A conversation with Jean-Michel Baudoin, Chief Information Officer and General Manager of the Technology and Transformation Department at the Inter-American Development Bank
“The question is no longer ‘Can I do that?’ It’s ‘Should I do that?’ You still get to decide what you want to delegate to an agent. It’s not Anthropic deciding for you.” – Jean-Michel Baudoin
In this episode of New Friction, host Douglas Ferguson interviews Jean-Michel Baudoin, CIO and General Manager of the Technology and Transformation Department at the Inter-American Development Bank. Baudoin explains the gap between what AI can do technically and how fast an institution can absorb it. He points to confidentiality, token costs, developer resistance, and process bottlenecks before and after code generation. He argues that measuring AI adoption misses the point, and that leaders should ask how work could be done differently, not only how to speed up the existing process. He describes the IDB’s tech labs and AI experimentation studio, a product-based IT organization with new roles such as data specialists and AI engineers, and use-case-by-use-case data curation. He closes on workforce transformation, saying organizations should decide what to delegate to AI based on their core values and take people through the change step by step.
Show Highlights
[00:00:00] Introducing Jean-Michel Baudoin and the IDB
[00:02:00] Technology Moves Faster Than Organizations
[00:05:00] Why AI Adoption Is Not the Goal
[00:09:00] Industrializing Experimentation in the Tech Lab
[00:14:00] Solving Data Governance One Use Case
[00:17:00] Product Teams, Reteaming, and IT Disruption
[00:22:00] Should We Delegate to AI Agents?
[00:29:00] Digital Fluency Across Every Function
[00:34:00] Workforce Friction and Step-by-Step Change
Links | Resources
Jean-Michel Baudoin on LinkedIn
Voltage Control
About the Guest
Jean-Michel Baudoin is the Chief Information Officer and General Manager of the Technology and Transformation Department at the Inter-American Development Bank. He leads the IDB Group’s technology operations and strategic direction, as well as its broader digital transformation. That work brings together technology, data, innovation, and new ways of working to help the institution deliver greater impact. He has worked in technology for about 30 years and describes his role as pushing technology, aligning it with business outcomes, and building the organization’s digital fluency.
Transcript
Douglas Ferguson: Welcome to New Friction. I’m Douglas Ferguson. AI just made execution almost free. So, why are organizations still stuck? Because the friction didn’t disappear. It moved, and it multiplied. It’s no longer in building. It’s in deciding what to build, how to align, how to move forward when the path isn’t clear. That friction, the human side of change, is what this series is about. Each episode, I sit down with leaders who are living it, navigating the real challenges of AI transformation, not the tools, the people. The task that took two weeks now takes two minutes. The work isn’t the bottleneck anymore. The conversation before the work is. That’s the work this show is about. Today, I’m with Jean-Michel Baudoin, chief information officer and general manager of the Technology and Transformation Department at the Inter-American Development Bank. Welcome to the show, Jean-Michel.
Jean-Michel Baudoin: Welcome. Thank you, Douglas, for having me here.
Douglas Ferguson: Yeah, it’s great to be here. It’s like we had a great chat over an executive dinner we threw in D.C. a while back and been looking forward to continuing the conversation.
Jean-Michel Baudoin: Yeah. Let’s continue to talk about AI. That’s great. Great topic those days.
Douglas Ferguson: Yeah. No shortage of conversations about it, right?
Jean-Michel Baudoin: Oh, yeah, no, definitely. I mean, the only moment that I’m trying to not talk or read about AI, I mean, read, no, I have to, but it’s like during the weekends, because from Monday to Friday, it’s AI from 7:00 a.m. to 10:00 p.m. So, trying to take a break during the weekends, but that’s not possible. It follows you to your house, to where you play golf, or you go to restaurant, it follows you everywhere. So, that’s the way it is now.
Douglas Ferguson: Yeah, that’s an interesting thing to maybe reflect on because I was talking to a colleague not long ago and they were making the point that at some point, we’ll stop referring to it as AI or SI or whatever. It will become so pervasive and so accepted that we won’t talk about it. When electricity first got invented, I’m sure people that were working in electricity were like, “Electricity this, electricity that.” And now, it’s just like you take it for granted.
Jean-Michel Baudoin: Yeah, it’s funny you say that because we did an AI session four years ago after the whole OpenAI and ChatGPT came out. We did a session with the board of directors that we have at the Inter-American Development Bank to explain what it was at the time, obviously four years ago, and when it comes to AI, it was like an eternity. So, we did a session, and they asked us questions about how do you see AI in two years? And actually, the answer that we provided was exactly what you said. I said, “Well, we think that in two years, meaning probably 2024, we will not be talking about AI anymore because AI would be in everything that we do and everything that we do. So, we will not be talking about AI in 2024.” Obviously, we were, I think, a bit optimistic. It’s going to take more time than two years, but I think you’re right. I think at one point it’s just going to be part of our life and that’s it.
Douglas Ferguson: Yeah. It’s a little bit hard to even reason about it. It’s fun to have those conjectures as we’re exploring new technology. And as we’re seeing, it took more than two years for that to manifest. But there are things that happen in those two years that we probably didn’t expect. I’m shocked at how good the code generation’s gotten in the past two years.
Jean-Michel Baudoin: Yeah. So, this is the thing when it comes to AI is things are moving super fast, but at the same time, super slow in a sense of we talk about things that could be done, but it takes some time to materialize and to become concrete. But having said that, when you think about four years ago and ChatGPT and being able to prompt a few things and getting some text, very quickly they move to image, to videos, to sound, and everything. And then again, suddenly code generations, Mythos, cybersecurity things. So, things are moving extremely fast. There’s a lot of things that you wake up one day in the morning, and then you realize that Mythos is here and they can detect vulnerability. You realize that Claude for code is here, and then you can develop much faster, much better, higher quality. So, things are moving pretty fast in terms of technology breakthrough, but at the same time, to put those things into the ground to make that a reality inside organization, it takes a bit more time, which is a bit frustrating because the technology move at light speed and exponentially, and the capacity of organization to transition to that is more logarithmic. So, there’s a bigger gap all the time. But yeah, the good thing is every two or three months there’s a breakthrough that is like, “Wow. God, we can do that now.” So, that’s pretty amazing.
Douglas Ferguson: I think there’s many factors that contribute to the slow adoption. Part of it’s an organization’s nimbleness, how they’re culturally and structurally aligned, be nimble or not, but also there’s even some, I would say, table stakes that organizations have to put in place to be prepared to even leverage the new tools. For instance, if data isn’t accessible in ways that the tools can access that is safe and guarded and permissioned in ways that the right people can get access, but the wrong people don’t, if we don’t have those systems in place, our adoption’s going to be pretty limited by that too. And so, I’m curious what other factors you’re noticing that contribute to the slowness and what we as organizations can start to do to prepare not only our organization, but our people to meet these moments more quickly?
Jean-Michel Baudoin: Yeah, so that’s a very good question, Douglas. And again, there’s a difference between the technical capacity of a solution and the capacity of an organization to actually embed that solution and that capacity into their processes. And this is where the rubber meets the road, as we say, and things get more complicated. I was in the Valley a few weeks ago with colleagues from other IDBs, and we met with Anthropic, we met with Cloudflare, we met with AWS, Google, and so on. And I think Anthropic was saying that 70% of code today is actually developed by AI, which is a pretty amazing number. And then came back to my institution, and I was like, “Are we close to 70% or are we close to 0.70%?” And I think we are obviously closer to 0.70% than 70%. And not again, because the technology does not work. Because you mentioned it, there’s a confidentiality issue like, okay, so if we want to use that type of tool, we need to have an enterprise agreement. We need to make sure that whatever we are putting into this machine remains confidential, and there’s a cost issue. We try to use the tool, and then suddenly you realize that, “Oh, my God, it’s going to cost some money as well. The token consumption goes like this.” And at the same time, you are not yet in a position to say I’m going to use less developer because you are in this transition phase. So, you’re in that phase where you’re piling up the cost, and you’re not able yet to cut some costs. There’s a people resistance. It takes a bit of time to convince a developer that they’re going to be faster and better developer using the tool. So, there’s a lot of legal ethics, so there’s a lot of different internal decision as well that needs to be taken. When do we want to use the tool, or when do we want to use human? So, there’s a lot of things that are slowing us down. And one thing in particular when it comes to code generation is if you really want to get the full value of AI when it comes to, let’s say, application development, it’s not just about code generation. Because what we have seen here, for instance, we have been using Anthropic and Microsoft to do code development, but the part where you develop is one part of the whole process. You have the whole design thinking things, you have the whole requirement gathering, you have the detailed design, then you have the code development, then you have the testing, the quality, and so on, and moving the code into productions. So, at the end of the day, what you want to do is you want to reduce your cost of delivery and your time to deliver, but the entire things. At the end of the day, if one piece of your process is working very well with AI, that’s great, but you have a bottleneck before, and you have a bottleneck after. So, ultimately, you are not necessarily reducing your cost to deliver or your time to deliver. You are reducing one piece of the process. So, if you really want to start seeing some benefit, you need to do that for the entire process, and that becomes a bit more difficult because your process are not necessarily adapted for AI and so on. So, there’s a lot of work to be done. So, it’s super promising. It’s fantastic. We are all going in that directions. We think that the quality is going to be better, the security is going to be better. We’re going to be able to reduce the cost, but it’s not that easy. That’s the point.
Douglas Ferguson: Yeah. I think the thing you’re mentioning there is a very key point that oftentimes AI is used to automate or enhance a vertical slice, but value flows horizontally through the organization. And I think it’s a natural evolution because these tools are very siloed, individual tools, almost like a superpowered IDE for developers or a chatbot that’s helping us craft an email or whatever. And the more we can move into applications, whether they’re autonomous agents or other tools that cut through the functions and follow the value stream, we’re going to unlock a lot more value. To your point, it’s really difficult because if we’re automating, we’re following the existing human process. If we’re starting to look how AI can play a role through the value stream, we’re starting to bump up the abstraction layer and think about how the work flows in a totally different way.
Jean-Michel Baudoin: Yeah, so you’re starting this talking about friction. That’s, for me, the first point of friction is we, and I say we, everybody, I think, is in positions where we have an obligation to move towards AI. And the easiest thing to do is to adopt AI. So, we are talking a lot about AI adoptions and we look at the indicators that we are putting in place. And again, when I say we, my institution, but I’m pretty sure it’s across the board in the CIO community. We are putting indicators about how much licenses of Claude we have, of Copilot, of Gemini, whatever, how many processes we have agent inside the step of the processes. So, we are looking at adoptions and this is good, and this is fantastic. And ultimately, what it’s bringing to an institution is, I guess, more quality, more efficiency, better risk management. So, this is good, but we are missing the point. The point is not adopting AI, it’s “How can I do things differently now that I have AI?” And going back to why I’m challenging my team is going back to the code generation is like, “Okay, what do we want to do here? Do we want to look at how we are designing, building, and maintaining an application? And then we are trying to look at how we can embed AI in different step of the process to bring better quality and better efficiency? Or are we trying to remove all of this and rethink about how we should be developing application now that we have AI, now that we have superintelligence, now that we have this, now that we have that?” Meaning I have no idea if we need to do a design phase anymore. I have no idea if we need to continue to have those step in the process. Because when you think about technology, the way that we have been building application in the past 30 years hasn’t changed that much. We’re still doing the same process. It’s like requirement gathering, design, detailed design, development, tests, and so on. So, this has not changed for the past 30 years. The things that have changed is changing methodology, moving from waterfall to agile. So, instead of having those long cycle, we have much shorter cycle. But the business process of developing and maintaining an application did not change. And now, what we are doing with adopting AI, we are trying to make that process better and faster, but we are not necessarily rethinking or reimagining how the process could look like. And by the way, I don’t have the answer to that question. I have no idea. Maybe there’s nothing. But what we are telling to our internal customer is this, is thinking about how you can use AI to make your process faster is great and more efficient is fantastic. But don’t forget that you have another option here, which is let’s rethink completely about your process in the world of AI. Adoptions might be one of the limitation of the real value generations of AI inside an organization versus reinvention, which is where we bring, I think, a lot more disruptions, but also a lot more value.
Douglas Ferguson: Yeah. I think those two things go hand in hand as well because it’s hard to imagine how you redesign the process without having first glimpsed at what the technology’s capable of. And you can’t necessarily really grok that from a demo or from some lightweight usage. It really does require some deeper use cases and getting your arms around it, and then stepping back and doing some systems analysis-
Jean-Michel Baudoin: Absolutely.
Douglas Ferguson: … on how could this thing impact some broader flows.
Jean-Michel Baudoin: Mm-hmm. Yeah, absolutely. There’s so many things that we don’t know. One thing that I think it’s important that we never really forgot about that thing, but we are really trying here at the IDB to put a bigger focus, more resources, on this is everything related to experimentation and exploration. I think that for various reason, because of budget constraints or whatever, many, many organization in the past decades have starting to shrink a bit their budget relative to innovation and experimentation and exploration, and learning as well, because there was not so much disruption in the world of technology. Now, I think it’s time to be able to reposition exploration, experimentation, innovation, learning, and more at the center of what we do because this is what you say. I mean, it’s very difficult to disrupt something that you don’t understand. So, giving the capacity to the IT department and the other department to try to experiment and see what can be done with the technologies is becoming critical. So, the big push on innovation, experimentation and experimentation, I think, is extremely important in the age of AI for an IT department and for any type of institution.
Douglas Ferguson: Yeah, I agree. And we’ve been thinking a lot about how to ensure or help increase the odds that our experimentation is going to lead somewhere. And so, I’m curious to see what things you put in place to… Because I believe just telling people to go experiment is not necessarily a fruitful path. And so, what sorts of things are you doing to ensure your experiments are bearing fruit, or at least you’re increasing the odds that they will?
Jean-Michel Baudoin: Okay, so scaling from experimentation to proper solution is always a nightmare in general. But I actually do think that saying, like, “Innovation is important, experimentation is important, just go and do it,” it’s really the wrong thing to do because people are going to be absorbed by day-to-day type of activity. So, you really need to industrialize innovation and experimentation, make it something that is part of your practice, make it a real thing. Actually, and it’s not on purpose, I don’t know if you can see beyond, but this is the tech lab. Okay?
Douglas Ferguson: Mm-hmm. Yeah.
Jean-Michel Baudoin: So, actually, what we have in the IT department is we created the tech lab, and we have one in Washington, and then we have four in Latin America and the Caribbean. But those are spaces where you can actually experiment and explore and do some research. This is where we have sandbox. This is where we have the capability to provide tools that are not standard. We are using Microsoft and Anthropic inside the organization, but here at the tech lab, you can use any type of tools that you want to do in a very safe way. There are some people around here that is helping you into this experimentation process because this is a process. There are some people that are telling you when to stop and say, “Hey, you know what? Exploration, experimentation is done. You fail.” I mean, not that you fail, but the idea that you had is not going to scale, so you have this. We actually recently, inside the tech lab, opened what we call the AI experimentation studio, which is more focusing on AI. And we provided resources for the people to experiment and to explore, and put them in the right track from the beginning for potential scaling inside the organization. So, we are focusing a lot on this. We’re putting a lot of resources, making sure that we have the right asset to do experimentation and innovation. But even with all of this, moving from a prototype to a scalable solution at the scale of an organization still remains very, very difficult because, at the end of the day, to scale, there’s a different team that is in charge of providing the solution for across the organization. And they have a backlog of activities, they have a lot of things to do. So, the connections between the experimentation and the scaling is still remain difficult. But at least we have that place where we can say, “I have an idea. I think AI can help here. Let me try.” And we are not letting people alone. We provide the environment so that they can experiment and innovate.
Douglas Ferguson: Mm-hmm. In addition to the environment, are you all providing any direction or focus or strategic vision to help align on the types of things they might experiment with or projects they might pick up?
Jean-Michel Baudoin: Yeah, I mean there’s a few things. One of the reason that scaling experimentation is difficult, I mean, let’s move on the side of the fact that maybe it’s a bad idea. So, if it’s a bad idea, it’s a good idea.
Douglas Ferguson: It is.
Jean-Michel Baudoin: But if it’s a good idea, very often what’s missing is upfront thinking about the data, for instance. Every applications are using data, AI even more. So, sometime the proof of concept and the experimentation is done here. And then when you try to scale, you don’t have the data anymore, you don’t have the quality of the data anymore. So, those are the type of topic that we are moving upfront. We are looking at, obviously, a technology. Even if we are open to use different technology, at one point, if you want to scale, you have to start realigning with the architecture standard, with the tool standard of the organizations if you want to be able to scale. So, we’re trying to put them in that directions. Cyber, how do we design a secure… I mean, exploring, we have a sandbox here, so we are a bit more open in terms of a cybersecurity standard. But then when you start moving into proper solution and scaling, then anything you need to think about cyber. The maintenance costs, how are you going to maintain? So, it’s the whole process, and there’s a few things that we are pointing. We are helping them to try to guide them through the process to make sure that the experimentation can be scalable. But yeah, it’s a whole process and sometime unsuccessful, but very important in the age of AI, very, very important in the age of AI. We need to refocus on experimentation and exploration.
Douglas Ferguson: Another thing that I’ve found helpful in experimentation is having some guardrails because when there’s endless options, there can often be option anxiety, or if we can try anything, then it’s like, “Where do we start? Where do we go?” And so, I’m curious if your governance framework or processes or guidelines ever play a role in helping people experiment or at least use AI more effectively because it’s helping them understand where they can play versus where they don’t play.
Jean-Michel Baudoin: Yeah. So, that’s what we’re saying, that’s what we do, and that’s why it’s very well placed inside the IT department because let’s not forget, because we talk about AI and everything, but when it comes to managing a large digital portfolio for an enterprise, for an institution, I mean, they are cyber constraints. There are architecture constraints. There are cost constraints. There’s a lot of constraints. So, this is not a very funny environment when you have 400 applications in productions managing the business process of the institution. I’m telling you this is not so much fun. How do you support this from an end-user standpoint? How do you secure that? How do you do manage technical obsolescence and so on? So, there’s a big part of the job of a CIO that is not… AI is fun and everything. That part is, I love it, but it’s not as fun. So, when we start an experimentation here, we really try to avoid as much as possible to restrain too much the environment. So, basically, you have to look at the experimentation as a process. So, that’s why we’re trying, like, “Hey, if you want to try different solutions here, we are not using Gemini, but if you want to use Gemini here in the experimentation, you can.” But step by step inside the process, we say, “Hey, by the way, are you aware that we have responsible AI guidelines? If one day you want your idea to be in productions, there’s a responsible AI guidelines, so we have to make sure that there’s a human in the loop, so start thinking about this.” So, doing the whole experimentation, you start step-by-step putting some safeguard and guardrails so that when we move into the productions, it’s less painful. We used to be very much into “Okay, open,” and then suddenly moving into a very restrictive world of IT operations. Now, it’s more like a process, like an experimentation process where you start introducing constraints into the idea.
Douglas Ferguson: Makes sense. I’m pulling on that governance thread a little bit more. I found that there’s a vast difference across many organizations as to how much effort they put into gathering up data and organizing it into data lakes and contextual layers that AI can access, because AI is only going to be as powerful as what it understands about the business and the challenges that you’re working on. To point back to the thing you said earlier, it’s one thing to ask the AI to write code, but there’s all this upstream work around understanding the problem and defining it, and doing the research. And if none of that gets passed off to the AI, it’s going to not do a great job of writing the code, or we’ll have to spend a lot of time really defining how we want it to work, and that’s pretty laborious. And so, I’m curious, has that been an initiative that you’ve had to pick up, or was that ongoing work that you were doing prior to the AI? How are you approaching this kind of business context layer problem?
Jean-Michel Baudoin: So, one of the biggest challenges that we have when it comes to AI is, and you mentioned it’s data. So, again, the technology is fantastic, but the technology runs on data. So, if the data is not at the quality and the level of governance that it should be, then it’s complicated. And we do have a lot of data, structured data, unstructured data. We do have a structured data lake as well, where we have the data where we’re putting curated data and that we can use. We have a lot of unstructured data with very good quality, but not necessarily govern the best way possible. And actually, prior, well, not prior, more or less at the same time of AI, we put in place a data governance structure across the institution, together between IT and the business, just to make sure that we were tackling the topic of data properly. Data governance, data quality, data taxonomy, all of those topics are very, very important topic, but they are extremely, extremely slow to be put in place. And the value generated by data governance is even slower to materialize. So, it’s a very long process, a painful process almost, but it has to be done. Okay?
Douglas Ferguson: Mm-hmm.
Jean-Michel Baudoin: So, the approach that we’re having right now is it’s the expression not boiling, not trying to boil the ocean. So, we are not trying to solve data governance as one. We put the structure in place, we’re putting the processes in place, putting the different taxonomies and roles in place. And then what we do is we try to solve the data issue use case by use case. So, basically, we have a use case on whatever, you need data A, B, and C, so we’re going to curate data A, B, and C. But we are not trying to solve the entire things because it’s trying to boil the ocean. So, we’re not trying to boil the ocean. Every time that we have an AI use case, we put a team together with data specialist, UI specialist, AI specialist, and we are fixing the data issue just like this.
Douglas Ferguson: Yeah. It reminds me, I think of it as treating it like a product. We try to establish a foundation that we think will scale to our needs, at least into the reasonable future. And then we build on that product and enhance it as new requirements emerge, and we don’t try to build everything at once. We can live with whatever we need today.
Jean-Michel Baudoin: Yeah, that’s it. So, we switched to a product-based organization inside the IT department a couple of years ago. Well, we started more when we are finishing it, brushing it up, as we say. And I think this was a very smart move to move to a product-based model, 100% based on agile, with different roles, because we can have those product teams that brings different type of expertise. And we can see already that when we moved to product-based organization and decided to move there, we can see already how the structure of the product team needs to change, needs to evolve with AI. So, we used to have teams that were a product lead, a technical lead, a UX designer, a change management specialist, the developers, and so on. And now, when we are building a new release of a product, we say, “Oh, shit, we need a data specialist. Oh, we need an AI engineer. We need this.” So, we’re starting to see new roles coming up inside the product team in order to be able to deliver what we are supposed to deliver. But that brings a lot of flexibility in the product structure and in the team structure of the product.
Douglas Ferguson: It’s funny you went there because right as you were leading into that I’ve wrote down reteaming because I’ve noticed that a lot of organizations are starting to embrace this reteaming approach rather than the annual reorg or whatever, just do it more real time. When a new product spins up, who’s best fit for that project, or allowing folks to be assigned on an as needed basis. And it’s interesting that I’m seeing more and more people are adopting this in the age of AI. What I haven’t seen is a consistent pattern on cadence and exact approach. I’ve seen some folks say they’re reteaming every quarter, which to me almost sounds like a reorg more than a reteaming, but curious to see if that resonates at all, or?
Jean-Michel Baudoin: Oh, yeah, no, definitely. And by the way, I just want to say something here very quick for all people working in IT department or whatever, somehow, I feel a bit sorry for them because AI is going to impact pretty much every functions of any type of organization, but the IT department is going to be the most impacted function. And it’s funny because we carry the AI transformation, but at the same time we are the one that are going to be the most impacted by AI. I think it’s going to change completely, as I said in the beginning, how we are designing application, how we do cybersecurity, how we do IT operation, how we do patch management, how we do service desk. So, it’s going to change pretty much everything inside the IT department. So, I just want to say that, not that I feel sorry, but buckle up, guys, because IT department are going to be shaked a lot. And then the reality is, we started to move to a product-based organization, I think it was in 2023 when we decided to do that. So, we redefined the roles of everybody inside the department. We used to have project manager and whatever, and we changed all that. So, we reintroduced agile type of role and so on. And by the time that we were done, we have to change the role again because, okay, we need those type of profile now, we need those type of skills, we need this. So, honestly, I don’t believe in doing reorganization every year because that’s very tiring. And in the institution that I’m working, probably takes two years to do reorganization anyway. So, if you try to do that every year, it’s not possible. But yeah, I do believe that going back on investing on learning and training for people, I think this is very important. I mean, the IT people needs to unlearn and relearn a new thing all the time. And yes, we’re going to see a lot of reteaming, changing the roles, adapting the roles. So, that’s a bit crazy. And the product-based organization is very good for that. I think it’s better suited for this type of rapid evolutions of roles and skills and technology.
Douglas Ferguson: Yeah, there’s a couple of things that all point in the same direction. You talked about training, unlearning, and I would argue unlearning is a extremely important competency in this age of AI because we talked earlier about how things are just evolving so rapidly. The stuff they told us around how to prompt last year actually doesn’t work so well this year. And so, we had to be able to learn things, but then don’t hold onto it too tightly and be willing to adapt. Also, another thing that is, because you talked about how much IT was going to be disrupted, I was at a Gartner event not long ago and they were showing their vision of the future IT department and it showed each IT worker having a team of five to 10 agents. And just to be clear, these are autonomous agents. These aren’t a little chat box project that you might call an agent because you gave it a little bit of extra context or gave it a skill or some extra knowledge files. I’m talking about something that you’ve built and can be proactive, email you, check on things, go do things, and come back. And if IT workers are managing five to 10 agents, that word management needs to be considered. And that’s often a skill that most individual contributors have never spent any time building. And one important sub-skill under management is delegation. And I think that’s going to be really key, people learning how to delegate, what they should delegate. And a simple way to think about it is like, what are we delegating to the agent versus what does it make sense to delegate to existing systems versus what do we delegate to humans? So, being able to discern those things and those are the types of competencies I think are critical to train on versus how to prompt or how the tech works.
Jean-Michel Baudoin: Yeah, it’s very interesting. And by the way, on the prompting, I give up because every six months is changing. So, I’m like, “Okay, I’m going to wait until this is slowing down.” No, but you bring a very good point, and not just for IT, but for every functions inside the organization. And I’m starting to have some discussion with my peers internally. And I told you, I met Anthropic and Microsoft and Google and all of those guys a few weeks ago. So, they’re doing two things. So, they’re really pushing the autonomy and the intelligence of the agents to a point that is pretty unreal. And obviously, at the same time, they’re putting a lot of governance and control because the more intelligence you give to your agent, the more risk you have that the agent is going to go rogue or whatever or do something stupid. So, you have to say more intelligence, more autonomy, and at the same time more governance and control, which makes sense. But the point is, where are we going in terms of technology is pretty much almost, I don’t know, whatever percentage, let’s say 90%, 90% of what a human can do can be done today or tomorrow with AI. So, the question for organization, for IT department, for any type of functions is not so much longer about what the technology can do for me. Because if you ask that question, the answer is, well, pretty much everything. It’s what kind of organization and institutions I want to be? What do I want to keep internally? What do I want to give to AI? And I’m going to give you one example because I had that discussion a few times. Think about the recruiting process, the process of recruiting somebody. It’s about publishing a job, you push the job to the market, people apply, you don’t select resume based on the job description, you interview the people, you negotiate the contract, you sign the contract, and the people is on board. Today, this can be done completely using AI, completely. You can create, you can say, I recruit… Going back to your point, I’m an HR director, I have multiple agent at my disposal, and I say, “Recruit an IT cybersecurity specialist.” The terms of reference, the job descriptions would be written, would be published automatically. People would apply, they would then select the resume. You can have an avatar doing the interview, asking questions like video avatar, audio agent, interviewing the person, making the proposal, signing the contract. And then the first time that you see a new employee is when he’s coming to your place, and you’re shaking his hand. Say, “Oh, you’re the guy that has been hired.” So, technically you can do that. It works. Okay, it works. So, the question is not like… And it’s probably cheaper than… Faster, cheaper. And in terms of quality, it’s probably very close to what you can do with human along the process. So, the question is no longer like, “Can I do that?” It’s “Should I do that?” Do you want to be an organization that hire people through AI agent? Do you want to do that? So, that’s more like a philosophical questions is, “Do I want to be an organization that hires staff and employees without any human interactions?” And maybe for some organization the answer is “Yes, of course, because we are a super tech company and we are an AI startup,” or whatever. Or “Yes, we want to do that because we have tons of attrition and we hire 10,000 people, so the business case is huge and so on.” But for some other organization, that’s not going to be the case. I don’t think so. For some organization, they want to shake hands, they want to see the person, not because they’re old-fashioned, because that’s the image that they want to display as an organization. I’m a human-centered person, I’m a human-centered organization. So, I think the whole point on technology, on AI, on agents, and so on is where we’re going is going to be fantastic, but we’re still in control of what we want to do. Whether you’re an IT department or human resources department, you still get to decide what you want to delegate, give to an agent, what level of autonomy you want to give to an agent, what level of control you want to give to the agent. So, you get to decide that. It’s not Anthropic who’s deciding for you. It’s not Microsoft, it’s not Google, it’s you. And I think that’s an important message, I guess.
Douglas Ferguson: Yeah, that’s super important. And really asking organizations to really be in touch to their values, not these platitudes that hang on the wall, but really shape their use of AI and their other behaviors based on what they value, what they care about.
Jean-Michel Baudoin: I’m telling them, say, don’t ask the IT guy. The IT guy’s going to tell you that everything is possible. Don’t ask the finance guy. The finance guy is going to tell you it’s cheaper. Don’t ask those guys. Ask yourself, what are your values? What are your core values? And that would probably help you decide more about what you want to keep for yourself and what you want to get to AI, when your CIO would tell you everything is possible.
Douglas Ferguson: Yeah. Also, your story about the fact that you could completely automate HR, and you were talking about the job description for this cyber role or what have you, and it got me to thinking that all of this is so circular, and one thing in one part of the business is going to impact something in another part of the business. And so, as we start to use AI more for our cyber, that’s going to impact how the job description for that role evolves, and then how we recruit and what we look for, and is that a good candidate that’s showing up, and what are we expecting of folks? And I think it’s just all evolving so quickly. So, I’m curious, that feels like a new friction in organizations is not only how we’re individually staying up with what’s going on, but how are we staying up to date across our functions so that HR understands, like, “What is the qualities we’re looking for right now in these new folks we’re hiring?” But that’s just one example. There’s got to be endless others as you start to look across a large organization. How are we all just staying current with what the needs of the functions are?
Jean-Michel Baudoin: That’s a fantastic questions and comments as well. And to set the expectation already, I don’t have the answer to that. But it is true, and I think this is the role of the CIO to actually push for that because we are the best position. So, I see my role as obviously pushing technology, bringing efficiency, aligning technology with business outcome and the strategy of institutions, and so on. But there’s one role that is very important for me is the role of increasing the digital fluency of the organization. Because I do think that, I don’t know if it’s a good news or a bad news, but it is what it is. And let’s take the example of an HR manager, but it would be the same for a chief marketing officer or whatever. But in any of your function today, if you are an HR person, obviously you need to know HR, but you also need to have technical skills. You need to have digital skills. I remember back in the days when you were hiring somebody, you were hiring a Douglas, and you were looking at the technical skills of this person, and it’s like, “Okay, he’s the great guy because he knows whatever, cyber.” And then, 20, I don’t remember exactly when, but 25, 30 years ago, we starting to introduce soft skills because we say, “Oh, my God, Douglas is a great guy, but he’s completely crazy. He cannot communicate. He’s the crazy guy. So, now, we’re going to recruit a Douglas that has very strong technical skills, but at the same time he has some very good soft skills because that’s important for him to do his job.” And now, we are moving to the phase where it’s like “We want technical skills, so we want an HR person that knows HR. We want soft skills, we want an HR person that is not completely crazy and psychopath or whatever and can talk to other people. And now, we want digital skills. So, we want an HR person that has technical skills, that has soft skills, but also understand how technology can change HR.” And I think this is very important that if you want to do your job well today, technology is so omnipresent, is so everywhere, is so impactful, especially AI, whether it’s on HR, on communications, on marketing, and so on, that if the person does not, on top of his job, understand how AI and technology can actually make his job different, then he’s not going to be a good employee or so on. So, I don’t have the answer to that. I think the answer is pushing digital fluency, and I think it’s important that we push that from the IT department because we are all nerds and geeks and we love technology, so we have to push for that. But it’s very important as well as that people start understanding that technology does not belong to the IT department anymore and that they should not delegate or outsource this knowledge to the IT department and to the CIO, but they start building their own because technology is going to impact, and AI is going to impact everything, everything, every type of job. So, one way or the other.
Douglas Ferguson: Such good points. I want to ask a question. I want to come back to friction. And we talked a bit about a lot of things that I would label as frictions that are emerging, things that we might need to consider or work around, address as AI is showing up more and more in our work. Have you noticed any frictions that AI could solve that you’re intentionally not solving with AI? You did mention the values piece. I’m curious, aside from the things we might not do because of values or governance, but just friction we’re intentionally leaving because there’s some value in that friction being there.
Jean-Michel Baudoin: Like what, for instance, what kind of AI?
Douglas Ferguson: One that’s come up for me is if it’s helpful in onboarding new employees, because if we remove that on-ramp and we made that really simple and easy, we might lose that experiential portion of their onboarding that’s so important. And even though we can make it more efficient, we might lose something in the process.
Jean-Michel Baudoin: In that case, we have tons of examples. I mean, right now we are moving to a phase, so we were in this AI adoption phase and using Copilot and Anthropic and so on to improve the individual productivity of a person. And now, we are moving to a phase where we’re looking at agentic AI and how we can build AI inside the process of the bank. And this is opening tons of frictions here, tons of frictions. And this is where you start realizing that AI is really about workforce transformation because at the end of the day, AI is shifting activities or intelligence into AI, an AI solution. So, you are stealing the job of people somehow. You’re shifting something that I’m responsible for this or I’m doing this like this and like that, and then suddenly it’s going to be done completely differently, or it’s not going to be done by me at all. As long as we are giving AI as individual productivity tool, which is, “Hey, I’m giving you a tool to do your job faster,” people are happy because they say, “Oh, my God, that’s pretty cool.” When they start using AI, people say, “Well, that’s very cool. I can do my job faster, better quality, it’s less tiring.” But when you start looking at the business process, regardless of the business process, whether it’s an operational business process, HR, whatever, then it’s a different story. You’re not giving a tool for the person to be more productive. Potentially you’re changing the complete process to be more productive, and therefore this is impacting roles, this is impacting responsibility, this is impacting workforce. So, here we have tons of frictions here, and this is where going back as well to, I think, your initial point, which is we also need time as human being to understand AI. And even if the technology is fantastic and we can do a lot of things, I think that the human being needs to go step by step. I mean, leapfrogging is a wonderful concept, but I think when it comes to AI, it’s complicated for the human being. So, I think that’s what you said at the beginning, giving them tools for individual productivity, this is great. And now, we’re going to look at the business process, and we’re going to start putting on AI in the business process, that’s okay. And then we’re going to move forward with completely reinventing the process and having autonomous agent doing this. I think we need to go through some sort of journey because I love the fact that you are using the word frictions because there are some frictions when it comes to AI, and I do think that we need to remove the frictions step by step and not just the Band-Aids. So, I think it’s easier for the human being to understand what can be done. It’s easier for the human being to understand the frictions and live with the frictions. And so, the frictions disappear step by step because otherwise we might get a better pushback. And there’s a lot of uncertainty as well. We didn’t talk about cost here, but that’s the cost of AI, the value of AI. There’s a lot of uncertainty. Even on the solution, OpenAI, Anthropic and Google, and Meta, there’s so much money put on the market right now. I have absolutely no idea how the AI market is going to look like in six months, two years from now. No idea, no idea. So, there’s a lot of uncertainty. If you combine uncertainty with fear sometime, with frictions, as you mentioned, then the recommendation maybe is like, “Okay, maybe let’s not go completely crazy. Let’s go step by step. There’s a fantastic potential, but let’s take everybody through a journey.” And I think that’s how we have to do it.
Douglas Ferguson: It gives folks a sense of stability and scaffolding in an environment that feels a bit chaotic or stormy, and so that can reduce anxiety for sure. So, as we come to a close here, I want to give you an opportunity to leave our listeners with a final thought.
Jean-Michel Baudoin: Oh, my God, a final thought on AI. Look, I’ve been working in technology for 30 years now. I wouldn’t choose another career. I think working in technology has been fantastic because things are changing all the time in technology. Since 2022, it’s been crazy, really crazy. I’m going to use a technology nerd things, but I feel a bit like quantum. I feel like I’m in two states at the same time. I’m super energized and super tired. I’m in the quantum state. Tired, because there’s so many things, and every single day something is coming up. We’re starting the thing with so many breakthrough every six months on AI, Mythos, code generations, and so on. So, it’s very tiring to keep up. A lot of expectation from our client, a lot of fears from the staff as well sometime. And people are looking to us. They’re looking at the CIOs and they’re looking at the IT executive, the digital executive of the world say, “Okay, what’s going to happen here?” So, it’s very tiring, but at the same time, it’s fantastic. I mean, the potential of this technology is great. It’s going to completely change how we work. I have four kids aged between 18, 19, and 23, so they are going to enter the workplace. And the next generation coming up, and my kids are already old, but when I’m thinking about all those next generations coming into the workplace, my duty, I feel that I have to prepare my organization to welcome the future workers. And this is super energizing, this is super exciting, super tiring as well. But yeah, I’m very happy to live this phase of technology.
Douglas Ferguson: Jean-Michel, it was a pleasure chatting, and thanks so much for joining me.
Jean-Michel Baudoin: Yeah, thank you, Douglas.
Douglas Ferguson: Thanks for listening to New Friction. If you enjoyed this episode, share it with a leader who’s in the middle of this right now. They’ll thank you for it. And if you want to go deeper, we bring leaders together through executive dinners and virtual masterminds. To learn more about our work or to inquire about exclusive executive events, visit voltagecontrol.com. I’m Douglas Ferguson. See you next time.