How To Stop Using AI Like A Search Engine, with Katelin O'Shea

In this episode of The AIQUALISER Podcast, John Bennett talks to Katelin O'Shea, who is a Project Manager at Dropbox, a Perplexity Business Fellow and has created a consultancy and training business called AI That Works.

Katelin saw AI being used in a fintech start-up in 2022, so she was already aware of the possibilities when ChatGPT launched. She put that to good use at Dropbox, where she has built systems that take a project scope from half a day's work to twenty minutes, and freed herself up to actually listen in meetings rather than take notes.

She explains what context actually means in plain terms, why rolling out AI to a team is just as much work as building the tool in the first place, and where she draws the line on what AI gets to do.

The episode closes with a question Katelin gets asked a lot: if you could only focus on one area of AI right now, what would it be? Her answer is skills, the reusable instruction files that document your processes in a format AI can act on, which can be built without any technical knowledge.

In This Episode

• Katelin's introduction to AI before the ChatGPT era, and what working with an early agent taught her

• Why people with management experience tend to get better early results from AI

• Transcription, structured folders, and custom AI tools: how Katelin manages projects at Dropbox

• The difference between using AI as a search engine and using it as a system built around your work

• What context actually means, with examples that require no technical knowledge

• Why AI rollouts fail, and why individual pain points beat company-wide mandates

• Filtering LinkedIn noise: building a system that turns saved posts into original content

• What Katelin will not let AI do without her own voice in the draft first

• Where to start if you have not started yet

• Skills: reusable instruction files, portable across tools, no technical knowledge required

Chapters

• 00:00 Introduction to Katelin O'Shea

• 03:53 Before ChatGPT changed everything

• 08:29 Inside a project manager's AI workflow

• 20:23 Context isn't technical

• 27:20 Why rollouts fail

• 36:40 Filtering the noise

• 44:00 Keeping the human in the loop

• 48:48 How to get started

• 57:40 Listener question: where to focus

You can find Katelin at aithatworks.io, on YouTube at @aithatworks_kate, on TikTok at @aithatworks, and on LinkedIn at linkedin.com/in/katelinoshea

If you have a question you'd like us to pick up in a future episode, you can get in touch at frmdb.ly/pod

Transcript

John Bennett (00:17)

Welcome to the AIQUALISER podcast where we try and balance out the AI noise with conversations about how people are really using it. And today I'm talking to Katelin O'Shea, who's a Project Manager at Dropbox, a Perplexity Business Fellow and runs her own AI consultancy. Hello Katelin.

Katelin (00:33)

Thank you so much, John, for having me on. I'm happy to be here today.

John Bennett (00:36)

Great, well it's great to have you on. And obviously we're get into AI, but first I thought maybe you can just give us a bit of background and how you ended up juggling AI consultancy, working at Dropbox and everything else that you do.

Katelin (00:48)

Yeah, well it's been a few years but I started off in ops and project management. My whole career has practically been ops and project management and I came into Dropbox around two years ago now but prior to that I had started working with AI

before joining Dropbox. But as we all know, the last two years have been a very busy time for AI and things have kept moving at fast, fast paces. And I guess working in a place like Dropbox, you get access to all sorts of tools and it's kind of hard to turn away from AI considering the opportunities that we have in big tech for applying different types of technology and the things that we do. So I have definitely done that. But as I said, I started using AI prior to that when I was an operations manager.

So in 2022 I was working with a startup and they were actually using LLM or a very smart agent to be able to help create debtor solutions between banks and private debtors. And I was just amazed by

how great of a solution this was. It was really helping the person on the other side come up with debt repayment solutions by using this agent that did all the negotiating on their behalf. And that really just like lit up in my head, I was like, wow, what is this LLM thing that they're using? And I think it was about three months later, ChatGPT came about as well. And this was kind of the, this is gonna change everything that we know. I think to be honest, because I had had that exposure with that particular use case,

seeing what it was already doing.

I just couldn't turn away and I had to start experimenting as an Ops Director within a startup. Everything was about efficiency and making sure that we could actually deliver the best results to our customers. And I started actually testing and implementing within our day-to-day processes, just with basic content creation at that point. So we were actually a grant funding consultancy. So we worked with startups to help put together very lengthy proposals that would get submitted to the government. So this was content creation.

was the primary area that first started to see really great results, especially from a non-techie person of using AI at that point. And we were able to bring down our...

processes that we were actually completing on a day-to-day basis probably by about 30 % in time. So we were able to get those proposals out faster, we were able to edit faster, we were able to write them faster and from there it's just meant...

I've just looked at how can I use this in every sort of way, my personal life at work as a project manager. And then it has evolved into me starting my own company, which I never expected to be doing, but I kind of fell in love with it. Call me nerdy, call me whatever you want, but I absolutely love using AI. And I think it's a brilliant thing if used in the right way can be such a great value driver for all of us, personal life, work life, whatever it is. So I like talking about that and I like helping people around that. So yeah.

John Bennett (03:53)

Brilliant. And there's quite a few things that came out what you're just saying there, but one that I really wanted to pick into, which is really quite interesting. So realistically, you were using AI before that kind of moment in 2022 when ChatGPT came out and we were all jumping on the same bandwagon. So that's really interesting because obviously you've had an experience of it kind of prior to where we all started from. how did that kind of feel in other words, comparing the

Katelin (04:10)

Yeah.

Yeah.

John Bennett (04:23)

bespoke tool that we're using to this thing that we all had access to. What were the differences?

Katelin (04:29)

Well, I guess if you think about it, they had created essentially an agent in 2022 that was utilising LLM technology. Like there are a lot of people now that still haven't got past using AI as a chat functionality like Google. you know, back then they had already created an agent that was able to perform tasks autonomously, use the LLM to communicate back to the banks, back to the private debtors. it was like magic.

I

was like, how is this even real? And then to be given access to a tool that yes, at that point was basic in comparison to what it is now, but it's still, as I say, it was still a huge unlock for me straight away, utilising it as a tool at work. So I guess it probably did instill a bit more excitement for me than the everyday user because, you know, like it takes time to start using AI and getting value from it, especially back then because it's

as I said, it's come a long way. We all know it's come a long way. ⁓ But because I'd seen what they had done with it and then what was possible, yeah, it was very hard to turn away from.

John Bennett (05:40)

Do

you think that kind of gave you some kind of, if you like, not necessarily advantage, but some kind of insight into how to make chat work for you back in those early days?

Katelin (05:48)

No, I don't think so, to be honest. think that it just gave me the excitement. It gave me the what's possible kind of look in you could say. And yeah, maybe that was a little bit of an advantage because most people were only going to be able to source that kind of stuff from YouTube and so on. And how much of that is true or just made up. I was seeing it in the real use case scenario, helping people. But I think majority of what I got from it back in those early days is stands true to

to exactly the same thing for anyone using it now. You have to experiment, you have to try, you have to fail, you have to iterate, and you need to continuously improve your systems that you're applying AI to. So I think majority of what I got back then really just came from playing with AI and then learning from those sessions and applying them in my work processes.

John Bennett (06:41)

Yeah, I think that's really interesting there, isn't it? That so many people for some reason kind of expect AI to be perfect. And yeah, that concept that you just talked about, about, you know, iterating, playing a bit and expecting it to get things wrong, but, you know, working with it.

Katelin (06:49)

Yeah.

Yeah.

yeah i think it comes down to the fact that

We have never had to essentially describe what we do. We all have learned what we do over the many years that we have been doing whatever it is in our jobs. Let's apply it to that. we've never had to describe that really to anyone, a lot of, unless you're a manager and they do say that managers tend to get better results on the first step with AI, then people that are not managers, say you're a subject matter expert. You don't really have to explain what you're doing to anybody. You just get your job done and.

everyone's happy. Whereas if you're a manager, you have to delegate, you have to describe the task, have to explain to your workers what good looks like, what are we looking to achieve here, what are we not looking to achieve. I think that's the reason why managers do better in the beginning and that obviously fell well with myself as well being a project manager that I had those skills and that got me good results probably quicker than others would find. But yes, you are 100 % right.

that you have to be willing to test, have to be able to iterate. And I think you have to be able to learn that how to communicate with the AI. And that is essentially describing what it is that you want to get done, giving it references. It's really breaking things down. I think that's what some people get stuck with a little bit and maybe don't see the results that they're hoping to get from this magical tool that's supposed to do everything perfectly.

John Bennett (08:29)

That's great advice. So how have you kind of used it in Dropbox? How has it kind of changed the way you do things there?

Katelin (08:37)

well in Dropbox, it has meant that I've been able to really minimise the amount of time that I spend on the administrative part of being a project manager. And if you know anything about being a project manager, there is a lot of admin. So we take very detailed minutes in meetings. We have to put together project scopes. We have to put together project plans. We've got Kanban boards. We've got, we've got all sorts of different documentation that we have to be able to keep up to date throughout the delivery of.

project. Now I have been able to automate a large chunk of those processes and what that has meant for me is that I can sit and deeply listen with the people that I'm working with on my projects. I can have deeper conversations with them instead of taking notes and making sure that I've got all the actions there and you know that the Kanban board updated with the dates that they've communicated or the the detailed technical terms that they've shared with me not noted down.

I can be just listening to the things that they're actually trying to communicate with me and be coming up with solutions there and using my brain power for that rather than the administrative. So that has been.

It's something that you don't see results in from the beginning because it also is a process or a change in behaviour that you have to go through because you come up with solutions, you test those solutions, you see that they're actually going to save you time. But from the beginning, you're not actually confident in those solutions. So you still tend to go back into your usual behaviors and you don't tend to get the real return from those new solutions that you've come up with until you're fully confident. And now I think.

having been about a year and a half of having a lot of those solutions in play, I have been able to really at least 50 % of my time I've gained back to be able to apply strategically and start going about my projects in a completely different way as to how I would have before using AI in my processes.

John Bennett (10:39)

I love that. So on the last episode, I was talking to Victoria Westcott and Victoria does many, things. But one of those is film production. And she was talking about, you know...

Katelin (10:48)

Yeah.

Okay.

John Bennett (10:53)

using AI in the areas that aren't your zone of genius, and keeping it out of your zone of genius. And I love what you just talked about there, that you're using it on that kind of mundane and repetitive stuff to give you the brain space and the focus to be able to really listen to people and really engage. I think that's wonderful. And how do you do that? Do you do some kind of AI note taker or how does it do that for you.

Katelin (11:04)

Mm-hmm. Mm-hmm.

Definitely. Yeah.

Yeah, it's a combination.

Yeah,

definitely. it's common, I'll speak through, guess, one of the, one of the first most tried and tested parts of the way that I use it, currently within my project management. So what I have is a project folder structure that I have templated and I use it across all of my individual projects. And that's where we start every project. So at the beginning of project, we have to have discovery. So we have to ensure that we document everything we need to know about the project, what we're looking to achieve, what are the different areas that we're looking to deliver.

upon what are the tasks in all of those pieces and that essentially is gathered by having a number of different interviews with all your different stakeholders on the project. So I record and transcribe all of my calls. Not one call goes ahead in my project without it being recorded and transcribed. If you're working with clients that are worried about confidentiality you obviously have the opportunity to have an NDA signed. If you are using it in

In a highly-governance area of business, you can run your AI systems in-house, on-site, to ensure that you're keeping that security high as well, or is as high as it possibly could be. And by using those transcripts and those recordings, they are fed into that folder structure. So it starts off, obviously, in the discovery and ideation section. Those interviews are then all fed into a custom GPT of sorts, or a project.

or a skill. I now use skills, you can start with custom GPTs, you can start with gems or projects. And they understand that that project understands how I write a project scope. So once I've had all of my interviews with all my stakeholders, I'm then able to connect that custom GPT to that folder structure and it uses it to be able to write my project scope for me. So putting together a project scope could have taken me before

depending on the size of the project, could be either half a day's work or it could be two days work. Now I've got that down to probably around at most, it would be an hour and a half's work, but majority of the time it's about 20 minutes. And that's essentially because you're able to give it the knowledge that it needs in order to deliver the output, which you have defined because it's got access to that very contextual data that is related to what you're doing. So that's been a huge win. It's a huge time saver. And once you've built it,

John Bennett (13:30)

Wow.

Katelin (13:47)

just keep using it and the return on investment continues so it's amazing.

John Bennett (13:50)

Yeah,

and this might be an unanswerable question, but the times when it takes you 20 minutes versus the times when it takes you two hours, what's the kind of the difference there? Do you think it's something that it's picked up on or is it just the complexity or?

Katelin (14:04)

Nor it would

Correct. Yeah. It'd be down to the complexity. would be down to, you know, I guess there could be nuances between different conversations that may crop up into something that, you know, you're not, not a hundred percent on, actually have to dive in and, and, and go back through a few of the transcripts to pull in some fine detail. Like when I'm saying, if it's a very technical project, there may be terms there that the LLM may not understand and may get a little bit wrong. So it's not to say that you can just step back and it's all done with.

without

you having to actually understand what was being said, know what was being said and you know, not have to edit. You definitely have to be very careful to sit down to look over everything in very careful detail before you put anything out. I think that's one of the big mistakes that a lot of people have made. seen some funny examples. Yeah, definitely don't put anything out into the public sphere without fully checking what you have done. You'll still save time, but it's a very

very important step unless you want to look like an idiot but you know that's up to you.

John Bennett (15:08)

Yeah. Yeah. I always say that

for me the watchword is, know, would I stake my reputation on this piece of work? you know, because the AI has got no skin in the game at all, has it? You know, it can be completely wrong and it just, you know, has no comeback at all.

Katelin (15:18)

Exactly.

No.

No, exactly.

John Bennett (15:28)

Yeah,

that's really interesting. So, and how else do you use it at Dropbox? So obviously you use it for that kind of the start of the project and then kind of what happens from there.

Katelin (15:37)

Yeah, I guess so that continues throughout the project. So every week we generally have a delivery meeting where you'd have your working group. I would be using the transcription and the recordings to then save again into the different part of the file structure. Once we get into our meeting section of that file structure, the AI is constantly connected to that live documentation, which is brilliant. So a Dropbox, we have a tool called Dash.

which we can connect into our Google Drive or our Dropbox and we are able to, you can actually ask it questions. Say a question came up in regards to a change that applied to the project and you can't exactly remember who it was that made that decision or when that decision was made. Because it's essentially connected live to all that documentation, I can actually just ask it questions and it will pull me the answer from all that documentation. So that's one way that is a great way of using it as a live source to get

get

answers to your questions that in the past would have, if you didn't have it noted down, generally not too many people would remember the exact details. So it could be just lost. Whereas now that is not the case, which is brilliant. Then it's Slack updates, it's follow-ups, it's communications that would have taken me probably a couple of hours to write now, take me, it could take me half an hour when I'm saying communications.

I mean, say for instance on a project, you've got a bit of a, you've got some ambiguity between two stakeholders or you've got an area of frustration that needs to be worked through or you've got timelines that need to be adjusted and you need to, you know, prove the business case behind that. Sitting down and speaking with your project management specialist that you could build that's built on your frameworks, how you make decisions, your references of what a good project plan is, the decisions that you've made in the past.

can literally sit down with that project manager expert and you can ask, okay, so this is the situation, blah, blah, blah, go into as much detail as possible. I always dictate, I would use the little microphone that you click and you can just chat away to the AI.

that is a huge time saver because it means that you can give as much context as possible. And I literally just talk through the situation and then at the end I go, so help me craft this into a communication that I can send out to, is it senior stakeholders? Is it, you know, mid-level stakeholders? Give it all that information. And a lot of the time, yes, it comes back with something that isn't exactly just hit send, but it's giving you a great framework to work with to then get it out in 10, 15 minutes.

rather than, you know, we all know how frustrating it can be to have to pull together all this information and then craft it into a communication that's not going to offend someone or, you know, be a little bit too brash or all these sorts of things. So that has been a huge win as well using it as a project manager in that sense.

John Bennett (18:39)

That's great. great. And one question, this might not be relevant at all but so obviously if you've got an online meeting, you know there's loads of tools you can use to record it. Do you use it in any kind of in-person meetings at all or is it always online?

Katelin (18:55)

So Dropbox is a fully remote company. So all of our meetings are online. ⁓ I have not used any of the personal dictation tools that I've seen are out there. I've seen there's a few that you can click onto the back of your phone and all these sorts of things. To be honest, I do majority of my work online, but I would see the reason for these types of, you know, in the exact same sense that...

Okay, data is so important and whatever data that you can capture in order to give as knowledge to whatever tools that you're going to build to help you in your business or your services or whatever it is that you do, the more of that you can capture to dissect into references that the AI can use, it's going to craft that natural copy or duplicate of who you are and how you do business. So capturing it in person would be

huge benefit. I myself haven't done it as yet.

John Bennett (19:56)

That's really interesting. Really interesting. That's a really good point as well about data, isn't it? The more data that you've got, it comes back to this magic box thing, doesn't it? a lot of people think it's just this magic thing, but actually it's the context, it's the data, and it's the structure that you give it that make it work. Interesting.

Katelin (20:06)

Yeah.

Yep, it is.

John Bennett (20:23)

Yeah, so ooh where do we go next? One thing I want to talk to you about was LinkedIn because...

I love your LinkedIn posts. I really enjoy seeing them. And I thought it might be nice just to pick up a few and chat through them if that's okay. Yeah, great. So let's have a look. One that really sticks out to me is you've talked about how most people use AI like a search engine, whereas realistically, the people that using it best aren't doing that. So can you talk a bit more about that?

Katelin (20:30)

Thank you.

Yeah, cool. Let's go for it.

Yeah.

Yeah, definitely. I guess search engine, let's think about it, you type in a question, you get an answer. And you could be doing that in any of the large language models. They've got that little type in section and you just go from there. But I guess the main thing that is going to change the outputs that you get rather than just being an answer, being a generic answer is that

like we talked about previously in some of the other questions is around context. So context is what is going to determine

the output that you're going to receive. So it's going to go from a generic answer that the LLM, so the large language model is going to pull from the world of training that it's based on and giving you an answer that it feels is good for what you're looking to actually get from it. for instance, let's talk about, say for instance, you wanted to create a birthday card for your mom. Like you should provide to it different examples of maybe birthday cards that you like.

that you've seen in the shops, take a photo of them. Well, take a photo of that one. I like that. And provide those, upload them as attachments. And then it will have a better idea as the type or style that you're looking to achieve in that output. Now, it's not to say that you won't get good outputs when you don't do this, but you will get better, more personalised outputs when you do. thinking about context in a way that, try not to overcomplicate it. Don't think of it as a

technical term. Think of it as like I just said, photos that you took on your phone in the shops, different examples of emails that you had written that maybe you felt really showed your, the way that you like to deal with challenging situations on your projects or whatever it is that you're looking to get an output for to use. You need to give it examples. And the more you get used to providing examples,

the better the return is going to be from the AI. And now,

If you want to go a step further, you could think about the repetitive tasks that you may be complete at work or at home. So for instance, I'll use a personal use case that I have seen great value from. I have built myself a tool called Mrs. Meal Planner. And what she is, is she is a very basic chatbot, you could say. She was built on a tool called Poppy AI.

Hopi

AI is a very visual tool. It's kind of like Miro and ChatGPT had a child and it allows you to build out chatbots visually. So I can have little sections of like, I've got my breakfast, lunch, dinner and snacks, and I'm able to drag in the, I'm able to type in text, which is literally just recipes from my memory. I'm able to drag in TikToks, YouTubes, and that is all connected up to some systems instructions. So they're literally step-by-step instructions.

of what I want the AI chat to do with all of those recipes. And each week we sit down and Mrs. Meal Planner in about 20 minutes plans out all of my meals, breakfast, lunch, and dinner for me and my husband, creates my shopping list. It now has a connection to Notion. It sends my shopping list straight to Notion and I go into the shops and I'm eating better. I'm saving on my food consumption because I don't waste as much anymore. And, you know, when I say take it to the next level, it's think about

those repetitive kind of tasks, build out a chat bot or a custom GPT they're also known as or a project or a gem depending on which tool you're using and fill that with context that's related to delivering the best output that you're looking to achieve for whatever use case it is. So that's where I think you go from using it as a search engine to a tool that's going to continue giving you results and be much more crafted to what you're looking to get.

John Bennett (24:51)

Brilliant

Superb.

I love the I love you meal plan example, but there was some there that you said that's that's really good. One thing I find with AI is that there's all these terms that we use and particularly, you know, those of us where we're in it all the time. We say these things like context and yet you've just completely demystified context there because, you know, it's another one of these words that we know what it means, but actually we could get hung up couldn't we I've to get the context right. But the way you describe it there is great. know, it's not a difficult

Katelin (25:16)

Yes. Yes.

John Bennett (25:32)

thing. It's just what is the sort of thing that you want it to do.

Katelin (25:33)

No.

No. And I think this is a huge, it's a huge thing that holds people back. And that's the reason why I created AI that works. It's working with people. I'm a non-technical person. I don't have technical understanding. I never went to computer science school or anything like that, but we have to get over, we have to understand that AI doesn't need to be technical at all. And yes, there are these terms, but we at AI that works try to break those down as, best as possible. And

Think about like in the simplest terms, if you are like confused about how to build a system or a workflow.

for something that is repetitive in your day to day. You can simply ask, you can turn on that dictate function, talk to it, explain what it is that you're looking to help or get AI to help you with and say to it, how could I actually bring AI into this? How could I make it more efficient with AI? These are the tools that I have, have subscriptions to. Could any of those help me? Like literally just ask the question and see what you get. Sometimes it won't work, other times it will.

But really just not thinking that you can't do something with it Just just try and see what happens because the technology is moving so fast that what you couldn't do two years ago You can definitely do now And that's something you also have to be committed to if you're a regular AI user is that something that may not work six months ago may work in May work today even sometimes something that didn't work three months ago now it works today, and it's just getting faster and faster So you really have to be committed to just try things just

try it out and don't be afraid to fail and then try again.

John Bennett (27:20)

That's great advice. And actually that leads us really nicely into my next post that I'd kind of clipped to talk to you about, is projects and why they fail. So I think the post was something like building the AI tool is 50 % of the work. The other 50 % is rollout. So just that kind of idea of the rollout and why projects fail.

Katelin (27:29)

Yeah.

Yeah, I think like kind of as you say, it's connected in what we were just speaking about there. But I guess for instance, the transcription to project scope.

system that I had created there and use every single time I start a new project and my file structure and everything. I was extremely excited about that and I delivered it to my team. I gave them a training on how they could use the tools and they could connect them together. And now you have to accept that this was around a year and a half ago. So things have gone from being much more ⁓

stuck together with sticky tape, you could kind of say between tools. And there was a lot of transferring and copying and pasting and these sorts of things. But still it saved loads of time, even if you did do that. But the excitement was there. I felt like I had created this amazing tool, delivered it to my teammates. And yes, there was excitement from some of them, but from others, it wasn't as like, there was pushback, you could say, not pushback, but hesitancy.

And I think that either comes from one of two places. Either it's that they're not at the same place that you are on your AI journey. And this seems too complex or too, it's just not where they are in their learning journey of AI. Or there is also the fear of being replaced. And as a project manager, a lot of the administration that we do is a huge portion of our work. So if you are unable to really step up into that strategic kind of place as

as

a project manager, there is fear that, where will we be in a few years time? Do we actually have a role? So you really, you know, and I guess for some people that is more scary than others because it's either a strength or it's not.

I guess the reason why I mentioned that 50 % of the tool is building it and then 50 % is rollout is because you have to be able to take your team along with you. And that means obviously starting off with showcasing whatever the tool is, but you need to be thinking that there is going to be, there is possibly going to be resistance due to those two reasons that I spoke about. And you need to address those and you need to take the questions. You have to apply that feedback in your future training.

you really have to expect that even though you might think it's a fantastic tool.

It's just, may not be picked up as much as what you expect just because you're excited about it. And I think one thing that we've been doing in AI that works is that we've been sitting down with teams individually, the individuals within those teams. So we not only do, say for instance, your foundational training around how to use AI, but we also then sit down within the team one-to-one and look at what are they doing on a day-to-day basis in their role? Where are the opportunities?

within that role to turn the task that you absolutely hate or that takes you way too much time into an AI supported process. you know, starting off with the one that they really dislike is probably the best place to start because who doesn't want to be doing less of the work they don't like doing? So, you know, that that kind of process of really sitting down and understanding their pain points, their individual pain points really gets people to experience the benefits of AI in a way that

will kind of hopefully take down those hesitancies and show them that there is gonna be real wins to this. And yes, also then there is the, well, but what about my role in the future? But I am a big believer that roles are going to evolve and roles are going to change and the future of work is, yes, uncertain. We can't say that we know because we don't. And I strongly advise leaders to state those facts. Don't lie about it to your people because they can see through that.

honest with we're all in an uncertain situation. Even our leaders are in an uncertain situation. So being truthful and being honest with your teams and take it, you know, as much as top down strategy for AI implementation is important, bottom up is just as important, if not more important, because you have to take your people along with you. have to get them to see the results and the benefits from using AI in the business before you say we need to apply AI to all of our business processes.

and get ROI. Like what

does that even mean, you know? ⁓ I feel like that's where that real people human centric application of AI is so important and is the reason why I think a lot of projects don't work out. Now I have missed one point that I should have made and that is in regards to when you talk about top down, they are like strategically your leaders will say, okay, we need to be using AI across all of our processes. We need to be getting time back. We need to be more efficient, but.

John Bennett (32:14)

Hmm. Yeah.

Katelin (32:42)

And we've got this new tool and it's going to solve all of our problems. Like say for instance, we've got ChatGPT or we've got Claude.

but there is no addressing the pain point. There's no addressing the problem. So they're bringing in a tool, but they don't figure out where is the pain point, where is the problem, and then the solution. They bring in the tool first. So you need to do it the other way. That is where you're going to actually get the rewards. You start off with the pain point or the problem first, and then you come up with the solutions. And things are changing so fast that you don't want to get locked into a particular tool and then have to figure out that, actually that other tool would have been much

better for us to have taken up a year enterprise subscription or a year subscription with. So that's another reason why I feel like, you know, they really fall flat on their face sometimes.

John Bennett (33:30)

It's interesting because one of the things that you said there really kind of tallies with some of that I've found and that's, you know, when I've tried to do a project and you start with a big team meeting and this is what AI is going to do and it almost always falls flat on its face and what I've found works for me better is to...

Find somebody within the organisation that's interested and as you say has got a pain point and work with them, do a project with that one person and then they become an ambassador then and help you roll it out elsewhere. It's interesting isn't it because it's that thing isn't it? It's that kind of getting people excited and you say solving their problems rather than a top-down edict of right we're using AI now.

Katelin (33:55)

Definitely.

Exactly, yeah. The champion teams definitely are a great strategy to set up within any business. And yeah, you are right. It's getting people to that point where they get to experience the excitement, they get to experience the win. And then they're like, okay, maybe I should think about this a little bit more, you know? So yeah.

John Bennett (34:34)

Hmm,

absolutely. And it's really interesting. Another thing you said earlier as well that really sort of chimed with me is that, you know, we get so excited about the projects that we because, you know, we build something and you go, this is amazing. You can do this. and it's, it's, you know, we have to realise that not every else might be as excited. So we've got to get them to that place first.

Katelin (34:46)

Yep.

No.

Yeah, my husband always says AI is a very, it's a very niche taste, you know, like, you know, you've got

John Bennett (35:01)

Hahaha ⁓

Katelin (35:03)

Not everyone likes AI, that's for sure. But we do, you know, I think the social media and that sort of stuff puts us in these little bubbles where you're surrounded by other people that love AI. like in the real world, it's like a lot of people have a lot of doubts around it. A lot of people have a lot of hesitancy around it. A lot of people think it's just going to remove all creativity. You know, it's, it's, it's a negative thing. And I guess, yeah, it's, that's the reason why I love breaking it down for people is I love to show that, you know, there are lots of positives

that can be taken from this. And there are lots of wins that I guess you could say, and this is the reason why I get very much so excited to work in small businesses is because the little guy now has the chance, which is, you know, that was.

People can be one individual person can easily putting out 10 times what they would would have been in the past with the use of AI to, obviously a more expert level, but this is now possible. ⁓ you know, and, that, that level of that, that barrier to entry has never been so low and it's helping people see that. And I think, can you say that that, you know, we, we can compete for once, you know, it's, it's and yes, okay. The companies are going to get ahead, but there is still opportunity.

there that was never there before and why not take the opportunity while it's there and make the most out of it. I guess I'm always a glass half full sort of person and yeah, make it what you will.

John Bennett (36:31)

Yeah, I think that's a

great approach, a really great approach.

So I think the other thing I want to talk you about on LinkedIn, so moving on from your post, I mean, there's a lot of this stuff on LinkedIn where people say, comment this to get my framework or comment that to get my toolkit or, and it's funny because we both quite often comment on the same ones and pick them up. I just wondered how often when you do that, do you find that the toolkit or the framework lives up to

the post or and how often is it hype?

Katelin (37:08)

I think it's probably about 50 50. I have to say, it's very hard to walk past something when you're building your own AI implementation business and think, okay, has this, has this person come up with something that I hadn't thought of? Do I need to see this? I need to see this framework. I need to see what they're doing and you just got to have it. And I guess it's very smart from a, um, from an engagement perspective of trying to build their audience. You know, everything's about engagement these days, likes comments and viralness of your.

or

of your content. However, I tend to steer the opposite direction. As much as I'll throw a little bit in there here or there, I don't like the hype. I think it's a lot of noise that none of us need in an already very noisy world and controlling that noise and figuring out what is the signal that you should actually pay attention to, especially as someone...

like ourselves or like individuals that are thinking, okay, I want to learn AI, but I don't know where to start. And all this stuff is coming at you from all different angles. And a lot of the time coming at it from an angle that is of a much more sort of expert level than what you're going to be at.

as a beginner, need to utilising AI, need to learn from the foundational level. you start with building context, start with building projects, then look to bringing in maybe some skills, then look to do some automations. It is very much so a kind of step-by-step building those foundations, much like anything is when you learn it in life. But I find that these frameworks kind of skip all the way to the end and they sound amazing, but if you have not learned those foundational

elements, you're really not going to get the results that they're promising. And a lot of it is hype. And unfortunately, I think it's very distracting. It's very noisy. It's very hypey. And I think the way that I get around it is I choose what am I looking to learn about? And who do I see as the best person to pay attention to in whatever the channel is that you are active on and only pay attention really to those creators posts, because then at least you're going to be kind of on a journey with that creator.

And

if that journey is in line with what you're looking to learn about then you're getting to kind of filter out that noise And I find that very helpful for myself So thinking about where are the areas I'm looking to develop on and then aligning with one or two creators max within whatever that social channel is and Avoiding the rest of it to be honest When it comes to frameworks that they do share and I guess utilising those lead magnets. That's what they're called if you if you want to use them in

your business, the one thing that I will say is customisation is the main thing. So whatever that you do take, if it is say, for instance, like I gave out a 15 skills business pack, I think it was last week or something. And I gave as part of that guide, a section where it explains how for you to get the best out of these skills, you need to customise them to your own processes. So taking whatever it is, the framework that's shared with you, uploading that into the AI and asking it to help you work

through this to customise it to your own business process. And then you'll end up with an output that takes the framework that they shared with you, but it's customized to what you do in your business or what you do in your work life or your personal life. So I find that's very, very helpful. And a must do if you're gonna use these kinds of tools that are shared online.

John Bennett (40:33)

Yeah.

Yeah, absolutely. think that's really good advice because you know what works for one person isn't necessarily the right thing for somebody else, isn't it? And it really worries me when people are copying and pasting stuff and just using it as is without thinking about what it does for a start. mean, critically thinking, what is this thing that I'm using? Yeah.

Katelin (40:54)

Yes.

Definitely.

Yeah, that's for sure.

John Bennett (40:59)

Absolutely.

And when you do find one that's useful, you kind of answer this already in terms of personalisation, but there is so much out there. How do you integrate a new idea or a new framework into what you've already got?

Katelin (41:12)

Yeah,

I guess when it comes to the development of my business, I guess there are areas that I'm focusing on. kind of plan out my year as to how I'm looking to build out the different parts of the business, bringing in AI across all those processes. Would you believe even in an AI business, unfortunately, some things I still have to do manually and I haven't got around to, to bringing in AI across all of them. So I kind of have plans around that and

Generally now one thing is a lot of these things are going to change with time as it's very fast-paced But I have my notion and my notion is literally like my second brain for all my planning for my business and I save these lead magnets into the notion doc planned with the different Deliveries that I'm looking to complete as part of the development of my business. I try and not just put them in the save folder However, I have actually built a tool which is brilliant that I'm actually getting a lot of value from and it is from

that saved folder that everyone has, I'm able to connect with the saved folder. So I've got an Appify scraper that scrapes down all my information from my saved folder and that then filters into my content creation agent that is able to help me with planning out my content as per my content pillars. So instead of wasting all those amazing saved posts that we come across that were like, that's such great knowledge. I'm able to put it into use.

through my content creation process, filtering down those ideas, then through my brand voice, then through my content pillars, then creating original context or content which is built on the great ideas that I've been inspired by that are also out there in LinkedIn, TikTok, YouTube, wherever it is.

John Bennett (43:00)

That sounds like a really good process, actually making it work for you. I've got something a little bit similar. I have a thing called my knowledge base and what I do with those things, because I have the same thing. I can't walk past one of these posts. You look at it and you go, oh, I'm sure it's hype, but what if there's that one thing in there that I don't know about? So I generally comment on them all when I'm on LinkedIn and then I'll drag them into a folder.

Katelin (43:16)

I know. Yeah. Exactly.

John Bennett (43:25)

And then I have this knowledge base and I have ⁓ a plugin that kind of looks at anything coming in into the import folder and goes, okay, compared to what we already know, is this new or is it a tweak on something we know or is it this is just, know, we've already got this, you don't need to worry about it.

Katelin (43:39)

Exactly. That's a great idea. So you're filtering it

before you even, you ever read through, even read through anything. That's brilliant. That's great.

John Bennett (43:46)

Yeah, yeah, absolutely.

And then of course, if you flagged it as something genuinely new, then I go, okay, well, I'm going to read that, you know,

Katelin (43:52)

No.

John Bennett (43:52)

and understand it myself.

So what wouldn't you get AI to do? You know, what are the things that you go, no, I'm not letting AI touch that, that is 100 % me.

not be in the future, but you know, right now, what are these other things you say, no, AI, stay off this, this is all me.

Katelin (44:15)

Yeah,

definitely. So I wouldn't say anything is 100 % me at this point in time. okay, there's personal things. I don't think it's a very good idea to be using it for any kind of therapy or any kind of personal deliberations. I feel like that is a dodgy territory that we should definitely keep 100 % human. yeah, read a book or something, please, if you're looking to personally develop yourself.

However, I guess one thing that I just won't let it do is I don't fully automate anything. A lot of my business is content creation, whether it be courses, webinars, posting online. I do have processes that do market research for me, do competitor research for me, pulling content as inspiration. All that sort of work gets done with the use of AI and the different context bases that I've built in order to perform those tasks.

but what it won't do is it will not publish anything online. There are people that run fully automated businesses, but that is just not what I believe in. There should always be a human in the loop and reviewing the output before it goes live. And then another thing that I will not do is I guess, is I won't create content from just...

previous information that's pulled in from somewhere else. I will always create my own long form content that will go along with the content pieces that I've been inspired by. So for instance, what I tend to do on a weekly, bi-weekly basis is I research different pieces of content that are doing well or different things that people are talking about online. And I sit down, it's like me reading my Sunday paper. I go through all those articles, I read them, I take notes,

that kind of pop up for me, different points that really stood out to me. And then I have a tool that I've built that actually I put in those articles and then it asks me questions like an interviewer would about those articles and I record myself answering those questions. That then is what I create my content from. So it becomes my draft, my long form piece of content that then becomes my content rather than essentially just taking content and

running it through a bunch of automations. So I feel that coming up with the draft or using a long form piece of content that is you speaking is essential to getting great quality on par voice content from your AI system. So I always like to collaborate with AI rather than use it as a fully automated system.

John Bennett (46:59)

I love that. you're taking a concept and rather than saying, look, here's my tone of voice, now write something about that in my tone of voice, you're actually giving the AI ⁓ your view or your response to that piece, which is your tone of voice. I love that. think that's great advice.

Katelin (47:16)

Cripp.

Correct. And just to add as well, to be,

just to

make the point there that I also have a tone of voice document. I, know, as much as I record and give it my tone of voice in the long form piece of content that I produce, I also have a tone of voice document, which was essentially developed from a very lengthy interview process where I've had that with the AI through a prompt that I'm more than happy to provide to anyone. If they get in contact with me, which will interview you on developing a tone of voice document from start to finish, literally ask you questions and asks you to provide

examples, it will ask you questions for you to dictate and for you to, you know, give that information back to them so that you can fully understand how you speak naturally. It's not about being, it's not about being specific like you're writing a formal email. It's literally how I'm speaking right now, trying to capture that in essence. And then paired with the long form piece of content, you really get a great output. So yeah.

John Bennett (48:20)

Yeah, that sounds a really good plan. I think when I put the description into the show notes, I'll obviously put your contact details, and how people can get in touch with you, but that's a wonderful offer for people who can get in touch with you to get a copy of that. So we talked a lot about a lot of stuff and some of my questions I wanted to kind of talk to you about, we've probably covered, but what I really wanted to get into is kind of, you know,

Katelin (48:30)

Yeah, great.

Definitely.

Mm.

John Bennett (48:48)

Like we talked about earlier, we're in this kind of bubble. We're in the AI world. We use it a lot. But there are so many people that outside that. And if somebody is kind of nervous about how to get started, or maybe they're using it in that way, like a search engine, what advice would you give them to get kind of beyond that and get started?

Katelin (49:07)

Yeah, definitely. I guess it starts with, you're going to have to get a piece of paper out or open up a notepad online and just put aside half an hour to 60 minutes, sit down specifically with the task of thinking about what you do day-to-day within your business or at home or wherever you're looking to first start off with AI.

And I would say, think about, like I said earlier, think about the things that really annoy you. Think that, think about the things that take too much time. Think about things that are frustrating and are also repetitive. Things that you do on either a daily, weekly, bi-weekly, monthly, quarterly basis. For instance, like for me, I hate doing my taxes. So I have created a system where I literally take a photo on my phone. It uploads that photo. It takes all the details down and it adds it into a spreadsheet.

which essentially does my bookkeeping for me. So it's like sitting down, giving it that time to think about where are those opportunities to...

take a repetitive process that you don't like doing and use AI to either make it faster or less frustrating or more accurate, whatever it is. And I would then open up one of the LLMs, so the large language models. So either Gemini, either Grok, either ChatGPT or Claude and say to it, I would like you to help me write a standard operating procedure or an SOP. You might hear it said. If you've got an SOP

this is for a business process, brilliant, you're already a step ahead, give it that SOP and ask it then, so sorry, I don't wanna jump ahead, but you wanna write the standard operating procedure. So you're literally just gonna ask the AI, help me write a standard operating procedure if you don't already have one.

And you're going to do that by explaining to it from end to end. What is that process? What are you actually doing? And I would do that either by typing it in, or as I said, hit that little microphone and just dictate to it. And it is then going to take that information. It's going to ask you questions. If it's Claude that you're using, it will generally just come back with clarifying questions. But another thing to learn to do with other models is, or just get into a good habit of saying, especially if you're using the dictate, it's just an extra sentence to say, I

and if anything is unclear, just ask me questions and I can give you further information. It's gonna take that and it's gonna write it into a procedure for you. Once you have that procedure, that's your context. is your AI is able to understand that process from start to finish. And once it has that context, you can then ask it, how can I use AI in this process to make it more simple or more automated or more efficient? And it will come back

to with suggestions and you can consider those suggestions. ⁓ It's gonna be different on every use case, but I would say definitely starting off with one that annoys you most and one that's not too complex as well to begin with. ⁓ And if you are depending how aware you are in your learning journey with AI, ⁓ if you understand what connectors are, so connectors are essentially like a universal power plug between an AI chat interface and your external tools. So for instance, like Canva or Figma.

or Google Drive or Google Calendar. And you could tell up what connectors you actually have access to. Because if you open up connectors within say Claude or within ChatGPT you'll see different connectors that you can switch on, which essentially gives the AI access to. And some will be read and write. So some will be able to read information and pull it back, or some will be able to read and write to them. So for instance, yesterday I made a couple of appointments to my physical therapist.

And I just opened up Claude and I said, hi Claude, can you please add my two appointments, 9.30 on whatever date and such and such, and added it to my Google calendar. It just went off and did that via that connector. You know, it's like those behaviors are going to be things that you slowly build up over time. ⁓ But that's where I would say to start, just think about a pain point, document a procedure, and then ask the AI how it can help you and go from there. And as I say, test and iterate, experiment.

not going to be straight away. Maybe it will be straight away. The tools are getting really, really intuitive now. However, you still have to be committed to learning and experimenting. It's just part of the process. And I can tell you from someone that's been doing that for the last nearly four years, it is so worth the journey because, yeah, the rewards are crazy.

John Bennett (53:51)

Brilliant. And at the same time when somebody's doing that, when they're kind of getting started and trying to build that sort of procedure, what are the kind of the key gotchas, if you like, or the key things to watch out for, or they might get wrong and they could look out for?

Katelin (54:08)

I guess just make sure that you're reviewing whatever it has said back to you. What was the output? Make sure that you read it through. know, AI tends to give you very lengthy responses at times. And you know, there's this, this, want to get things done quickly. So there's this want to just copy and paste and go to the next step. Take a deep breath, realise that this is, this is not going to be something that if you want to do it properly, you need to put aside the time to actually, getting these things set up well.

and just make sure you read through things carefully. You tweak it if necessary before you move on to the next step. You know, it's.

It's important because the thing is every system is going to build on itself slowly but surely to hopefully give you a workflow from end to end so you go from your start to your finish. And if one of those pieces is wrong because you rushed through it, it's going to mean that it's all going to have an effect on the next step and so on. So just take your time. I think that's a key point to we all want to rush these things, but you need to give it time as well. You need to give it careful consideration and make sure that you

read through all of your outputs before you move on to the next step.

John Bennett (55:21)

That's great advice. One thing I do for that is I always, if I'm building a system, always ask this question. It goes something like, what's the thing we're going to wish we thought of now later on?

Katelin (55:33)

Yeah,

John Bennett (55:34)

because you know that

Katelin (55:34)

great one.

John Bennett (55:35)

and it quite often brings out something you oh, I haven't thought about that. You know, if I do it this way, if I do it that way, it's going to impact all these things. But I'm terrible with that. You you get a long answer and I look at it and I just want to go, oh, press the button and do it. You have to make yourself read it or quite often I'll just go to it. Give me the short version or know, explain that to me in plain English because sometimes it does like to really be verbose, doesn't it? And just say, you know.

Katelin (56:01)

⁓

John Bennett (56:03)

I trained as a barrister years ago and there used to be this kind of popular myth that barristers were paid by the word because that's why they gave such long speeches. Sometimes it feels a bit like that, it's almost like it's trying to justify being there because it gives this long long answer. and you think, there must be a shorter version of that that you could give me.

Katelin (56:05)

Okay.

Ha!

Yeah.

Definitely, I think you just made a point pop up there in my head. think.

Whatever this word is sycophantic. I'd never heard about it before, but that's new. The new word for AI going along with you. And no matter what you ask it to do, it's going to tell you you're brilliant and fantastic. it's such a great solution, a great product. You're just amazing. Sycophantic they call it. But in the end, what I like to do with AI is always like when I propose anything to it and it kind of goes along with it. Like it's saying that they do, they want to keep you on the platform. They want to keep you engaged. So they want to make you feel good.

If they told you you're an idiot, you probably just go, okay, I don't want to use this tool anymore. you know, asking it to propose the opposite argument or like you say, a gap analysis. What did I miss? What should have I considered could be actually a risky part of this process. Amazing. It's so good at doing that. And then I think when you've got both arguments, then you as a human can make your decision as to what you feel is best to do next, rather than leaving those decisions up to the AI. You don't want to do that.

John Bennett (57:31)

Yeah, that's great advice. Absolutely great advice.

So one thing we quite often do is we have a listener question, but we've chatted before, we decided to do slightly different today and we thought we'd tackle something that you get asked a lot instead. And if I'll get it right, it goes a little bit like this. Yeah, people are asking you, I get constantly lost in all the AI content in my feed and find it hard to keep up with the features launching on all the big name AI tools. So if you were to pick one area of AI to really focus in on, what area would

Katelin (57:52)

Yeah.

John Bennett (58:10)

be.

Katelin (58:11)

Yeah, definitely. I feel that the biggest feature launch or

move in the technology recently is definitely skills. So you might've heard of skills, skill MD files, markdown format. That's the format that they're written up in. They were created by Anthropics, Claude, well they were created by Anthropic and Claude would be the tool that you would have probably heard of. ⁓ And they are literally the biggest game changer or unlock to the non-technical type of person.

So before you could use tools like N8n or Make or any of these automation platforms to create different steps for your workflow and on triggers move from one step to the next and you create an AI automation. However, there's really no need to do that for a lot of the non-technical kinds of workflows or processes that you or I are probably going to use John in the things that we do. You can now create skills.

skills

are essentially an instruction manual for your processes, which the reason why they're so important is because skills, and this is a little bit, I guess it's kind of meta to think about it, but skills can call on other skills within them. say for instance, you've got an end to end workflow and at the start, so using a client proposal workflow, at the start, you're going to have your discovery. And then from your discovery,

it's going to go into writing up the proposal off that transcript. So you're going to have a skill that is going to take the discovery transcript and it's going to send it to the next skill, which is the taking the transcript and turning it into the proposal. And then you're going to have another skill that's written up in that instruction manual that is going to call on the next step, which is your QA, so your quality assurance. Is that proposal written up to the framework

that we like to follow. So it is essentially a non-technical way of being able to set up automations within your AI tools. And the exciting part is that skills have now been adopted as ⁓ multi-tool application that ⁓ you can take from Gemini, you can take it to Claude, you can take it to OpenAI. So you are able to document all of your skills. So literally instruction manuals that document

your processes in a format that the AI can act upon without hallucinating and literally replicate your business and apply it to any of the AI tools. Before, you as I say, you had to learn platforms like N8N, now you don't and skills are really, it may sound technical what I'm saying, but...

The way I'm creating majority of my custom skills right now is as simple as this. I sit down with AI. I've got a process that I'm looking to complete. say for instance, it's writing a piece of content for LinkedIn. I'm going to take my context. I'm going to, so for instance, the article that I would like to reference, I'm going to then dictate the different types of points that I would like to make. And then I'm going to go through the process of creating the final output.

At the point of which I have reached that final output, I'm going to say back to the AI, I would like you to now write a skill for me to be able to use next time I perform this task. And it's just going to go about creating that skill for you. So then next time you're not going to have to go through that back and forth. Say for instance, it came to you with an output that you didn't really like and you had to edit that. It's going to essentially learn from that skill, from, from that process.

and then dictate it in a way in that skill that you can avoid that back and forth in the future. And another step from that where skills are allowing so you can have skills that actually review your day's processes at the end of the day and they will learn from all the different changes that you made throughout the day and update all of your prior existing skill documentation with whatever developments you've made in your processes. So I know it's a bit of a long-winded answer

but skills are essentially the most amazing technology of recent times that I would apply to learning about. They're just huge, definitely. Yeah.

John Bennett (1:02:37)

Skills,

Brilliant, brilliant.

And that kind of leads quite nicely. I wanted to wrap up by maybe talking about your AI consultancy, what you can help people with and if they're interested, how to get in touch. So maybe tell us a little bit about that.

Katelin (1:02:54)

Thank

Brilliant. Yeah. So I guess AI that works, we run foundational and department specific training services. So we can take your teams or your departments within larger businesses or in smaller businesses, sort of, you know, a group of 20, we can, we can go through foundational learnings of how to adopt AI and their processes. Or say, for instance, it's a larger business and you have a marketing department, we can provide bespoke training for that particular department as to how to use AI as a marketer in business.

and so on, financial departments, operations departments, and that sort of thing. And then on the other side, we also have individual one-to-one consultancy, either working with sole business owners or also working with, as I described throughout our chat, teams. So sitting down with those teams and one-to-one throughout a program, working with those individuals to help them adopt AI in their processes. So if you're a business and you just don't know where to start with AI and you're wanting to be that business,

going to address your people and ensure that they are learning in the way that they need to be in this fast moving world that is AI. That is what AI that works does. We help those individuals get started with AI, start seeing results from using AI, and we do that through those particular activities. So if you are interested in anything like that, please do reach out to me. ⁓ I am active on TikTok, LinkedIn, and YouTube. I'm pretty sure John will put all those links in

description but yeah I am it's been a pleasure speaking with you John it's been really really great.

John Bennett (1:04:35)

Brilliant, great. And

there's been some great things there that you've discussed. really, I think for me, like my key takeaway, I love the way that you've kind of demystified some of those terms that we all bandy around that maybe aren't as big and scary as they might seem. So that's been wonderful. So thank you so much for your time. It's been great chatting to you.

Katelin (1:04:46)

Yeah.

Awesome. Thank you so much, John. I really appreciate it. Thanks. Bye.

John Bennett (1:05:02)

Thanks. Cheers. Bye

bye.