A Look Inside How OpenAI Deploys Enterprise AI With Stuart McMeechan, EMEA Deployment Engineering Leader at OpenAI

In this episode of A Class Act, Michael Young sits down with Stuart McMeechan, EMEA Deployment Engineering Leader at OpenAI. 

Stuart leads the team that OpenAI sends into the biggest enterprises in Europe to help them get AI into production and he’s seen first-hand why most never make it past the pilot.

In this conversation, they get into👇

☑️ What OpenAI’s AI deployment engineering team actually does and the two objectives behind every project
☑️ The three types of customer OpenAI works with
☑️ The difference between organisations actually scaling enterprise AI versus those staying stuck in experimentation
☑️ What’s really blocking the move from pilot to production (most of it isn’t technical)
☑️ What separates organisations genuinely ready for AI from those just experimenting
☑️ The culture changes required to make AI implementation actually work and how to find the “wizards” already driving AI inside your teams
☑️ Where humans stay irreplaceable, and how roles are already being redrawn
☑️ Is AI moving faster than organisations can adapt?
☑️ Why OpenAI’s release cadence went from 15 months to six weeks and what that means for planning

Guest Information👇

Stuart McMeechan is EMEA Deployment Engineering Leader at OpenAI, where he leads a team of AI architects and engineers helping Europe’s largest enterprises adopt AI at scale. His team works directly with organisations to define AI strategy, design target architectures, build implementation approaches, and develop evaluation frameworks - turning OpenAI’s capabilities into real business impact.

Connect with Stuart McMeechan on LinkedIn:
https://www.linkedin.com/in/stuartmcmeechan/

Please note: The views expressed by the guest in this episode are their own and do not reflect those of their current or former employers.

About Your Host👇

Michael Young is the Founder & CEO of MBN Solutions and host of A Class Act: Conversations in Data & Leadership. With nearly two decades of experience helping organisations build high-performing data, analytics, and AI teams, he brings frontline insight into what makes exceptional talent and leadership.

Connect with MBN Solutions👇

If you’re hiring for data or AI leadership roles, scaling AI capability, or rethinking how talent supports your AI strategy, the team at MBN partners with organisations to build data and AI teams that actually deliver. Visit MBN Solutions website at www.mbnsolutions.com

MBN Solutions on LinkedIn – Join the conversation and discover how MBN Solutions is shaping the future of data leadership. Visit the company page here: linkedin.com/company/mbn-recruitment-solutions

For questions, guest recommendations, or to connect:
Email: michael@mbnsolutions.com

Subscribe & Follow👇

Enjoying the series? Don’t miss an episode. Follow A Class Act wherever you get your podcasts and leave us a review to help more listeners discover these conversations in data and leadership.

Transcript

0:01

I often hear from leaders that many AI initiatives never make it past the pilot phase and are left wondering: where did we go wrong? Today, I'm speaking with someone who's helping organizations across EMEA move beyond pilots and into real-world AI deployment. Stuart McMeehan leads AI deployment engineering for OpenAI in the region. If you're leading a data or AI function or thinking seriously about how to scale capability, build the right team, and navigate Of modern AI adoption, this conversation is packed with practical lessons for anyone leading AI, data, technology or transformation. You'll leave with ideas you can apply immediately. Before we dive in, don't forget to hit subscribe, whether you're listening or watching on YouTube you stay up-to-date with insights from the people shaping the future of data analytics and AI. Stuart, welcome to A Class Act.. , let's start with your own journey. You've worked across consultancy, public sector, supply chain. What first pulled you into the world of AI? I've always worked in the kind of tech space, I started my career in software development in financial services and that was around the time of the financial crisis around 2008-ish. So I did a couple of years in that kinda field, and then pretty much since then I spent most of my career in consulting and it was all very much kind of data-heavy projects we were helping our customers with in My consulting career. Initially analytics, data science-heavy type projects Where our customers had big kinda business questions, and we would have to go in and capture a load of data, run a bunch of analysis, present some findings. and then the last kind of eight, nine years of my career have been a bit more in, in terms of like implementation, so where a customer needs to, you Access and analyze data like at scale, in areas like, Financial crime or supply chain. and we would Go in and plug in a system that they could use as part of their day-to-day job and helping them to do their area of the business, a lot better. So kinda throughout all of that, the more traditional types of AI, like machine learning, et cetera was very much baked into, to all of that. But I would say the more kind of GenAI side of AI, obviously as it's ramped up over the last two, three years I think I got my first kind of flavor of ChatGPT I was in my last role, and I think we were one of the first kind of big customers for OpenAI in terms of ChatGPT Enterprise. So I was using it as part of my day-to-day job fairly early, that was the first like glimpse of GenAI in action. And, an example from our side was, we would do a lot of kind of reverse engineering of customer systems and understanding data and all of that, and that would typically involve doing lots of Digging around, experimentation, et cetera, reading documentation in some cases. And, ChatGPT Enterprise, even in the very early days, was able to do a lot of that already and tell you the answer. And I think for me that was the first glimpse that, knowledge work was going to change and this was an area that made sense to go all in on. Excellent. Then an opportunity obviously came up for you to join OpenAI. Is that been just over a year now? Yeah, approaching like a year and a half, I think. It's gone- Year and a half. Has it already? think. A year and a half. It's gone- It's already that time A year and a half already. It's gone quickly, yeah. I can remember seeing an update. It was like a year, but that felt as if it was like two minutes ago. Just time just flies. I know. It does. Before we really get started with the podcast as well Can you, tell us what the AI deployment engineering team does at OpenAI? Yeah. Yeah, absolutely. So We're a customer-facing team. We have a team of like leading AI architects and engineers, and our-- we have two main objectives. One, we help our customers to build solutions on top of our models. So this could be, A bank trying to, improve their customer support function so they can get back to customer queries in a better way, or they can give more, insight to their customers about their spend. so building solution, building that type of solution with our LLM models that we make available to the market through our API. or it could be, A company trying to automate part of their back office function, like document processor or something. So we're essentially there to help our customers get from their initial idea and their use case through to a production deployment, basically. So, so something actually running in production. And, What that actually involves is everything from high-level kind of architecture advisory, so advising them, what's the target design? How is this gonna work? If it's like an agentic solution, How do you break down the different convergence, and how is that going to be controlled, and what will tool calling look like? Things like that. So there's a bit around design. There's a bit around like we do a lot of work around evaluations- Yeah … so helping our customers understand how are you going to measure this? How are you going to know when it's ready for production? How are you going to measure it when it's in production? So a lot around evals. And then we essentially D- advise the customer's engineering team as they build a solution, so helping to unblock them. And as we get new models come out, we'll advise on whether they should change direction or not. And then we help them with things like optimization when it's in production. So how do you improve latency and all these things that become super important when a solution's in production. So that's That's our core job. Yeah. We also have a side objective where We basically gather intelligence from the market to see like what's working well where our models are still struggling in areas, so very like specific areas and specific industries what's difficult in, i- in regards to deploying our models into production, and then feeding that intel back to our product and research team so they can Improve models through the post-training process. But also just from a general like platform perspective, so like security compliance, all the things that are important to our customers. Feeding that intel back to our product who can improve the product and make things that easier to deploy in the future. Ah, brilliant. we get straight into this then, the reality is many organizations talk about AI, but very few scale it. From where you sit at OpenAI, what's separating the organizations that are scaling AI from those that are still stuck in experimentation? Yeah. So From our perspective, and we we do this as our job, right? So every customer we work with is deploying AI to production. So from our perspective, we're seeing organizations scale AI, like a lot across the globe. So we globally track how many solutions that, we've helped a customer get into production, and the numbers are, a big number, and it's going up very quickly. Yeah. within OpenAI we group our customers into three buckets. So one i-in three kind of segments. So we have startups. And as you'd expect, startups are moving the quickest, and the biggest startups right now are all powered by AI. So if you think about the kind of, Lovables of this world who are doing-- Of helping customers to build websites and software, the kind of Harveys, Lagouris in the legal tech side there's lots of them, right? And they're all built on AI, so they all have AI running in production, like at scale, and they've moved Very quickly. the other segment we focus on is what we call digital natives, which is like the-- it's one up from startups in terms of maturity, so they've been around a bit longer. But, they were maybe-- they've maybe only been around for Ten years, so they don't have the legacy issues and systems, and their head count isn't in the hundreds of thousands. So they're moving pretty quickly as well. Most of them will have pretty mature AI capabilities. They've got some good things in production. and often the work we do with digital natives is on The more complex side. So like how do we you're pushing the boundaries basically. how do we make this solution available to more people and get the latency down by Ten milliseconds or whatever? It's fairly mature, but they're, they've got AI in production at scale. And then large enterprises, which is the third segment we focus on, probably where I spend most of my time just now. these are probably the furthest behind, right? Because they've got the big organization, lots of people. They've got different systems and processes in different countries. They've got, it's harder to build things because there's more processes to get through, and in some cases, they're working in regulated environments as well. So there's a lot, there's a lot more challenges. and so the trend we've been Last year was very much around like piloting and, Experimenting, seeing what's working, maybe testing the boundaries of what they can build. And for The large enterprises that really go, going with that last year and made good progress, this year is really around like scaling up to the other kind of use cases they've got across the enterprise. Our kind of top clients, and we've got Some amazing customers like BBVA, the bank based in, in Madrid they started very early. They've been rolling out ChatGPT Enterprise to, like hundred, hundred and twenty thousand- of their employees, And we've been, like my team and the FDE team have been working with them on Four or five, like really big use cases where they're building something custom that's going to really change them into a, like an AI-native bank. And customers like that really got a head start last year, so they kinda went in, they took some risks, and worked on our earlier models. Yeah. And this year they're in a really good position because they're then thinking about like they've done the first five, six big use cases. They've got the learnings, and now they're thinking about how they scale that up to the next Two, three hundred cases across the bank. So I think the general theme for large enterprises that it's less about Like deciding if an organization is going to do AI, and the conversations are generally about like, how do we accelerate? how do we scale up? how do we pick the best use cases to focus on next? It tends to be more of those types of conversations. No, it's good to hear, and it's good to hear it's really moving along as well. I've still been in some, like, leadership meetings and, like roundtables and stuff where it's-- the people are still talking about like someone's not logged into a license for five weeks, and how do we get them to do that? So I love conversations- Yeah … like this, where you really hear some just a lot's moving in the market. And from your perspective, what are the biggest myths companies still believe when they start exploring AI deployment? I think one would be that it's, that it can be deployed like a typical technology project. Yeah. So w- we've seen a bunch-- we've seen some kinda companies in the market who have kinda almost tried to deploy an AI solution like they would with a traditional piece of software. A big design phase where not, Nothing's been built, but there's lots of like paper and architecture diagrams and things like that. Then there's like big sign-offs- around security and architecture. then there's like a massive build phase, right? And with AI, with gen AI specifically, it, it doesn't quite work right because w- for a couple of reasons. One, a lot of this is new, right? A lot of what we're doing is new with gen AI, so There's a lot of experimentation that's required, and we encourage our customers to experiment early. So don't spend a long time like design, on design. Yeah. Get some dummy data, Run some tests, understand like how good is this, is, What we're trying to set out. If we just deploy like some of our, Our latest models like 5.5, which is really good at a lot of, like a broad range of things. Just like out of the boxes, just try what you're trying to do with 5.5 on some data and see how it works. And that gives you an indication pretty quickly of whether you're onto something and it's gonna be feasible. it's a lot more like iterative. You need to, try things out, get a feel for if it's gonna work then start thinking about your e- evals and getting those in place so you can start measuring and then scaling up from there to some sort of solution. I also think- You know, obviously think like tools like Codex, I don't know if you've managed to play around with the Codex app. yeah, you should try it- Yeah … if not. But, capabilities like this are As you'll know, have been really speeding up the process of basically writing code and building software. So we're seeing this like crazy flywheel just now where, like our team will be working with a customer around something like, A back office automation project or improving customer support or improving claims processing, for an insurance company. And, the AI models will be there to like, To enable that. Yeah. But then you've got tools like Codex, which means that you can build a solution like super quickly. Yeah. Um, so a lot of the customers we work with, within two weeks or something, we've got something running that they can see and feel and, we encourage our customers to do That part themselves, but it can be done using Codex, and it can get them to a very rough kind of version one within weeks rather than- Yeah rather than months. Brilliant.'Cause there is still a lot maybe people still getting stuck at the proof of concept stage as well. What do you see as the biggest blockers moving from pilot to production? Yeah. Two-- a couple of examples I would say. One, one is picking The right use case to work on. And I'd say Where we are in the market, especially in the larger enterprise type side of the market A lot of enterprises are doing this for the first time, right? They're deploying gen AI into production for the first time. And One you have to you should pick a big use case- Yeah or two or three big use cases. So is something going to move the dial on revenue? Is this something that's gonna mean we can service our customers better and we'll, we'll increase our revenue? Is this something that's going to reduce costs or reduce the process? So again, we can maybe like a banking example would be getting decisions around products and, the risk process and things. Can you do that in a shorter period so you can get better outcomes for your customers? So as like a big use case Is important for a couple of reasons. One You kinda need to have pretty senior buy-in and leadership for these projects in th-in this kind of early stage Of gen AI. So if it's a strategic kind of priority for the customer, then the C-- you'd expect the C-suite To be involved, and that's important because, for again, for large enterprises particularly they're gonna be doing-- they're gonna have to Change some processes. They're gonna have to unblock things. They're going to have to like shuffle around people. So we want the kind of business people involved in the kind of build process as well as technical. There's lots of things that's kinda happening For the first time. And I'd say for a lot of our customers, we've, We've had to see We've encouraged them to, kind of unblock things in processes when they happen. So again, rather than following the kind of traditional build process, which often has lots of gates and lots of approvals and things like that having to like reshuffle that to Suit and cater for the more iterative type development. pick the right use cases- Yeah … is super important. Have the senior buy-in, super important. And then I'd say probably thirdly, third example would be coming back to the eval topic. You have to have a clear view on how you're going to measure this. Yeah. So when it comes to this big decision about whether we put this solution into production there's a thing on screen which says we're good to go because the… If it's something straightforward like a customer, Chat bot on your website or something, it might be like a score around does the responses fit to our brand and quantifying that. What percentage of questions get answered well by the bot? The things like this we look at on a dashboard together with our customers before they go to production. But if you're sitting ready to go to production and you don't have any of that, it's gonna be a very difficult decision. No, brilliant. Thanks very much. And if you are, as you will do, advising board or C-suite today about launching a serious AI initiative what are the, The main questions they need to be asking before they start? I think I've covered some of it already. Yeah. So Is this the right use case? I th- i think kinda why, like why are we doing this- Yeah… is an important question. So again, it comes back to the use cases and, a lot of our, quickest moving customers, Y- you can go and speak to a member of the C-suite, obviously the head of AI and but others as well, and in some cases the chairman or- Yeah chairperson, and they will tell you what their top five use cases are for AI and why it's important to them. Yeah. And in some industries that will include them saying that,"If we don't get these right, We might not be competitive in the market in a year's time." So our, our best customers have a real grasp on on that picture. A- and it's different per sector, right? How fast each are moving and how disruptive AI can be. But having a clear picture of like, why they're doing it is super important. Yeah. And if they don't have that, then it's a fairly kind of clear indicator that it's … they maybe haven't seen where this is going, and therefore all the problems you have to, Unblock in your first run at getting gen- into- gen AI into production is gonna be very hard- Yeah and it will probably take a while. I'm guessing a big part of your job is maybe sitting down with these kinda C-suite and leadership teams and actually talking to them about the case studies and advising if it's good or bad, or, maybe even at times going in and presenting case studies where they could get value from that as well. Quite an exciting part of the job. Yeah, and I think As we … every month we see new- Yeah … use cases and obviously we work as part of a … Although I look after our Europe team- Yeah … we're part of, obviously, a global team, and there's just so many use cases coming out that in different industries that you see, Results and therefore we are flipping to be more, Turning up to a customer like an insurance company or a health company and saying,"Here's three things we kinda, kinda need to be doing now because we've seen it work in others. And then here's the big picture." And, and kind and Beyond the next two, three years, it becomes a bit of a harder conversation because it's hard … It's a bit more like, where do we think things are going And- Yeah …it becomes a bit more different per depending on the personalities and the companies in the sectors. but certainly the next, One, two, ish, three-ish years there's becoming a fairly clear picture of the use cases that different sectors should really be, Yeah should be moving on. No, brilliant. Lot of different cross-sector initiatives and cases I can imagine as well. The one thing we spoke about this quite a lot at The Edinburgh event when you were on the panel, the one thing I do find fascinating right now is that AI success seems to be less about the technology itself and more about the people, leadership, and organizational readiness. From your perspective, what separates organizations that are genuinely ready for AI adoption from those simply experimenting because they feel pressure to enterprise readiness is- Yeah… is a big blocker, right- Yeah to you, to rolling out AI and, I think, rather than, like I think our models are have got to the point where you can do like most things with the models pretty well. Yeah. And we'll maybe touch on this later, but our kind of cadence for releasing new models has reduced from 15 months to six weeks. Wow. So pretty much every six weeks there's new models coming out that are even better. Wonderful. So For bigger enterprises It's not really about waiting for like new technology. Like you say, it's more about like, how do we actually make this happen? There's a few things. I mean, The implementations we work on, there's Two, two different like sides of an AI implementation from what we see, And one is the kind of bottoms up, get AI in everyone's hands. and for us this would be like rolling out ChatGPT Enterprise and Codex, and this is just like getting it available to everyone so They come in on, in the morning to their job, and they've got a tool that's there to accelerate. Draft emails, do research- Yeah … build slide decks, things like that. And we'll maybe touch on this later, but like Codex is another part of that which is- Yeah … becoming even more, it's even more powerful and can do even more. But it's very much around like helping people day to day. And- For that It's all around like enablement. So first of all, like actually getting it in people's hands and getting it connected to the right data sources so that they can interact with their documents and emails and things like that. and we've seen that it, it becomes a bit of a, Leadership thing. Like each different function in the business, so like a finance function, sales, et cetera, there needs to be good, leadership and we often call them like wizards in each of those functions to help figure out like What are they finding most impactful, In this, kind of like day-to-day, AI use cases. Someone in each of those teams will have figured something out where it's like, "Oh, I used to spend like a day reading these contracts and like summarizing them in a spreadsheet, like what the risks are," actually like ChatGPT Enterprise can basically just do that, Out of the box. So like discovering those things, getting the word out, encouraging that person, maybe rewarding that person and figuring out how we get others in that function to- Yeah … to think like that and also to use some of these like patterns that have been tried and tested. So there's a lot around like finding the right people in each of those functions. and there are always people In every function who will use a tool like ChatGPT Enterprise or Codex and very quickly like figure out how it can help In their job. So on that side, it's all about enablement. On the like big use case side where we typically focus, which is okay, we need to completely change how we do work in this area or s- or service our customers, and we have to build something new to do this, that's some very different challenges. Yeah. That's things like, Pulling the right team together. So what we've seen A team building a new solution like this in a business needs to have a good mix of like business people- technical people who are willing to, Have had their hands dirty with AI, Are willing to try new things, are willing to, like, change direction when inevitably like new things come out in three months and you have to change your design and change course. So there's a bit around, like flexibility and kind of willingness to challenge processes as well is, like super important In the enterprise. So If, this team who's building this new AI solution gets told that actually we need to wait for three months to get something signed off the kind of confidence, I guess, To raise it to super senior level and push back and say "Okay that's great. Like, why do we have to do that? And is there another way that we can work that out? what's holding it up?" So a lot of our customers have the pressure to get these done, so we need to figure out work around these kind of traditional, maybe out-of-date processes to make it happen. but I would highlight going back to my point around like including the business is, like super important. One of the banks we're working with on like commercial and investment banking solutions to make, like, bankers more productive, so like automating creation of pitch decks, prospecting, things like that. We've got junior bankers basically on, on the team who are giving real-time feedback every week on, like, this."No, like that's not worth pursuing. Let's go in- this other direction, et cetera. So i- it's all around speed, and therefore you have to have the business involved. Yeah, definitely. I love all the kind of large scale kind of enterprise stuff, but I do love just having conversations with people on a daily basis where it's really helping maybe some of the younger managers coming through who may be struggling a little bit with, like, communication and talking to the execs within a business, and some of them are actually losing maybe bonus because they're not able to really articulate theirself right And tell the business what they're doing. Then I know they're using like the things like Chat GPT just to produce like weekly reports on their team, what they're doing, wh- where we're getting progress, where we're not, and just those examples are amazing as well how it's really helping individuals enhance their jobs. Yeah. I've got another one of my events in Glasgow next week as well It's just amazing, like- Yeah … like just trying to speaking to the audience, getting marketing out, and doing things where I can just run it through Chat GPT, right? I need to let the, let everybody know the address is covered next week. Just basic things like that. Yeah. Put it in with the website for the event, and it's just producing stuff in what's, a second whereas that took me a while just to go type out and think about and Yeah, and it's crazy how quickly that, that kind of like general knowledge work-… has changed so much- Yeah …this year already. I think like in in OpenAI we often refer to This term like feel the AGI. Yeah. I don't know if I mentioned that to you before- Yeah … but it's like you, you see a model or a tool do something that you didn't think was possible and we refer to those moments as feel the AGI. And I think in OpenAI, there was a lot of that around January, February this year when our Codex tooling got Very good, and basically- Yeah Codex is, uses our underlying models like you have access to in Chat GPT. But Codex has like a, what you call like a harness around, a really good, strong harness around it, which is like the model in the center but the kind of tooling which lets it do tasks like a user- Yeah computer, access the web, and do all kind of end-to-end type things. And, Around January and February, We realized it was just like unbelievable. Basically Building spreadsheets, building slide decks. Yeah. Like our team have gone through a transformation ourselves in the last few months just because of Codex, and so things like building architecture, technical design document diagrams, it does extremely well. And then like you say, just Like day-to-day things- Yeah … like when I report up to my, to the global leader, like we have a fancy dashboard that was just like created using Codex and refreshes automatically every few days. And yeah, just like our ways of working have really changed this year. Definitely. I I did a handover with one of of my senior consultants last week, but she's a, absolute genius at delivery. there's a large-scale project with a consultancy, 15 people, so it's volume. So what she's produced over three weeks, it's a lot of candidates, a lot of paperwork, a lot of organization, and it's She's doing it manually at the moment, basically. So it's I think we, as a, probably a business leader, I've spent a little bit of time with her because I had to and I've really understood what she's doing. And I know for a fact 90 to 95% of this could all be automated, and it would be amazing for her job. And even it would really change the type of people I would hire into my business moving forward. I'd probably be able to bring in more graduates and trainees and really embed them in the business. No, fascinating. Yeah. And- Yeah … and it's interesting what, freeing up time is, like- Yeah … a massive priority- Yeah … for a lot of our customers. And, some of our customers are, like, large enterprises are doing things like, I don't know if you've heard about kind of code modernization projects, which a lot, of the big banks and others are often doing, and this is where you're taking an old legacy system that might- have been written in COBOL or something- and converting it into a more modern system to, like- Yeah …save time and ma- cost of maintenance and things. And these programs can take a year That's a different thing … to two years, maybe even longer. Yeah. And they've got, Their best engineers on them, and the projects ultimately are just like, recreating a system to be the same but just built on a better foundation. Yeah. And so we found that, our FDC team have been doing an amazing job on some customers and have seen these projects come down by, 60, 70, 80% in terms of, It's done quicker. And these organizations will free up some of their best people to do other things that they didn't, probably didn't think they would ever have time to do. That's exciting- Yeah … as well, and we don't really know what that will lead to. Amazing. That is amazing. Fascinates me. And do you think organizations are underestimating the cultural change required to successfully deploy AI? I think possibly last year. I think everyone's been on like a learning journey and, We, we hold a lot of like closed-door events for our customers like C-suite, head of AI, et cetera. And A lot of the conversation is just around like sharing stories on what's worked well, and going back to my like two, two approaches to AI implementation on the kind of like ChatGPT Codex route. It's be- it's becoming clear what like what's working and it's- and it's- things like ru- running hackathons within your function- Yeah … showcasing what's working well, rewarding your people. And then there's some Metric monitoring as well. So Per like part of the business kind of monitoring like who's using tools like ChatGPT Enterprise or Codex like every week versus not, and you start to see which parts of the business are doing, starting to use them well and which aren't. So you know, Culture side is like super important. And as we touched on earlier, I think the bigger the organization, The harder it is just because there's more people, right? And there's more processes as well. I feel like the enterprises that have done the best from what I've seen over the last like year or so have been where they've Plucked out the people who are, for the big like use case builds- and picking out the people who are kinda, kinda get it like now. Yeah. And they've figured it out. They know like things are changing, they're building quickly they're using Codex, things like that. Making sure those people are focused on the top four or five initiatives, and then everyone else will follow right over time. It's gonna speed up quickly- Yeah … in the next like year or two. No, definitely. And this might be quite useful for some people as well, 'cause I know a lot of bus- a lot of businesses are like, they're maybe asking do we need an entire new AI team, or do we need to evolve the teams that we already have?" do you have a view on that, what you're maybe seeing in the market as well? Different organisations are doing different things. Yeah. And It's quite interesting. There, there are a few different kind of capabilities that's needed to make these things happen, and the way I've seen it work really well for some enterprises is where they have They break down like their AI transformation into like- Yeah say five or six kind of work streams. So one of them might be roll out Chat GPT to everyone in the company, …for example, to, to, and then they'll have some me- success metrics around it. Met- success metrics around it. That there needs to be a team arou- like a small team who are leading that. And that's all around the things we've just discussed around- Yeah … like enablement wizards, like rewards, things like that. you need a specific team around that. And then the other, the others might be like, there might be one around turbocharging or software development or engineering functions using AI and, That generally doesn't need a team, right? That's more just like a technical group in an enterprise anyway, so that's more about kinda leadership and picking whether you use Codex or something else and getting it in the hands of developers and then there's some process things that they, those functions need to do. Then for the big like build processes where we're like, "Okay, we're gonna transform our customer support function using like voice models- … or we're going to change this back office process that's taking six weeks and try and get it down to three days." those need like specific project teams, and that will be, business, like we mentioned, people AI engineers, et cetera, and they're generally like plucked from across the business. And they will work like In a kind of project sprint-type team. So I think Where I've seen it work really well is not create an AI function and they do everything. Yeah. It's more about like clustering around these like initiatives- Mm certainly in the first like year or so. and then often in those examples you have like more like an R&D AI team who are like really thinking big pictures in okay where might be- we be in like two years time? We often work with those types of teams to Showcase our new models that's coming out in like the next Like month or two and they can run testing and things like that. So that's the kind of structure. But I think, once an enterprise has got things into production, so maybe they've wrapped up their first three big implementations- …for example, and it's running in production I think the way it's probably going is that there's going to be like AI capabilities in each function- across the business, certainly in the big enterprise. And then there'll be like some sort of central centralized AI function. But their job is not necessarily to build everything. It's probably more around Releasing like reusable patterns where things like- a company might… You d- you don't want an enterprise building like a document processing solution like 100 times- … in different parts of the enterprise. So they might Scope out the different solutions and patterns that are gonna be reusable across the organization And lead the charge on that. And then things like methodology, evaluation, like you said, You don't want everyone to have to figure out what an - evaluation is like every single time they build something. So these kind of like repeatable repeatable patterns, methodologies, maybe the R&D is bundled into there as well. I could see that being like in a more centralized type function. think that's where things are going- Yeah but I think it's still pretty early days. So maybe the kinda future is embedding AI capability directly into every function over time? I think so, yeah. And we mentioned like FDEs, et cetera. I kinda see- Yeah … a future where you- Yeah you probably have an FDE type- person or team in every function. Certainly the big functions, and these are the people who are spotting the opportunities- Yeah … building quickly, prototyping- … and where things are working, like getting something into production very quickly. And then Pulling on other capabilities across, across the organization where needed. But like proper like transformation within each function Is Probably the way it's going. And I think the cool thing about things like ChatGPT and ChatGPT Enterprise and Codex is that, Day-to-day like work in each of these functions, you can actually do a lot of these complex things in Codex, but- the challenge then becomes like how do you scale that up? So I think- where I see things going potentially is the, in a typical organization, like someone will stumble upon something within a function just using Codex, where it's like,"Oh, I just Realized you could do this." Like you can build this report or something- Yeah … like just by you feeding it some data and a good prompt and things like that. And then maybe like an FDE or an AI team-… builds that into a solution because they realize that there's 100 other people in the function who just does that as their day-to-day job. So it's wrapping it up into an actual solution.. No, brilliant. So it feels like we're still very early in the AI journey, even with all the noise and investment happening globally. From where you sit at OpenAI, what changes do you think business leaders are still underestimating? I think some enterprises are still underestimating How much AI could transform their business and their industry. a lot of organisations are Trying their first use cases, and they might be picking the things that are fairly obvious, which makes sense. This thing is taking way too long, so we need to speed it up, and we can now do it using models. or, we need to scale up our, this sort of function, but we can't be hiring 20,000 people, so we can now use models to do this. So there's some almost quick wins even if- Yeah there's strategic use cases. It's fairly obvious the use cases that can be picked. But some of our customers are, like, thinking beyond the next two, three years. And we obviously talk about this a lot internally within OpenAI, and I think, I think this is where it starts getting a bit tricky for businesses, and it requires some, Crazy thinking around, you know, if we assume that, even just from an OpenAI perspective we keep releasing models every six weeks, and maybe that comes down to days, who knows- Yeah in the future, and you can basically do anything What does that mean for certain sectors, and what does that mean for us as a, as an organization? So again going back to something like insurance or banking or something like that, what does a bank look like in five years? On the retail side I don't know. My, my assumption is that I'll be interacting with my bank through, like, Chat GPT, On my phone and …asking it to like, my, my assistant to send a payment to someone. The bank's still there, but the way we interact with it as a consumer might be completely different. I think for the enterprises who are maybe still thinking about, like, how do we automate this, or how do we, scale this thing up that's amazing, but, you need to be thinking about, the next three, four, five years. And I think that's where we see some enterprises maybe underestimate how- Yeah quickly things can move like this, things are moving Very quickly. And so what does that mean for us, I think is sometimes not talked about enough, I think. I always like to ask people that if we could try and think and predict where things will be in the next five years, but I just don't think that question's even worthwhile now because it's just try maybe breaking that down to three months or s- or six months or 12 months. Yeah. It's just happening so quick. Everything's really changing, and when you try to predict- Yeah … five years away, it's just, I think it's impossible at the moment. It's cra- it's crazy. It really is. We've had some projects- Yeah … with customers even this year where we, like, We set out to do something in January- Yeah and it was gonna take maybe six, seven months or something, and then Codex came along and this was like- …completely cha… and then our models got better. Yeah. Some of the solutions even we've been involved with our customers have completely changed direction Be-because of the, they've maybe they've become a lot easier to do, or it's Actually you don't need to build something anymore. You can use our new agent functionality in Chat GPT because It's power- more, powerful enough to do it. So I agree. I think it's like thinking- Yeah in three months- Think so …chunks makes sense, yeah. No, definitely. And one thing I do hear a lot when I'm speaking to people is like AI is moving faster than organisations can adapt, adapt. Do you think that's true? in the larger enterprise side for sure. Yeah. it's just the way it is, right? It's the big organizations, lots of people. Everyone's busy doing their day jobs. It's, right now it's not possible to, for enterprises to move at the speed at, at which the models are coming out. And but I think a couple of things, like one, We've been investing a lot in terms of like our FDEs and we announced our Deployco, which is our k-kind of consulting arm. Yeah. We work with a lot of partners, like our partner organizations who come in and help us with c- like us and our partner and the customer working together. so there's like an increasing number of people who are helping enterprises to basically navigate this and share learnings from what we see from other customers. And then there are just like organically, like more people in the market who have done a GenAI implementation. And so like we'll see things move quicker and also organizations have been changing the processes to cater for this like new type of development. The engineering functions are starting to use tools like Codex so they can build quicker. So we'll see this like flywheel over the next few years where some of the blockers that exist today will Largely fall away, I think. And then there are some areas that we're trying to tackle as OpenAI with products. Yeah. As you'll know, like in a large organization, there's a big challenge around like access to systems and data. So you could have built this like amazing app, but if it has to connect to, for, For instances of Salesforce and 20 instances of this like Other types of systems which is in some other country and hosted on s- it's like it's difficult, right? You can't just you can't just do that automatically. And so we've been building some interesting technology will let-- which lets you connect and discover data from across the organization in a bit of an easier way. So there's a bit around like product and we're like figuring out what's the biggest blockers to getting AI deployed and see-- figuring out how we like build a product around it. And, but then we've also got like the people side, the FDEs, et cetera, to help on the ground to Make things happen. So like we're everyone's like throwing everything at it, I think. Yeah, definitely. And I was gonna cover a little bit about human, like AI, it will primarily like enhance human capability, but I think that's been a theme throughout the whole podcast. But where do you think human beings remain irreplaceable regardless of how advanced AI becomes? I think customer-facing role's super important. Yeah. Obviously in, like obvious examples would be, In like health, health sector, People want to speak to people. And I think like on, on the health side, like even just G- speaking to GPs, for example, I think a lot of people still want to speak to a GP, But if a GP has some sort of like real time co-pilot that's built on intelligence from every single medical case in history that's giving them real time guidance, like that combination is super powerful. there are certain kind of obvious cases where like the human touch is still- Yeah Is still like super important. I think, creative sectors. I think just coming up with, in some cases like new ideas and new strategies while we can obviously use AI to, to, To help with that. I think understanding local markets and what will work well w- versus well owned, I think, human touch will still be key. And yeah. And ultimately like what we've seen is there's still like lots of work to do. Yeah. This is not getting rid of everyone. A lot of what we've discussed today is challenges that need- Yeah … to be fixed with people. Like How do we figure this out, and how do we like connect these? Uh, There's lots of like things that need people to do, and I think When, as the capabilities get better and better, I think it'll be interesting to see what humans can do with them. Like you say, what does a recruiter do when actually they don't have to do any admin anymore? Yeah. Or maybe even do First interviews, you don't need to do them anymore. It's just like a voice model- Yeah … can do them. So what does that mean? You could be speaking to more companies. who you might be able to offer your services to like things like that. Yeah. We'd hope that people are spending their time on more productive things, and we've seen that ourselves, like in our team has transformed in the last few- months- Yeah like you mentioned. So we get spend more time With our customers, like in going traveling around rather than sitting making diagrams and things. Yeah. So I think humans are still like super key. Yeah. No, brilliant. Which is great. Yeah, definitely. There's a new kinda job title coming in recruitment. It's like some kind of engineering team. Everybody's talking about and even small- smaller firms like recruiters, and they're starting to look at bringing people in to build out the stack and get things moving. So it's now interesting. But finally- Yeah … if we were sitting here in 2030 looking back, what do you think people will say we completely underestimated about this moment in AI? AI? I think probably like the speed at which things are moving. Within OpenAI, and we're seeing kind of things play out in real time, and we see what's coming up in the next month or two months, three months. I think like I'm still often surprised and, have to take a second sometimes just to be like,"Whoa," like things- Yeah Things are moving very quickly. and so even just thinking where we might be by Christmas, I like-…I think it'll be stuff we can do that we can't do now. our current models will seem pretty old school even by Christmas, which is crazy. I think there'll be some like amazing stories probably in like health, like drug discovery. even just like from a consumer perspective, I'm super excited about having A super assistant which can just do all my life stuff for me- Yeah and I think we're like pretty close to that. I think a lot of things like that will happen between like now and Christmas. Yeah. and so I think if we look back in a few years' time, I think, I think we'll probably realize that we d- like we didn't really think… we might be saying we, we know it's moving fast, but I don't think we'll really have grasped like what that actually means and the kind of things that, We'll all be able to do and what it means for, Organizations, how we do, how we do our day work- So yeah, I think that's what I'd say. Amazing. Thanks very much, Stuart. Honestly, it's been a master class. I could've sat and spoke to you for ages. I had loads and loads of questions I didn't even get around to asking conversation was flowing. It was fascinating. I really enjoyed listening. Yeah, I can't wait to get this one out to all the viewers and people who have subscribed to the podcast and stuff. So thanks very much. And just as always before we finish where can people find you? Where's the easiest place if people are looking to talk to you or find out more about OpenAI or what you're doing? yeah, LinkedIn. Yeah. I'm mostly, on Some other places like X and things, but mostly in a kind of read-only capacity. so yeah, L- Link- LinkedIn's the best place. I usually do the odd post around what I'm up to and things. yeah, happy to connect. Amazing. Thanks very much for your time. If you found this episode useful, hit the like button, subscribe for more conversations like this, and reach out to us at MBN if you'd like to appear on the show. But thanks for listening to A Class Act, and I'll see you next time.