any task that I've got.

So where is this discrepancy coming from?

People are saying they're using it all the time, and yet companies are saying we're not really using it after this because people are not telling their manager, their leaders, their colleagues when they're using it.

It's kind of under the radar a little bit.

You know, oh, I'm just going to ask ChatGPT or Claude.

It might be seen as cheating.

And often this is because certainly when I'm speaking to teams, I'm not sure what the rule is.

I'm not sure whether I'm allowed to use it, so I'm definitely doing it because I know that it gets me where I want to go faster, but I'm not being very public about it.

So one of the things that you can do with creating a short, pragmatic, positive AI policy is we are leading into AI.

We encourage your use.

We encourage you to experiment with new tools, but be sensible.

Here are some guardrails that you might want to follow so that you know we're not sharing confidential data or so on, but having that policy where we're setting what's right and what's wrong, but reinforcing to everyone we are leaning into this and we want you to lean into it is it's a really good approach.

And I can't remember if I asked this question before, but it's something I really want to know about.

And it's probably like the next layer of depth rather than like, should you have a AI policy?

But being in tech, everybody says that 2025 is the year of a genetic.

It's all about a genetic AI.

And yet we mostly just talk about the chat bot and using ChatGPT to do a bit of thinking for us, rather than to actually start to automate our lives and be proper agents.

I really want to understand about a genetic, because it's like the year of people creating agents to sell to other people, but if you're in a business and don't want to buy 500 million different vertical agents, how do you start to actually automate with agents and are there any good platforms to do that?

Agent says the password.

So first let's define what an agent is.

And as everyone's talking about it, suddenly it becomes just like this fake.

No one's quite sure.

For an agent to have agency, it has to have the ability to choose what tools, if any, it is going to use to complete the task.

The opposite of that is what we might think of a workflow or an automation.

So if people are building things with it's happier if they're building things would make if they built integrations with Boomi or Informatica.

If those things are predetermined, even if it is an alarm call and then another alarm call that is still more of a workflow or an automation because whatever happens has been predetermined.

The true agent is that I've got this alarm tool and it is able to decide now I'm going to go and check a stock price.

Now I'm going to create a document.

Now I'm going to go and ask, you know, another alarm for for something else.

And I'm going to decide because I've got the agency about what to use and when.

And so for any company that is thinking of building agents or is being sold an agent by someone else, that's the important thing to dig in.

Have we really got an agent which is able to make decisions, or we just got a workflow?

And why that's important is that if we're saying 2025 is the year of the agents.

What does that mean?

We start to think about true digital workers, digital employees that can make decisions.

So if we I don't know if someone working in HR as an HR administrator, I'm able to as a digital agent, I'm receiving a request or a job posting request from a manager.

I'm then able to go and use another tool to go and write up the job description.

Then able to go and decide if we're going to have that budget and when we're going to post it, I'm then going to choose which job boards we're going to post it on.

Then maybe if and I'm going to choose which candidate I'm going to shortlist and get invited in for an invite, that would be one or a series of agents that are making decisions as opposed to a true workflow.

But that's when you're starting to really replace humans.

So that's a great definition of agents.

This year is supposed to be the year of a genetic AI.

By the way, genetic is a word that's been basically made up.

That means agent, IC agent.

Like now everybody in here, like I think I saw a genetic for the first time about nine months ago.

And I was like, what is this word?

Is it a typo?

And now, like my entire life is about a genetic.

It sounds great, I want some if this is the year of a genetic, I said, this is a year that all software companies create agents that other people can buy.

Or is this the year that I can suddenly have a million agents in my business by using ChatGPT?

I'll answer that by talking to maybe some of the steps that a CEO or a company might want to go through when embarking on trying to make that decision, both on anthropic docs, website documentation, and also on eyes.

They talk about passing the the intern test, which is you should think of I as a brand new, eager, but poorly informed intern.

And if you provide that intern with a badly defined instructions, you will get what you deserve in return.

The original request is quite poorly defined, and that's often because the person that's doing the job today has been doing it for a while.

They kind of know it.

There's lots of sort of assumptions or well, I'm sure you know what that means.

And this is often why you see in companies that when someone asks, you know, how do I do this?

Or, you know, need and you can speak to Gary.

He's been here for 20 years and he knows how that works.

You can't take what's in Gary's head and just give that to the LM, because there's so much of Gary's experience and knowledge that's wrapped up in there.

So I find myself writing a lot of long, prompt documents that are maybe anything from 10 to 20 pages long, which is first we do this, then we do that.

Then we do that.

This is what that means to pass the internal test.

If I gave that Google doc to an intern, could they do what I'm asked them to do and I would get the output?

And with anthropic and OpenAI, when they say pass the intern test.

If the answer to that is no, that if I gave that document to an intern, would I get the response I wanted?

Then how is an El Alam going to do that?

It's not.

So I'd say a large proportion of the time that I spend is in getting the ask correct.

Once you've got the ask correct, then creating whether as a custom GPT or whether it is building out some form of agent that is able to access different function calls, that becomes relatively easy, because what you've said to the intern is, let's do that HR hiring example, right.

The first thing that you're going to do is you're going to be on the lookout for job requests that come from hiring managers.

The second thing that you're going to do is you're going to take all the details of that job request, and then you're going to start drafting up a more detailed job description.

This is what our job descriptions should look like, and all of the details of what should be in each section.

And then you're going through all of that.

We use some examples of how that H.R.

Hiring person would have the choice of choosing which job boards to post on.

Well, how do we make that decision and how do we access those job boards.

And that's really the instructions that you're going to end up providing to the AI.

Now, if you don't need to access any tools because the human wouldn't access any tools, then you're thinking more of just a simpler implementation.

If you are looking at accessing tools or you need to access your HR system or your payroll, then you're thinking more of, I need to create some kind of agent that can choose when to use these tools.

And how do you create the agent?

That's my question.

My recommendation is a tool called the vessel AI SDK.

So vessel VR CEO and vessel is a web hosting and deployment company.

So anyone that's an AI product management product development role may have come across V0.

So that's a v not dev.

That also comes from vessel.

And that is a AI generative UI.

So you can design your applications.

But there's vessel AI SDK is a really neat way of plugging in to open AI and to anthropic and to Gemini and to any of your preferred LMS, and it gives you the ability to create these calls out, these tool calls, which is what makes an agent.

Well, because I guess if I step back to what I'm thinking about and maybe it can help is in the past, and particularly as my experiences as Crow, which I complain about bitterly.

Not the crow part, but the number of tools I had to buy.

Because you had this tech stack, the revenue tech stack and everything costs like 20 K other than Salesforce, which of course costs shitloads more.

But it'd be like, okay, now it's 20 K for this and 20 K for this and 20 K for that, and you add it all in, and then suddenly you had a tech stack of, I don't know, 20 different things and half 1 million pounds.

What I don't want to do when I'm looking at my IT stack for the future, is to have that 20, 30, 40 pieces of tooling that are critical for the business, and some of them are big and some of them little, and they're all whacking some AI on, so they all have a chat bot.

Nobody seems to have a particularly compelling vision.

I know that we are not all that impressed by GitHub copilot, you know, versus Verses like cursor and some of the others.

It's I don't want to blindly buy all of these tools that say they have something, that maybe have an agent, and I still just have to pay shitloads.

What I would like to understand is how to think about what the future tech stack should be for a scale up.

So a couple hundred people to a thousand people.

Technology's moving really quickly.

This is the year of the agent.

What should I be buying today or what should I be evaluating?

Because my ideal and my vision, and I'm guessing that other people would have this vision, is that everybody has their own assistant who can do all of this shit work, and it's really easy.

And you just say to your assistant, can you look through all of these CVS for me and stick them and highlight the best ones, or contact them or, you know, and be able to write The foreign intern prompt that's 20 pages long.

But once I do that, I never have to do that work again.

And I want everybody in my company to be able to do that, not just the individual who understands how to use VSL or API's.

How far away are we from that and what are the steps to get there this year?

It is such a fast moving space that it is just impossible to say, you know, use this tool or use that tool because so much is coming out, whether that's models or whether it is tools that are set over the top of these models that help to create these workflows is speaking to a CEO.

This would be my advice is that there is never been a better opportunity right now to internalize some of these skills.

Previously you might have said, you know, build versus buy while the building is just too complicated.

We're not going to go and set up some massive infrastructure and start investing in people to run that.

So we're going to buy we'll buy a CRM, we'll buy an air platform, whatever that might be.

But I think we're seeing now that pioneering companies, and that doesn't have to be just like some funky startup.

But actually, I mean, you saw this with Klarna recently and saying, you know what with some of these tools, whether that is universal, whether that is V0, whether it is a new development platform that people are going to, you know what, I can take non development people.

So a product manager or someone in an operations team and I can start building out proof of concepts and building internal products without having to go and get a SAS subscription from someone else.

And suddenly I'm relying on their roadmap rather than our own.

So I think there's going to be, you know, some companies that just don't get involved in this at all and just stay.

Let's leave AI aside.

Then there are going to be some that go.

We want to buy stuff.

So let's rely on someone else's a platform for stars or for marketing or whatever.

But I think there's going to be an increasing number of companies said, you know what?

We have got insight into what are our challenges.

And we think it's quite unique to us.

And we're just going to have a go at building something internally using these publicly available tools.

It's about using as many tools as you can come across, as they keep getting developed to improve your learning and the way of working.

This all feeds into that first bit, which is about, you know, AI positive or a negative AI going to lean into AI or lean out of it.

And again, checkers is important.

So having done that then the primary driver of a good change program, and everyone will know this regardless of what it is you're trying to change, is having empowered and inspired senior leaders that are at the top of that program.

So I would then be looking at who's the rest of your executive leadership team.

So if you are the CEO, I'd be looking left at the CFO.

I'd be looking at that chief Revenue Officer.

I've been looking at the Chief People officer and a level down into those VP's and figuring out quickly, how do we get those people comfortable with using AI for their own tasks?

It is so important to have that team really understand how I can help them personally, because that immediately filters down to the rest of the team.

So you've got to get that team on board so that could be just running.

So inspirational workshops for those leaders.

Maybe have a breakout session at your next leadership offsite.

And it's not about encouraging them about how their teams should be using AI.

It's about how do you use AI?

If you're a c o, you have a set of direct reports that you need to manage and inspire.

You have got colleagues who run finance or on HR, or who run legal.

And how do you better understand their view of the world as a CEO?

Or you may be just going through a merger or acquisition, and you may be having to figure out, how do I align this whole new set of people and processes and data into our organization?

So it's thinking about how does ChatGPT, how does Claude, how does Gemini, how does that solve your personal, daily and monthly working processes?

And if you can figure that out immediately, it gives that permission to everyone below to explore and experiment as well.

I tried to just have one subscription, but I now end up having to have two.

So I have both Claude and Betty because I just find Claud writes things better.

But ChatGPT has a wider range of capabilities and functionality and also is attached to the internet, and I use it when I'm stuck thinking, but and kind of in those use cases you talked about and like the team use it for ideas for deal reviews or how to write a better proposal.

But what I really want to use it for is all of the hard things that are boring.

And I hate like org charts, but I ended up spending about 2.5 hours trying to get it to write me an org chart, and I could not.

And I was asking it to help me figure out how to tell it, to write me an org chart, and I could not.

Is this what it's worth, going for some sort of AI training course?

Or is it just really bad at making org charts no matter what?

Because for whatever reason, they're incredibly difficult.

So there's some basic tips that you can follow.

So whether you're using code called ChatGPT.

Gemini.

Any of these tools.

Firstly, in both Claude and ChatGPT, you can set up a project and set a project as a wrapper around a task or a set of chats.

So as a CEO, if you're doing anything repeatedly.

So that might be a set of chats about one of your direct reports.

It might be a set of chats about an acquisition you're going through.

It might be a set of chats about a policy or process that you're crafting.

Create a project.

You then upload background information so that could be documents.

It could be just your instructions about how you want ChatGPT to respond to you.

And then you get into how do you craft a prompt.

And I'll give you some specifics here.

So my typical prompt, especially something when I'm reusing it, is anything from 6 to 8, nine, ten pages long ago.

That's that's longer the little chat window there.

But because I'm, I'm using that prompt again and again, it's worthwhile.

So I break it up with XHTML tags.

Now you don't need to be a developer to know this.

I'll just explain it super simply.

If you don't want to add XML tags.

So a left arrow kind of opens it.

And then you write the word objective and then a right arrow to close it.

So that's your opening tag objective.

And then to close that XML tag you do exactly the same, except there's a backslash up at the start after the first opening bracket.

And I'll put some of the notes about this prompt and guidelines so that you can have it in the show notes.

Now when you're writing your prompt you're basically structuring.

So I say objective.

This is the first part.

The objective of this prompt is to develop a job description or a merger plan, whatever it might be.

Then the next tag might be open the tag instructions.

Right.

This is how I want you to do it.

The user is going to provide you with this information.

I mean, I'm the user, but I'm going to provide you with this information.

I then want you to ask me some questions.

And here are the questions which I put in another question tag.

Then you get to the two most important parts of a really great prompt.

And they're often the hardest bits to put it in.

And this is what takes their long time and makes it a ten page document.

So examples.

So think about the intern.

If I ask you to build me an org chart, but I don't show you what a good org chart looks like.

2 or 3 of them, you're just running around in the darkness.

If I don't show you a bad org chart, you don't know what bad looks like.

So whether it is, you know, a job description, whether it's a spreadsheet table, whether it is a project timeline, whatever it is that you want.

A good prompt should always have at least 2 or 3 examples, a good one, and then a bad one.

So that can take up quite a bit.

The final tag that I always have in there is exceptions.

So what should the AI do if you don't have certain information that you've asked for?

So for example, if I'm drafting a job description and I've asked the AI and the instructions to ask me for the salary range, let's say I haven't got the salary range yet.

Well, what should the AI do in that situation?

You might say if the user does not have a salary range, then know that this is to be verified later and at the bottom of your response, put some next actions or follow ups or whatever.

So this sounds quite detailed, but if you're doing one to ones with your direct reports, you do that regularly and you're going to do it a lot.

So it's worth spending an hour writing the prompt so that you've got that in there.

So there's just a bit of guidance.

If you put the the right in front to really pass the intern test.

If you gave that to an intern, they'd give you a good response.

Then I think you'll start to find you get better.

Whether that's an org chart has on that, because maybe you're trying to get a visual, which it probably wouldn't be very good at.

But in terms of the hierarchical structure would probably be pretty handy at doing that.

We in our company kick off this year, we had everybody working groups to identify use cases along the customer journey where applying.

I would be really cool.

And every single team, they came up with loads of different ideas, but every single team came up with one idea that was the same, which is basically putting all of our internal data in a data repository and putting ChatGPT on top, or a GPT on top.

So like all of the historic slack information, all of our Google drives.

Then you have Salesforce data, etc., then be able to from that repository, ask like, give me a summary of what the customer relationship has been like to date.

Give me a summary about like usage of that customer, all of those types of things.

So that we go think about the build versus buy.

I'd be taking a look at glean.

I'm not sure if you come across Glen GLE and.com.

And it's exactly that use case.

So founded by some ex Googlers.

How do we use AI to search across our existing data sources.

And at first glance you say, well, you know, the challenge is quite easy because we just go and, you know, plug it in via APIs to all these different things, but you need to preserve the data privacy rights of the source material.

So if I go search customer X, how much did they spend last year.

Well, I should only be able to get that answer if I would be able to get that information in the actual source system, because I've got access to that data or whatever.

Anyway, so I had a chat with one of their team just before Christmas and it was a super interesting use case.

But that's exactly what Glenn is targeted, that.

When we first kind of reconnected and started talking and you talked about your Power Hour, and I think I have told everybody I know about it, and I thought, why not share it with the rest of our listeners?

What's your power hour?

So my Power Hour started during Covid, and like many of us, I was a commuter before Covid, jumping on a train.

God knows what hour into London.

Covid came along and suddenly I had all this time in the morning and my wife would take the kids to school and, you know, I'd be up and about, you know, having taking the dogs for a walk.

7:00.

And then the kids are off.

And then I had to sort of hour from eight till nine when nothing was really happening.

And so I decided that I was going to start to write a book during that hour.

And so I would just write, you know, a chapter in an hour, having thought about what was going to write in the shower.

And suddenly I found it was like super productive.

Like by the time I got to 9:00, I had done a ton of work and probably my most creative work of the day.

I was lucky enough to finish published that book, and I just kept that power hour going.

And it's the hour that I work for myself, whether that's on sort of personal writing, whether it's recording a YouTube video.

And I'm very humbled that I get lots of lovely comments from people like, I've got no idea how you create so much stuff.

You know, there's always a new e-book that you've published, or you've recorded a video for YouTube, or you've done a podcast, or you've published a book and you know, how are you doing all this stuff?

And you've got four children and two dogs and you've got, you know, your work and everything, and it's that Power hour has been super helpful.

And once you get into the habit of it, then it's very difficult to break.

It doesn't actually seem like work anymore.

It's just a routine I find myself.

Weekends included at my desk from eight till nine.

Just crank through some great work.

The book that inspired that was Atomic Habits, which came out pre-COVID.

I'm sure many people have read it was an Amazon bestseller, but it was all about, you know, if you create the framework

at 8:00, 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.