June 1, 2025

Tom Hewitson: How to Build AI Literacy in Organisations

Tom Hewitson: How to Build AI Literacy in Organisations
Tom Hewitson: How to Build AI Literacy in Organisations
AI Pathfinder for Private Equity Podcast
Tom Hewitson: How to Build AI Literacy in Organisations

Summary In this episode of the AI Pathfinder for Private Equity podcast, Steve Budd and Tom Hewitson discuss the journey of firms in adopting AI technologies. They explore the evolution of AI, the importance of building AI literacy, and the challenges organisations face in moving from experimentation to structured capability. Tom emphasises the need for leadership alignment, effective incentives, and metrics to measure success. The conversation also talks about the concept of 'vibe coding' an...

Summary

In this episode of the AI Pathfinder for Private Equity podcast, Steve Budd and Tom Hewitson discuss the journey of firms in adopting AI technologies. They explore the evolution of AI, the importance of building AI literacy, and the challenges organisations face in moving from experimentation to structured capability. Tom emphasises the need for leadership alignment, effective incentives, and metrics to measure success. The conversation also talks about the concept of 'vibe coding' and how AI can empower individuals to create their own solutions, ultimately transforming the way work is done in private equity and beyond.

Takeaways

Private equity firms are transitioning from AI curiosity to action.
AI literacy is crucial for organizations to thrive in the future.
Culture change is necessary for successful AI adoption.
Leadership alignment is key to driving AI transformation.
Incentives should focus on organic usage rather than bribery.
Measuring AI usage frequency is a critical success metric.
Organizations are in a discovery phase with AI technologies.
Vibe coding allows non-programmers to create AI solutions.
AI can significantly enhance productivity and efficiency.
The future of work will involve everyone writing code.

Links

Tom's Linked In and General Purpose's Website

AI Pathfinder website and more details on the Private Equity quarterly breakfast briefings here

AI Pathfinder Private Equity AI Strategy Meet-ups:

https://www.aipathfinder.co/private-equity-ai-events

Have a question or would like to suggest improvements, please contact steve@aipathfinder.co

SPEAKER_00

Well, hello and welcome to the AI Pathfinder for Private Equity podcast. I'm on a mission to help private equity firms go from AI curiosity to confident action by building a network that shares what's working, what's not, and how to unlock meaningful value with AI. A lot of firms and portfolio companies have already taken their first steps with AI. Maybe they've trialed tools, run a few pilots across different use cases. But the question now is how do we move from isolated experimentation to structured capability across the business? That's exactly what we're going to explore in today's episode. And joining me is Tom Hewitson, Chief AI Officer at General Purpose Training. Tom works with organizations of all shapes and sizes to build practical AI capability. He's also behind the VIB coding session at our forthcoming private equity AI strategy breakfast briefing. And uh we'll get into what skills matter in 2025 and how to scale AI literacy across your organization and why hands-on learning beats theoretical every time. So great to have you with us. Welcome to the podcast. Hi, Steve. Thanks for having me. And what a great intro. No problem, no problem at all. Well, look, first of all, let's learn a little bit more about you and general purpose. And perhaps, you know, why you find where you are today? What's your catalyst?

SPEAKER_01

So I used to work in the AI space. So I've worked in AI for about 10 years, uh, just over 10 years. Um, and I worked for a long time actually building uh these systems. Um, first uh kind of traditional natural language understanding systems, uh using intents, etc., um, and then uh large language model-driven systems. Uh, and that gave me kind of an early sense that something was happening. Uh, you know, when kind of GPT-2 became GPT-3, and then GPT-3 became GPT 3.5, you could really sense that hang on, something is going on here. Um, it was not a surprise when ChatGPT came out to me. For the rest of the world, they were like, wow, what is this thing? You know, I thought chatbots were rubbish, and now somebody's made one that's actually good. But for those of us that have been working in the space for years, we've been seeing that kind of incremental progress happening. And so when it happened, it was really, really exciting and kind of the validation of kind of seven odd years of belief that this kind of conversational technology would be really useful one day. One of the things that I noticed very early on was how useful these large language models were for helping me to do tasks. Even when we were still kind of working with APIs, I would be using it to help me summarize documents and whatever. And so when Chat GPT came out, I implicitly understood what a big impact this was going to have on the way that pretty much everyone works. I mean, if you do a job that involves writing emails, talking to other people, or any kind of basic knowledge work, it's gonna be impacted by these technologies. And I kind of saw, okay, here is something that is going to change pretty much everything. And people are gonna need a hand actually making the most of that, you know, advantage of it, rather than potentially being left behind by it. And so decided to found General Purpose as a kind of company to train people on how to do use AI. You know, that was very much at the start around kind of prompt engineering, you know, that was what I was doing most recently when I was working with large language models and, you know, how to kind of write things in a way that gave you more reliable responses. And in the kind of earlier versions of Chat GPT, like 3.5 and 4, uh, the earlier kind of large language models, that kind of prompting capability was really, really important. And there were kind of specific techniques and specific phrases that you could use in order to get better results. However, what we kind of started to notice as the technology developed is that good prompting is actually more than just copying and pasting a set of phrases from, you know, a hundred long series of use cases. It's about developing good taste. You can tell when you look at a prompt whether it's a good one or a bad one if you have practiced enough, if you have seen what gives you good results and what gives you bad results. And so we came to the understanding that actually the most important thing was really getting people to actively use AI every single day in every single task that they do, because that would be how they would develop that taste. So the company kind of ever so slightly shifted its focus towards really trying to drive culture change, to get people to adopt an AI-first mindset to make sure that whenever they started something new, their first thought was, I wonder how I could use Chat GPT or Claude or Gemini or even a custom model to do this better, more efficiently. And we found that that approach has been very successful. And it also gives something that's much more durable because these models changed the whole time. I mean, you know, just in the the last kind of 10 days, we had Microsoft Build, then we had Google I.O. Then we had Claude for uh Opus come out last night, which has currently beating all the benchmarks. You know, it has shifted like multiple times in the last two weeks what was the best model to use. Um, developing this kind of taste uh around what is the right way to work with AI allows people to adapt between the different technologies much more flexibly than if they just learn a set of phrases that they can kind of copy and paste that might have been really relevant in GPT 3.5 but don't make sense at all for a reasoning model like O3.

SPEAKER_00

Yeah, yeah, okay. Well, look, I I think that I put myself in the in the bracket of being curious, and that's driven, you know, I came to one of your workshops a little while ago, and and I can imagine that's driven a lot of your um sort of initial customers to curiosity, but we've you know we've got to move beyond those that are curious, and and certainly when we look across organizations, you know, we need to try and build that capability as a baseline. So what what what are the the challenges that you're seeing to to sort of achieve that? And and obviously you're speaking to to those that want to achieve that in their organization, but they've got to take some steps to to to deliver that, I suppose.

SPEAKER_01

Yeah, it's so interesting. So many of the organizations that we speak to uh at the kind of start of the process, they describe themselves as dabbling, and that is something that we see time and again is that organizations they've been trying some stuff. Maybe some people have got Chat GPT like a little small pilot, maybe they've tried some kind of custom tool. Perhaps if they're a bigger organization with some technical capability, they might have tried to build something internally, but really they're just kind of doing things here and there around the edges, and they tend to engage us at the point that they've decided, okay, we need to actually become an AI-first business. How do we make that happen? Particularly when there are a good chunk of people in our organization who are actively against AI. I mean, I'm sure you're kind of familiar with uh, you know, 20% of people are early adopters, then you've got your kind of uh early majority, late majority, and then you've got the laggards. And really, there is a kind of battle going on between the early adopters and the laggards for who wins over those kind of people in the middle. What we discovered really quickly, and firms tend to figure out for themselves, is that those early adopters tend to be terrible champions of AI for those people in the middle because they're like, look at all this amazing things that I've done. And the people in the middle go, Well, yeah, but I don't know how to do something really basic, and I'm now scared that I'm gonna lose my job because you're showing me how you've automated this entire workflow, and I can't even rewrite my email using this yet. And then you've got the kind of the laggard saying, This is really unethical, it's trained on trained on kind of stolen data and there's environmental impact and whatever. It's very, very easy. And you know, we've seen multiple organizations where they've mismanaged that kind of rollout and adoption process, and they've ended up with like 80% of the organization being pretty anti. And you know, once you get yourself into that situation, it's quite hard to drive something like an AI transformation, and you know, it takes a lot more effort, so it really is kind of worthwhile getting it right from the start. And you know, when we kind of go into an organization, we really do kind of the whole package around it. So we do training for leadership uh to start with. So we think about okay, how do we get the leaders to be aligned over what their AI strategy is? Because in pretty much every organization we've ever worked with, the leadership team are at different levels. Often they are less AI savvy than some of the people that work for them. And although we notice we we conduct surveys when we go into organizations, both with all the staff and the leadership team, and we frequently see that leadership teams are more optimistic about AI than the all-staff surveys, that doesn't necessarily translate into those leaders actually being any better using AI. I think they probably just feel more confident that uh their jobs aren't the ones that are at risk. Uh yeah, so we we we kind of try and get those people to align because one of the things that we see, I don't think we have ever seen a successful AI transformation where the organization hadn't nominated a dedicated person to actually own it and to make sure that it happened. And in so many of the organizations we go into where they're in the dabbling phase, nobody is actually owning, making this happen. You know, there might be an innovation manager who's kind of doing some stuff. Perhaps the CEO is really interested. It's very common actually to find that the CEO is the person that's very interested, but they've got a million things on their plate and they're never going to be able to give this the time and attention. And, you know, sometimes it's been delegated off to head of HR. I actually think that HR is a really great place to run this transformation out of. Um, and maybe we can go into that later if we need to. Um if it's just given to someone on top of the things that they already do, it's not going to be prioritized and it's not going to actually drive anything forwards. And we see that time and again. And so the first thing we do when we work with a leadership team is make them commit to picking that person and then adequately empowering them. And that usually means giving them some budget to spend because we've also come across plenty of times where you know the innovation manager's trying to make something happen, they haven't got the money to pay even for like basic Chat GPT or copilot licenses. And you know, it's that they're trying to kind of borrow money out of the IT budget, and so then the IT director is really opposed to it. And there's so much kind of internal politics to navigate in terms of getting that kind of transformation started. So that's something that we kind of help organizations do because we've seen it a lot and we kind of know what needs to be done in order to kind of get things going. And actually, honestly, sometimes I think it's beneficial to have an external come and tell you what to do when you feel like you probably could figure it out yourself, but just having someone that you could go, right, yeah, they've told us to do this actually makes it makes it a bit easier to go on and do it. Then we can kind of go through the organization, upskilling everyone. We always tell organizations to adopt a T-shaped AI strategy that is to go broad across the organization at the surface level, make sure that there's some kind of basic level of capability. There's kind of two reasons for that. If you are going to make AI usage an expectation of people's roles, which actually a lot of the kind of more forward-thinking companies are, you know, Duolingo's done it, Ethereum have done it, uh, Shopify have done it, they've had these big memos out. The one that got the least pushback was Duolingo because they said we're going to offer you support and training as part of this. And, you know, if you are going to expect your staff to actually actively use AI and you're going to make it part of their performance management, you're going to have to be able to point to how you trained them adequately to do it. Otherwise, you're saving yourself or you're setting up a lot of problems downstream when you come to that performance management process. Um, so so going broad is kind of very good for that. It also helps you to spot where the opportunities actually lie in your organization because people will organically try and use AI for the things that they think they want to use it for, or the things where they think it's beneficial. We went, we've been into far too many organizations where they've built some kind of internal tool based on something that some kind of innovation product team thinks is going to be really useful and then nobody is using it. And they say to us, Oh, can you train people on this tool to get people to use it? We have to say, what user need did you establish for people to kind of want this tool? And often, unfortunately, the answer is none, which is why when we're in those training sessions, people are like, I don't really get what I'd use this for. And buying in a kind of a standard general purpose assistant like ChatGPT or Copilot or Claude or whatever, and seeing how people use that first will give you really, really strong, good signals about where you should be investing in AI in your business. Because every business is going to need some kind of AI capability internally over the next 10 years. But where you're going to put that investment is really, really unclear. And the best way to discover it is to let it bubble up organically from the bottom of the organization rather than trying to guess as the leadership team what what to do.

SPEAKER_00

A couple of things have popped into my my head as you've been talking there. One is around incentives and what incentives you've seen work not work so well. So I was at a legal tech event a couple of weeks ago. And a legal firm has apparently offered a million dollars going into the bonus pot if they see that a million prompts have been entered into their uh into their chosen um uh AI at all. So that that was that was a very interesting approach. And legal is a very interesting area, you know, an area that hasn't changed for decades is suddenly sort of waking up to well, they could be uh fully disrupted here. What are your thoughts?

SPEAKER_01

What have you seen? Find that absolutely insane that they that they would do that. I mean, that just tells you absolutely everything you need to know about that organization. You know, I used to work at Facebook. Did we pay people to use our apps? No, people used our apps because they liked them and found them useful. We were with hundreds of organizations using AI. I've never seen a need to bribe staff to use AI tools because people, everybody has bits of their job that they don't like that they wish that they could get done faster. And if you show them how to use AI to do that, they'll run at it with arms wide open and grasp a hold of it as tight as they can. I mean, I just it is absolutely hilarious. I mean, they'd have done better giving us a million quid to come and train all their staff would have been a much better, much better plan, honestly. Um, I mean, I can say that, but you know, I I think that the thing that's kind of most interesting about that is the fact that they felt the need to do that tells you everything about whatever internal tool they're using. Uh that it clearly isn't meeting the need. Like if you if you if it was meeting an organic need, there would be no need to try and bribe people to use it. Um, and and I I really think at the moment, uh organizations should be thinking about how do we implement AI in a way that is going to be well received by the organization. It may well come in the future. I don't think, you know, I'm a realist, and I think most people are that if you have a set of tools that can replace a significant chunk of the work that people do, that is going to have some level of impact. But we don't need to tackle that as the first step. As the first step, let's get everybody doing things that they find useful for themselves and see where the opportunities lie, see where the efficiencies lie, and then we can have conversations about okay, do we need this level of resource or whatever? You know, and I think that that is a much better way to go than it is to try and bribe your staff to use AI. Um, and actually the thing that that's quite entertaining about that is that most of the organizations that we work with, they're always having internal fights about where they're getting the budget from just to pay for licenses for people to have access to these tools. So the idea that there are organizations out there that are willing to throw money at bribing people to use those tools is very, very interesting.

SPEAKER_00

Yeah, yeah, yeah, indeed. Well, look, the the other question was around um metrics and measures of success or failure with all of this has been quite tricky. But you talked about the survey that you you set out at the start. Is that one way you measure um you know the impact you're having?

SPEAKER_01

So so the kind of the key North Star metric the that we look at, I mean, we look at lot lots of things, but the key North Star that we look at is how many times it's being used a week, um, and then how many times it's being used a day. Uh, and before we go into an organization, the average usage is probably about three times a week. And that's not at all evenly distributed. If you thought uh thought of the kind of curve, it would be the exact opposite of a bell curve because you've got your early adopters who are probably using it pretty frequently, and then you've got uh a large number of people who are not using it at all, and you might have the kind of the odd person in the middle who's using it kind of once or twice, but invariably they are just playing with it, they're not actually doing anything of meaningful value. And our kind of key goal when we come out to those organizations is to see that people are using it multiple times a day, that the average number across the organization of usage is multiple times a day, like three, four times a day. And we think that if you can start demonstrating that everybody is using it three, four, five, three, four, five times a day, that is going to make sure that they are doing enough practice that they are going to develop that taste, that they are going to develop that AI first mindset, that they are going to continue to feel excited by the potential and explore what is possible. Because every single time they are using uh an AI assistant, that is them effectively exploring what can be done with that assistant. And you know, one of my favorite sayings about kind of large language models is that large language models are much more like a discovery than an invention. This thing came along and we don't actually know exactly how it works. It's not like the steam engine where it was invented and we it works in exactly the way that it was designed to work. It's more like discovering America and it's like, oh wow, okay, now we need to go off and see what's there. And oh, this works really well, this is really great, and this, you know, it hallucinates. It's like, you know, there's kind of I think that's something that's really, really interesting. And the only way that you are going to adequately explore all those use cases is by having um a lot of people doing it. And the people who are going to be most experienced in which of those use cases are going to be best done by AI are the people who are actually doing them themselves at the moment.

SPEAKER_00

Yeah, I mean, it's like any sort of well, uh coming from product background, you know, it's all about usage and the frequency of the use, isn't it? Really important metrics. Um, but obviously then it's about what ideally, we don't need to go into this, but it's ideally identifying those that are delivering most value, I suppose, um, uh as well. But let's let's let's move on.

SPEAKER_01

I think that I just to on that point that most organizations that we work with are at the kind of stage where getting people to use it just for anything is kind of the precursor, but you can't worry about prioritizing which use cases you're going to tackle until you know what use cases are out there, until you have a kind of comprehensive idea of the space. You know, AI as a technology might be racing ahead, but as organizations looking to adopt AI, we are very much in discovery phase. We are not in alpha, we are not in beta, like we are in discoveries phase. We are figuring out what is possible at the moment. All that happens if organizations try to force that bit is that they end up prioritizing the wrong things and then wondering why it didn't work, when actually they would have been better taking a step back and saying, okay, what are the things that the people in our organization are telling us that they find really useful?

SPEAKER_00

Yeah. It was Ethan Moloch, who writes a uh a news regular newsletter, has a book.

SPEAKER_01

He was the one that said that uh discovery versus invention phrase.

SPEAKER_00

Okay, yeah. Well, you know, the jagged frontier he talks about and and what's in and outside of the frontier, and and we'd only know that by you know using it and experimenting.

SPEAKER_01

One of the things that I find utterly fascinating about the kind of jagged frontier is because you've got lots of different types of models. So you've got the kind of the base foundation models like GPT-40 and 4.5, you've got reasoning models, you've got multimodal models. They all have different frontiers, and so you've got this kind of weird overlapping mishmash of frontiers that you know they're good at some things, bad at others, it's not easily predictable what what's what unless you try. And I think that that is definitely needs you to kind of adopt an experimentation approach as an organization um in order to kind of handle this properly.

SPEAKER_00

Yeah, and look, you know, for someone that that's um I I you know I feel I'm fairly competent, I'm definitely not an expert, but all those different models, you know, struggling to to to understand and connect how how I Use them. It just shows the importance to be keep um you know getting updates on right, okay, this week this is this is happening.

SPEAKER_01

Someone in a class uh the other day that to the uh in fact yesterday, and so they were like, Oh, so you're using 04 mini high. Why use 04 mini high rather than 04 mini or 03 or for that matter 4.0? And I was like, Well, OpenAI has done no one any favors uh with their naming schema uh for for their for their models. But it is quite hard to if you are not following this space really closely to understand that 04 mini high is halfway between uh 4.0 and 03 and is slightly better than 04 mini, but takes longer to give you an answer. And it and it's all about that kind of trade-off between the quality of the answer that you want and the speed at which you want it. Um, and you know, and I think that that that is not very clear to people.

SPEAKER_00

No, well that you you should yeah, you should definitely be writing the naming invention in the future, and that would be helpful for sure. No, what I'd like to do just over the next say five minutes is just try and bring it into the sort of P world. And I know that what's interesting is that you've got um a view across all sorts of different industries, different um roles as well, but you could but also you know you're closer to the community that I serve as well, which is private equity. Is there anything particularly different that you're seeing in this space or anything that you can give us that would perhaps point us in in a direction that maybe the the community should be uh thinking about more so?

SPEAKER_01

Yeah, interesting. So something that I've noticed in the kind of the P firms uh that we've worked with is that lots of the clients that we work with, they're quite like different internally. They've got like marketing teams and sales teams and that like and lots of different bits of the business work quite differently. I feel like the P firms are kind of they're more kind of consistent within the firm in the kind of the way that they think about the world and the way that they approach the world. They're also, I think every single person we've ever met at a PE firm is like ferociously intelligent and you know they're they're kind of very like focused on thinking about stuff, but they are also incredibly time poor. And this is something that we see kind of consistently with our PE clients is that finding the time to do anything is actually really difficult. And so, in a way, they are being hampered from innovating, from kind of actually adapting to AI, not because they don't have the kind of the intelligence or the capability within the organization, but because they don't have the time. And I think that that is something that as an industry they're gonna really have to like think about are we willing to take a step back on some stuff in order to invest in this AI transformation properly, even if that means saying to a deal team, you know, someone on a deal, okay, you've actually got to make time for this training session or whatever. I mean, you know, they are all smart enough that they could easily figure out AI for themselves. They could figure out how to use Chat GPT for themselves, reading documents or watching YouTube videos. You know, I think the reason that they work with us is because they're time poor, it's actually beneficial just to have somebody come in and say, okay, these are the key things that you need to need to need to think about. But I think that that is a real challenge for them. I think P are also in a kind of very unique position compared to our other clients, that they're not just thinking about themselves and how to make themselves more efficient. Lots of them are kind of experimenting with that as a way of trying to build up learnings that they can then uh put onto the kind of the portfolio companies. And I think that is the really kind of interesting opportunity for PE firms is to kind of sit back and say, all right, what would it mean if our portfolio was an AI first portfolio? Like, would that what would that do to the value of those companies? And I think that the first kind of PE firm to really kind of get behind that and make that happen, I think there's going to be a really big opportunity there.

SPEAKER_00

Yeah, I think it's what their role is that that's perhaps what they're trying to get their head around is is um you know, how do they, you know, do they actually create a a new fund almost to that that enables their their portfolio to drive things and accelerate things, especially with those that where there's clear uh opportunities, but that's still you know, definitely you know, I get from my conversations that they're not quite sure what role they play because it's not following the same playbook as other areas.

SPEAKER_01

I think the thing that they are probably also missing at the moment to an extent is that right now this is really couched in terms of this is an opportunity, get the portcodes to you know use AI and they'll kind of be more profitable and worth more and whatever. There is also a very big risk that if the portfolio companies don't do this and somebody else comes along in that portfolio company's market who is AI enabled and eats their lunch, that's going to be a real pain for the PE firm. And they could, because this is such a broad-based technology that's gonna apply to every kind of corner of business, that effect is gonna happen to every portfolio company at the same time. It's not like you'll be able to say, oh well, this one portco's gone bad, but the others are doing well. You might find that they all go bad all in one go, driven by AI. Um, and so the kind of level of risk associated with doing nothing is so high that I think PE firms have no opportunity but to get involved in making sure that their portfolio companies do something about this. And you know, there are lots of different levers to how they could do that, but I think that for the sake of their their own kind of position, they'll need to actually play an active role.

SPEAKER_00

Yeah, I totally agree. Well, uh, going back to finding the time, um, the briefings that I put on every quarter is is concentrated three hours for them to take out the office, breakfast is to come along and and and almost an intervention, you know, is to force themselves to come along to this, lock the door, right, it's is a download. And it you know, it's really important that that um it's grounded, it's accessible for them, it's not adding to the hype for the media that they read. And and to link it to the session that you're running is to for them again just to get a bit a bit closer to what's what's possible. And I wonder if you could sort of give us a little bit of a view of of what you're what you're doing, not so much as a as an advert, but more in terms of you know, vibe coding, which is what you're going to a session that you're going to be running with with uh with the group is you know a little bit more about what it is and and why should we be interested in that.

SPEAKER_01

Yeah, cool. So I'm I'm not gonna give away the uh the secret of what I'm actually actually going to going to do, but vibe coding, yes, I am going to be vibe coding uh an app live, and that's basically using AI to create or to write code for you. I'm gonna be creating an AI assistant, an agent that is going to help me to manage some of the kind of work that we do. We actually have something similar running internally um at General Purpose to kind of help us do tasks. And I'm going to show them how you can create something like this pretty quickly. And if you aren't necessarily a programmer, I mean, as we were kind of having a conversation ahead of uh ahead of this recording, Stephen, you were saying, how do you kind of take um vibe coding from kind of experiments into production? And I said, Well, really, you need to be a software engineer. Um, and you know, I I have that advantage of I do know how to code and you know, I do know how to kind of make it take that kind of fight, like step into deployment or how to debug if the coding agent gets stuck, you know, I can actually go in and manually figure out what the issue is myself. But that doesn't mean that they can't get a lot of value from vibe coding as it stands today. And the reason that I wanted to showcase it is that I think almost the vast majority of kind of the economic benefits of AI is going to come from the fact that the cost of software is trending towards zero. You know, in the in the kind of the past, if you wanted custom software, you either had to hire a kind of a development team and that was a kind of large budget, or you had to buy in something from a vendor. You know, SaaS has been a big feature of the space for for kind of last you know, 15 years or so. And now SaaS startups are really, really struggling because organizations are saying, oh, I can actually build this tool internally that meets my needs better than your SaaS startup can, and I can have it for free rather rather than rather than having to pay you 20 bucks a month or whatever it is. And I think that that is going to be incredibly interesting because there are millions of use cases in organizations, things that could be automated but only benefit like four people. And so you're just never going to justify spending a developer's time on it. They're too rare, they're too expensive, you need them focused on the thing that's going to affect like 100,000 users. But when you can actually create working software through vibe coding, people can create the workflows for themselves that make them more efficient. And I think that that is going to be one of the primary drivers of efficiency through AI is people vibe coding uh workflows that do tasks for them that they don't like doing, or you know, that they're like, oh, it'd be so great if instead of me having to kind of go to this place, get this bit of information, this place, get this bit of information, write them up in the middle and send it to this third person, that you could do that all automatically by designing a work, designing and vibe coding a workflow. That is not very far away. That is going to be a proper feature. I mean, you can already do that with the tools that exist now if you know what you're doing. And what we have kind of seen as the models have got more and more powerful is that the kind of the skill level needed in order to make that happen has reduced. Uh, and there is no reason to expect that by the time kind of GPT-5 comes out, you know, I mean, we were talking about Claude 4, Opus coming out last night, the, you know, early benchmarks seems to be much, much better at software engineering. It's kind of covering like 80% of the use cases on the software engineering benchmarks now. It might be that we are already at that moment, and that's going to fundamentally change how organizations work. I think that's really, really exciting.

SPEAKER_00

And do you think the the vision is that anyone inside a P firm could could be doing this?

SPEAKER_01

Yeah, I I mean I think that is the kind of the promise of vibe coding, is that basically everyone will write code and that it will become much more important to be a subject matter expert to know about the thing that you are trying to write code around and how to describe the problem clearly and you know all the other kind of things that go into that, but the actual process of writing the code will be fully automated through AI. I think that that I would be very certain that that's going to happen. There will still be roles for software engineers, but those roles will be largely orchestrating AI agents that are writing code, setting kind of rules across the organization, so making sure that the kind of the various agents and things that people are using adhere to kind of security standards that are saving into the right databases, etc. But I think that we can look forward to a future in five years' time where we are all writing code all day long as part of our jobs, regardless of whether we're a software engineer or not.

SPEAKER_00

Yeah, well that's that's uh you know fascinating uh thought, isn't it? And um, well, for those that are gonna be in the room in a week or so are gonna be uh well hopefully inspired to uh to start uh experiment experimenting themselves. I could spend a lot longer 30 minutes talking to you, Tom. But let's leave it there. Let's leave them wanting more. Maybe we get your back you back on at another time as things are moving so fast. Really energizing, enjoyable conversation. So so thank you for that and uh look forward to you uh hosting a session at the forthcoming breakfast. Right, I'm looking forward to it. Thank you, Tom. Um and uh we've got your LinkedIn details in the show notes and links to your your website so to check t Tom out there. But uh for now, thank you for for listening.